GOAT labs
StudiesPolymarketBenchNEWModelsDomainsMethodologyFAQAboutContributorsCareers
Log inContactBecome a contributor
Legal

GOAT Labs — Privacy Policy

Effective Date: October 26, 2026

GOAT Labs ("GOAT," "we," "us," or "our") provides tooling that helps organizations connect approved coding and observability Sources, receive cashback based on eligible usage, and contribute trace data to a deidentified dataset and research program. This Privacy Policy explains how we collect, use, disclose, retain, and protect personal information and how we process trace data.

This Policy applies to personal information GOAT processes as a business or controller — including account, website, onboarding, fraud-prevention, payout, and support data — and explains how GOAT handles trace data that may contain personal information when GOAT processes it on behalf of a contributor under a Data Processing Addendum.

Terms of ServicePrivacy PolicyDPA
Contents · 15 sections
  1. 1Notice at Collection
  2. 2What We Collect
  3. 3Where the Information Comes From
  4. 4How We Use Information
  5. 5How Our Trace-Data Processing Works
  6. 6De-Identified Data and Licensing of Datasets
  7. 7How We Disclose Information
  8. 8California Privacy Disclosures
  9. 9GDPR and UK GDPR Legal Bases
  10. 10Your Rights and Their Limits
  11. 11Retention
  12. 12Subprocessors and International Transfers
  13. 13Security
  14. 14Children
  15. 15Changes and Contact
On this page
  1. 1Notice at Collection
  2. 2What We Collect
  3. 3Where the Information Comes From
  4. 4How We Use Information
  5. 5How Our Trace-Data Processing Works
  6. 6De-Identified Data and Licensing of Datasets
  7. 7How We Disclose Information
  8. 8California Privacy Disclosures
  9. 9GDPR and UK GDPR Legal Bases
  10. 10Your Rights and Their Limits
  11. 11Retention
  12. 12Subprocessors and International Transfers
  13. 13Security
  14. 14Children
  15. 15Changes and Contact

1. Notice at Collection

At or before collection, we inform you that we may collect the categories described in Section 2 and use them to provide and secure the Service, calculate cashback, review for and respond to suspected abuse or fraud, comply with law, communicate with you, measure and improve our marketing website and dashboard and find and fix errors in them (Section 2, "Website and dashboard analytics"), send emails about our datasets to people at businesses and research organizations, see which organizations and which of the people we emailed respond to our outreach emails and decide whether to follow up with them (Section 2, "Outreach emails and visits that come from them"), handle the calls you book with us (Section 2, "Booking a call"), and — for trace data you connect — create De-Identified Data and Corpus Products. We retain each category as described in Section 11. We do not use the website and dashboard analytics data described in Section 2 for advertising, except that, for a visit that arrives from our outreach domains, and for the records of our emails to the people we emailed, it can help us decide whether to send a follow-up email to an organization or a person we emailed (direct marketing; Section 2, "Outreach emails and visits that come from them"). To exercise privacy rights or to opt out of any processing you can opt out of, contact us at privacy@goat.ai (Section 10); you can also change your analytics choices yourself at any time (Section 2).

2. What We Collect

Account and organization data: your email address and display name; organization and project names; a coarse company-size selection; and teammate invitation email addresses. Accounts are created via Google sign-in (we request only basic profile and email scopes — not Gmail, Drive, or Calendar) or email magic link. We do not store a profile photo.

Authentication and credential data: login events; the authentication identifier; and API keys, CLI tokens, device keys, and connected-Source credentials — stored only in hashed or encrypted form, never as plaintext (plaintext is shown once at creation).

Trace and telemetry content: prompts, completions, model reasoning, the full content of tool/function calls and their results (which can include complete file contents, source code, diffs, command output, pasted text and images, and attachments), tool and MCP-server definitions and configuration, system prompts, tokens, cost, model and provider identifiers, timing, session and conversation identifiers, provider response identifiers, and repository/file context — including working-directory paths, git branch and commit, and the repository origin/remote (with embedded credentials stripped; the host, organization, and repository name are retained). When you connect a local coding tool, historical (pre-existing) sessions may be uploaded at installation. When you connect an observability Source, we may also collect the end-user identifier supplied by that Source (e.g., your application's customer or user ID) and any user-supplied metadata, tags, or feedback. Unlike contributor identifiers, these source-supplied end-user identifiers are captured and retained as provided (encrypted at rest) and are not pseudonymized by GOAT; the contributor is responsible for the lawful basis and for any notice or consent owed to those end users.

Multi-party collection. Where a contributor installs repository- or organization-level capture, or connects an observability Source, we collect trace data generated by other individuals who use that repository, tool, or Source — including teammates, employees, and the end users of the contributor's own product — even though those individuals have not separately installed our software, signed in, or interacted with us. The connecting contributor is responsible for any required notices and authorizations (Section 10 and the DPA).

Contributor identifiers: for repository/local capture, a one-way pseudonym derived from your git identity (we do not store the raw git email); for anonymous capture, a pseudonymous machine fingerprint (a hash of your operating-system username and home-directory path), used to attribute and rate-limit anonymous capture. Although the raw git email is not stored, we maintain internal mappings that let us associate trace data with an account for attribution, payout, support, and fraud-prevention — including (a) a reverse mapping from a contributor pseudonym to your account where you sign in and link a device, and (b) the association of every project and trace with the organization that owns it and that organization's designated payable account. As a result, trace data is reasonably linkable by GOAT to an organization and, in most cases, to an individual account, and trace identifiers should not be treated as anonymous.

Network and security data: we read your IP address transiently to rate-limit certain requests; for a visit that arrives from our outreach domains, we also use it to look up which organization operates your network and the approximate location of your IP address, and we include your IP address in our team's alert about that visit (see "Outreach emails and visits that come from them" below); and we record the actor IP address on audited security, access, and administrative events (for example, internal data access, exports, key and project management, changes to your dashboard analytics choice, and payout actions). Apart from the daily visitor code described under "Website visit counting" below, and the 10-second check for a second copy of an outreach visit (see "Outreach emails and visits that come from them" below), we do not use browser fingerprinting, and we do not build tracking profiles across websites. (The machine fingerprint described above is a single hashed value used to attribute and rate-limit anonymous capture, not a browser or behavioral tracking profile.) Trace data records the name and version of the coding tool, and of the GOAT capture component, that produced it; we do not send these trace records to analytics providers (dashboard setup events name only the type of Source you connect).

Payout and finance data: your payout destination (a crypto Wallet address or a Coinbase email — stored as you provide it), payout amounts, on-chain transaction references, and payout status. We do not currently collect tax forms or taxpayer identification information, and we do not perform identity, sanctions, OFAC, anti-money-laundering, or KYC screening. We reserve the right to request tax, identity, or sanctions documentation in the future and to condition, delay, or withhold a payout until it is provided.

Support and communications data: contact-form submissions (name, email, and optional provider, domain, monthly-spend, reason, and message), which we receive by email and in a Telegram message to our team (described below), and do not store in our application database; and other records of your interactions with us. The contact box in the footer of goat.ai pages sends only the email address you enter, which reaches us by email like other contact-form submissions. The contact form at goat.ai/contact also sends, with your message, how you reached goat.ai during that visit: the goat.ai page where you arrived (or the goat.ai page you came from, if that is where the link information below was), recorded only if it is one of our site's pages; the domain of the referring website; the campaign tags (such as utm_source), referral codes and campaign identifiers in the link you followed; the names of up to ten of the link's other parameters, never their values, and only names that cannot be a code or key (the rest are only counted); and whether the link carried an advertising, social-network or email-tool click identifier, never the identifier itself. A line under the Submit button shows this, with details, and lets you remove it before sending. If you declined analytics cookies, or your browser sends a Global Privacy Control or Do Not Track signal, it is sent only if you choose to include it. Your browser keeps this information only in its memory while the page is open, and we do not pass it to any analytics provider (they receive only what "Website visit counting" below describes). Our other contact forms, including the footer contact box, send none of it. When you submit a contact form on goat.ai, we also record an analytics event for it (see "Website and dashboard analytics" below) that says which form you used and, where you chose them, the reason and data-use option — never your name, email, or message. A request sent with the reason "Privacy or data request" carries none of this, goes only to the people who handle such requests, and is never recorded as an analytics event. When you submit a contact form on goat.ai, our server also sends our team a Telegram message with what you entered in it, which form you used, for the contact form at goat.ai/contact how you reached goat.ai as described above (unless you removed it), and whether the email reached us. If the email reached us, a long message is shortened in Telegram (the whole of it is in the email); if the email could not be sent, you are shown an error, and the Telegram message, split into more than one if needed, carries all of what our server accepted (up to 5,000 characters of a message and 100 of a first or last name, as for the email) and is the only copy we receive. For a request sent with the reason "Privacy or data request" on goat.ai/contact, the Telegram message says only that one arrived and whether the email reached us, without your name, email address, or message; a request sent with another reason or another form reaches our team in Telegram like any other submission. These messages go to the same chats as our outreach alerts and to our team's main Telegram group (Section 12), up to 30 an hour for submissions whose email reached us (beyond that, those reach us only by email) and, separately, up to 30 an hour for submissions whose email could not be sent (beyond that, those do not reach us), and stay there like the outreach alerts (Section 11); our server keeps nothing of them once they are sent.

Cookies and local storage: on the dashboard (app.goat.ai) we set strictly necessary cookies for sign-in and your authenticated session. They include two short-lived cookies, goat_terms_intent and goat_analytics_intent (up to 30 minutes each), that carry your acceptance of our terms and, if you tick it, your Product analytics choice through sign-in, and are then deleted. The dashboard also sets one interface cookie that remembers whether its sidebar is open (7 days). Our marketing website (goat.ai) and the dashboard use browser local storage for interface state, such as the color theme, a calculator value, or a selected date range (which the dashboard also adds to the page address); this interface state is not sent to any analytics provider, although a session recording on goat.ai, if you accept it, shows each page as you saw it (see "Analytics cookies" below). We use no advertising cookies. The analytics cookies and storage we use, and the goat_consent cookie that remembers your choice, are described below. If you open our booking page in its window on our dataset pages, Cal.com, whose page it is, may set its own cookies and storage entries for it (see "Booking a call" below).

Website and dashboard analytics: we use two analytics providers, PostHog and Google Analytics, to measure and improve our marketing website (goat.ai) and, only if you turn it on, the dashboard (app.goat.ai), and to find and fix errors in them. No analytics provider receives trace data, prompts, completions, file contents, API keys, DSNs, or tokens, and none receives your email address or name, unless we sent you one of our outreach emails or you book a call with us on Cal.com (PostHog then receives your name and email address with the record of those emails, or your email address with the booking; see "Outreach emails and visits that come from them" and "Booking a call" below). What they receive depends on the surface and on your choices, as described in the next five paragraphs.

Website visit counting (goat.ai, without cookies): when you use goat.ai, the page sends events to PostHog through our servers: page views, a few clicks with the part of the page they were in (for example "Sign in", with the coding tool the link was for, if any; "Contact"; an email link; a dataset link or copy button; a "Book a 15-min call" link, with where on the page it was; or a tab or view in a dataset preview), contact-form submissions, and, for a visit that arrives from our outreach domains, the outreach event described in "Outreach emails and visits that come from them" below. Each event carries the page address, the referring page, your browser and its version, operating system, the kind of device (such as a desktop or an iPhone), screen and window size, language, and time zone. Query parameters are removed from the page address, except campaign tags such as utm_source and, on the contact page, the topic you arrived with; for another website, only its domain is recorded as the referring page. Page views also record how long the previous page in the same visit was open. Our servers also pass on your IP address, and the page passes on your browser's user-agent string. PostHog uses your IP address and user-agent string only to compute a one-way code, together with a random value that it creates for each day and deletes within a few days; it does not store your IP address or user-agent string. PostHog stores each event under that day's code. This lets us count visitors and see which pages one visitor viewed on one day, but not who you are or which visits on other days were yours; for a visit that arrives from our outreach domains, our team may also learn the visitor's IP address, its approximate location, which organization operates the visitor's network, the visitor's device, browser, time zone and language, and what the visitor did on goat.ai during that visit, and may be able to guess who visited, and PostHog may then hold that guess under the email address of a person we emailed (see "Outreach emails and visits that come from them" below); and if the visitor books a call with us through one of our links and the booking carries that visit's reference number, the reference number links that visit's events, in PostHog and in our team's chats, to the name and email address of the person who booked, which is much more reliable than a guess, though a copied or forwarded booking link can carry another visit's reference number (see "Booking a call" below). PostHog derives no location from your IP address. Nothing is kept on your device for this count. We do not measure pages of our PolymarketBench section; at most, analytics records that a visit came from or went to that section, and an outreach alert can say that a visit arrived on it (see "Outreach emails and visits that come from them" below). You have the right to object to this count (Section 10, "Your right to object to website visit counting").

Analytics cookies (goat.ai, only if you accept): a banner on goat.ai asks whether you also accept analytics cookies. In this Policy, "analytics cookies" means both cookies and similar entries in your browser's storage, together with the session recordings and error reports described below, which run only with them. Only if you click "Accept analytics cookies":

  • we load Google Analytics 4 (Google LLC) from Google's servers. Your browser then sends Google your IP address and browser information, the same cleaned page address and referring page, the page title, how long each page is in view, and the same clicks and contact-form submissions. Google uses your IP address to derive an approximate location (country and region) and to filter spam, and says that it then discards the IP address. We have turned off city-level location, detailed device data, Google signals, and advertising features. Google Analytics sets the _ga and _ga_<id> cookies on goat.ai, which expire 13 months after your last visit.
  • PostHog keeps a random identifier, together with your current session and the cleaned first page, referring site, and campaign tags of your visit, in your browser's local and session storage (entries whose names begin with ph_ and __ph_opt_in_out_), so that it can recognize later visits from the same browser. These events still pass through our servers, which pass on your IP address; PostHog does not store it, and in this mode it does not receive your user-agent string.
  • PostHog also records how you use goat.ai pages (a "session recording"): what the page showed, your mouse movements, clicks, taps and scrolling, and the size of your window, so that we can see where pages are confusing or broken. Anything you type into a form field is hidden in your browser before the recording is sent. We do not record our contact pages (goat.ai/contact and goat.ai/talk-to-research), the PolymarketBench section, or a page you arrive at from our outreach domains (see "Outreach emails and visits that come from them" below); recording also stays off on the goat.ai pages you go on to from that page, until your browser loads a page afresh (for example, when you open a goat.ai page in a new tab). Recording pauses while the page address contains anything other than the campaign tags described above or a link to a section of the page. A recording does not include your network requests, your browser console, the content of canvas elements, our booking page shown in a window on our dataset pages (see "Booking a call" below), or anything else outside the goat.ai page. A part of a recording that has not been sent yet may be kept briefly in your browser's session storage, under an entry whose name begins with ph_. Recordings pass through our servers, which pass on neither your IP address nor your user-agent string.
  • If code on a goat.ai page fails, PostHog receives an error report: the type of error, its message, shortened and with email addresses, quoted text, long codes and long numbers removed and web addresses cleaned like page addresses (on our two contact pages, no message at all), and where in our code it happened, together with the same page and browser details as the events above. Error reports pass through our servers like those events. The code that makes session recordings and error reports is loaded only after you accept.

Your choices on goat.ai: we remember your choice in one strictly necessary cookie, goat_consent (set on goat.ai only; 180 days). If you decline, session recording and error reports stop, anything not yet sent is discarded, and we delete the analytics cookies and storage entries at once; visit counting without cookies continues. When your choice expires, we delete them the next time you visit goat.ai, and the banner asks again. You can change your choice at any time with "Analytics preferences" in the footer of goat.ai pages, and you can delete these cookies and storage entries yourself in your browser settings. If your browser sends a Global Privacy Control or Do Not Track signal, we treat it as "Decline", even if you accepted before, and we do not offer the option to accept while the signal is on; visit counting without cookies continues.

Dashboard analytics (app.goat.ai, only if you turn them on): no analytics run in the dashboard unless you turn on "Product analytics". We offer it as an optional, unticked checkbox when you sign in (including when you approve a CLI login) and when you accept our legal terms, and as a switch in the dashboard account menu. Leaving the checkbox unticked does not change your current setting. Your choice applies even if your browser sends Global Privacy Control or Do Not Track. If it is on, PostHog records the dashboard pages you open inside a project and during onboarding, and the onboarding step you reach. Addresses are recorded with project and other identifiers removed, and with query parameters removed except campaign tags. It also records these setup steps: creating an API key or DSN, approving a CLI login, connecting a Source (with the type of Source), receiving your first traces (with their number only as a range, such as "2–10"), and sending or accepting a team invite (with the role). It also records errors in the dashboard's code: the type of error, how it was caught, where in our code it happened, and the page it happened on (recorded the same way as the addresses above), never the error's message, which could contain what was on your screen. An error report may also carry a short error code that our servers generate. Events and error reports from your browser carry your GOAT user ID (a random internal identifier for your account), the ID of the project you are viewing, the dashboard version, the referring website's domain, your browser, operating system, the kind of device, screen and window size, language, and time zone, and how long the previous page was open. The setup-step events our servers send carry only your GOAT user ID, the project ID (except when you accept an invite), and the details named above. No dashboard analytics event carries your email address or name. We do not make session recordings in the dashboard. PostHog keeps a profile under your GOAT user ID that groups these events. Requests from your browser pass through our servers, which remove your IP address, cookies, and other request headers, and our servers send the setup-step events directly, so PostHog does not receive your IP address. The PostHog script keeps nothing in your browser; the only dashboard analytics item on your device is the goat_analytics_intent cookie described above, set only if you tick the box while signing in. When you turn Product analytics off, collection stops at once in that browser and within a few minutes on your other devices where the dashboard is open; events already collected are kept as described in Section 11. If we materially change how dashboard analytics work, your earlier choice stops counting, and they stay off until you turn them on again in the account menu or with the checkbox offered when you next sign in. We keep a record of each change of your choice: your GOAT user ID, whether you turned it on or off, when, where (sign-in, the terms screen, or the account menu), and which version of this Policy it referred to (the most recent version that changed how dashboard analytics work); the audit event for each change records the IP address it came from. These records are append-only (Section 10, "Immutable records").

Website and dashboard kept apart: we do not link your goat.ai visits to your dashboard account. The two are recorded under unrelated identifiers, no analytics cookie or storage entry is shared between goat.ai and app.goat.ai, and the website never sends an account identifier to an analytics provider. We do not use analytics data for advertising or to build profiles across websites, and we do not use it to make decisions about you, except that, for a visit that arrives from our outreach domains, the outreach alert, its replies, and the visit's events in PostHog that carry its reference number, described in the next paragraphs, and, for the people we emailed, the records of our emails in PostHog, the messages about clicks on links in our emails, and the daily file, also described there, help us decide whether to send a follow-up email to an organization or a person we emailed (direct marketing).

Outreach emails and visits that come from them: we email people at businesses and research organizations about our datasets, including from the domains goatlabsai.com, goatlabsi.com, and goatailabs.com. These three domains ("our outreach domains") are ours and send their visitors on to goat.ai. They are hosted by Netlify (Netlify, Inc., United States), which receives your IP address, your browser's user-agent string, and the address you open when it sends you on. Their links to goat.ai can carry campaign tags (such as utm_source and utm_content) that name the domain and the short name in the address after it. Some of our emails may contain a link to, or suggest, an address made of one of these domains and a short name for your company (for example, goatlabsai.com/example for someone at Example Inc.), so that a visit through it tells us which organization we sent it to; the outreach domain adds that short name to the link to goat.ai as a campaign tag (utm_content), which PostHog and, if you accept analytics cookies, Google Analytics record like other campaign tags (see above).

We send these emails with Smartlead (521 Products Pty Ltd, Australia), our email-sending service, which records when each of our emails is sent, opened, and replied to and when a link in it is clicked. Smartlead learns that an email was opened when your mail program loads a small image in it, and that a link was clicked when the link passes through Smartlead on its way to goat.ai; each time, Smartlead receives your IP address and your browser's or mail program's user-agent string (or those of a service that loads images or checks links for your mail program). Every few minutes, our server reads from Smartlead, for our current campaigns, each recipient's name and email address, which of our emails they were sent, when, and from which of our outreach domains, and when it was first opened, first clicked, and last replied to, and how often it was opened and clicked, and whether they unsubscribed, their address bounced, or we blocked them. Smartlead's answers also contain the text of our emails and of any reply, and other fields; our server discards them as it reads them, without keeping, logging, or sending them. Our server keeps what it reads only in its memory: the record of each email for as long as that email is part of a campaign we read, until our server restarts, and the email events it makes from them for up to 26 hours. It sends these events to PostHog under the recipient's email address, together with the recipient's name, the domain of their email address (unless it is one of the common personal mail providers), our campaign, and, with an open or a click, our guess whether it was automatic (an open within two minutes of the email's delivery, or a click within five minutes of it, often comes from an email security scanner or from a mail program that opens every email as it arrives), so that PostHog groups them in a profile of that person; and when an alert or one of its replies (below) names a recipient, it also sends PostHog our guess, marked as a guess, that the visit with that alert's reference number was theirs. Our server also uses the domains of those email addresses as part of the list of organizations we emailed (below). Our server drops, as soon as it reads them, the records of anyone who unsubscribed from our emails, whose address bounced, or whom we blocked in Smartlead (we block in Smartlead everyone who asks us to stop emailing them), and of everyone whose email address belongs to an organization that objected (Section 10); our server reads an unsubscribe or a bounce within a few minutes, and when we block someone in Smartlead we also unsubscribe them from our campaigns, so that our server reads it in the same way. It does not keep such records, name such a person in a new alert or message other than the daily file described below, the message about a call that person books with us (see "Booking a call" below), and the message about a contact form that person sends us ("Support and communications data" above), or send PostHog anything new about them. When that happens to someone our server knew, it forgets them at once, except that it keeps a coded form of their email address in its memory until it restarts, so that they stay excluded, and their email address itself until a daily file described below has listed it, so that we delete what PostHog holds about them by hand. Because a restart makes our server forget whom it knew, the first time it reads Smartlead after a restart it lists in the same way everyone it then finds excluded.

When you arrive at goat.ai from one of our outreach domains (we tell this from the campaign tags in the link or from the referring site), the page tells our server which of our outreach domains you came from, whether the link's campaign tags or the referring site showed it, the goat.ai page (only if it is one of our site's pages), and the short name in the address after the domain, if it is made of letters, digits, and hyphens. The page also tells our server your browser's time zone and language. Our server reads your browser's user-agent string from that request to name the type of device, its operating system, and your browser (for example "iPhone, Google app" or "Windows, Edge"); it does not keep or send the user-agent string itself.

Our server reads your IP address from that request and looks up which organization operates the network it belongs to (for example, your company or university, or your internet, cloud, or VPN provider). It does this in a database from IPinfo that our server downloads and queries itself, so IPinfo does not receive your IP address. From that database it uses only the name and domain of that organization. When it makes an alert (below), it also looks up the approximate location of your IP address (city, region, and country) in a database from DB-IP that our server downloads and queries itself, so DB-IP does not receive your IP address either; from that database it uses only the city, region, and country. Such a location can be wrong, or be where your internet provider's equipment is rather than where you are.

Our server then sends our team an alert with the outreach domain, the short name in the address after the domain, your IP address and its approximate location (city, region, and country), the name and domain of the organization that operates your network (and whether it looks like an internet, cloud, or VPN provider), the page, the type of device, operating system, and browser, your time zone and the local time there, your browser's language, our server's automatic guess whether the visit came from a person, an email security scanner, our team or one of our test links (made from the time zone, the device and browser, the network, the location's country, the short name in the address, and whether a second copy of the visit arrived), a random reference number, and, while our server can read Smartlead, the names and email addresses of up to three people we emailed from that outreach domain whose email was opened, or a link in it clicked, in about the 60 minutes before the visit (with a note when the domain of their email address is that of the organization that operates your network or of the organization the visit matched). For 15 minutes after an alert, our server adds a reply to it when Smartlead records an open or click of one of our emails from that outreach domain made in the 60 minutes before the alert or since, naming one more such person in each of up to three replies. If a second copy of the same visit (from the same outreach domain, with the same short name in the address, from the same network block, with the same user-agent string) reaches our server within 3 seconds, as happens when an email security scanner opens a link, our server sends no alert for it and instead adds a reply to the first alert saying so (unless the alerts about its organization were stopped, Section 10); for this it keeps a coded form of the outreach domain, the short name, the network block (the first three parts of an IPv4 address, or the first three groups of an IPv6 address), and the user-agent string, and the reference number, in its memory for 10 seconds. It sends these alerts to private Telegram chats: each team member's own chat with our alert bot (a member of our team's Telegram group who starts the bot receives them), or a private group of our team. When someone clicks a link in one of our emails (those links take them to goat.ai without a campaign tag, so the visit itself makes no alert), our server sends the same chats a message with that person's name and email address, which of our emails it was and from which outreach domain, when it was sent and when the link was clicked, and our guess whether a person or an email security scanner clicked it, unless a reply to an alert, or an alert about a visit from that outreach domain made within 10 minutes of the click and before our server learned of the click, already named that person. Once a day, our server sends the same chats a file that lists the alerts of the previous 24 hours (what they said, without the IP address or its location, with their reference numbers and the people they named), the email events of those 24 hours with the recipients' names and email addresses, for an event that an alert or a reply named, that alert's reference number (our guess), and, for a click, our guess whether a person or an email security scanner made it, and the email addresses of the people described above whose emails were unsubscribed, bounced, or blocked, or whose organization objected, that no earlier file has listed (after a restart of our server, everyone it then finds so). For this, our server keeps a list of the Telegram chat numbers of the team members who receive alerts; it holds no information about visitors. Our server also holds a list of the domains of the organizations we have emailed: a list we keep with no names or email addresses, and the domains of the email addresses it reads from Smartlead (common personal mail providers left out). If your network belongs to one of those organizations, judged by its domain (a network of example.com or de.example.com counts for example.com) and never for an internet, cloud, or VPN provider's network, or if the short name in the address names exactly one of them (as example names example.com), the alert also says that the visit probably came from someone at that organization, gives its domain, and says why.

Apart from the alert, our server keeps your IP address only in its memory: while it makes the alert, and for at most 10 minutes to limit how many of these requests one internet connection or network can send. It does not store your IP address in our database or write it to our application logs, and it does not send it to IPinfo or DB-IP (as with any request to our servers, the IP address can appear in our web server's error log if the request fails); it sends it to Telegram only in the alert. It keeps the location, the organization's name, your time zone, language, and user-agent string, and the device and browser named from it in its memory while it makes the alert, or, for a second copy of a visit, for at most 20 seconds while it waits to add the reply described above (apart from the coded form for 10 seconds described above), and the organization's domain while it makes the alert and for 30 minutes afterwards, so that repeated visits from the same network to the same address send one alert and many visits from one network do not keep notifying our team; except that, while our server reads Smartlead, it keeps what each alert it sent said (the organization's name and domain, your time zone, language, device, and browser, and the people it named included, but not the IP address, its location, or the user-agent string) in its memory for up to 26 hours, for the daily file described above. It writes none of them, nor a matched domain from our list, to our logs or database.

Later in the same visit, the page tells our server, with the reference number, each click and contact-form submission it records for analytics (those listed under "Website visit counting" above), by their type, the part of the page they were in and, where the event has them, the dataset, the link, preview tab or view chosen, and the form and the reason you picked, each with the time since the visit was reported, never anything you typed; the goat.ai pages you open (never the pages of our PolymarketBench section: once you open one, it stops reporting the clicks, contact-form submissions, pages, visible time, and scroll depth for that visit, and it reports nothing while you are on one; the separate replies described next can still follow after you leave it); and, each time you leave the page or switch away from it, how long the page was visible since the visit was reported and how far down the page you scrolled. Our server writes them into one reply to the alert in our team's Telegram chats, which it updates as more arrive, and it still adds a separate reply when you open our Hugging Face sample or the dataset viewer, open our request form, send one of our contact forms, or open our call booking page (see "Booking a call" below). For this it keeps the alert's reference number, the numbers of its messages and of those replies in our team's chats, and that list of actions in its memory for 2 hours, and their totals (how many clicks and pages, the visible time, and the deepest scroll) with the alert for the daily file, for up to 26 hours. The alerts, their replies, the messages about clicks on links in our emails, and the daily files stay in those chats, which, when an alert is sent, only our team members and our alert bot are in (Telegram stores the messages on its servers, Section 12); we do not delete them automatically, and we delete those about an organization that objects, and, when you ask us, those that name you, from the chats of our current team (Section 10). When someone leaves our team, we stop sending them alerts and ask them to delete their chat with our alert bot; alerts already in that chat stay in it until they do. An outreach visit that arrives on a page of our PolymarketBench section still makes an alert, which does not name the page (it shows the page as unrecorded); the page then reports nothing more about that visit to our server.

With the visit's other events, PostHog receives an outreach event that names the outreach domain, whether the campaign tags or the referring site showed it, and the same random reference number, and from then on each event the page sends PostHog during that visit (page views, and the clicks and contact-form submissions above) also carries the reference number, until you decline analytics cookies or turn on Global Privacy Control or Do Not Track. Our server also sends PostHog, with the same reference number and under no person's profile, its automatic guess described above whether the visit came from a person, an email security scanner, our team, or one of our test links, and nothing else from the alert; for a second copy of a visit, or a repeat of a visit it alerted about shortly before, which get no alert of their own, it makes that guess in the same way from what it already knows, without looking up the location, and for a visit from an organization that objected (Section 10) it sends none. PostHog does not receive the organization that operates your network (if the address you used names your company, that name is in the page address PostHog records, as described above; and if an alert or one of its replies names a person we emailed, PostHog holds the domain of that person's email address, which can be that of the same organization, with our guess described next). If an alert or one of its replies names a person we emailed, PostHog also receives, under that person's email address, our guess that the visit with that reference number was theirs (described above), so that the visit's events can be found from that person's profile; it is a guess, and it can be wrong. It reaches PostHog like the events above. The reference number lets our team find in PostHog the pages that visitor viewed and what they clicked (on that day, or, if you accepted analytics cookies, also on other visits from the same browser). Google Analytics does not receive this event or the reference number. For this, and for the replies to an alert described above, the page reads only its own address, the referring site, your browser's time zone and language, your analytics choice (the goat_consent cookie), your browser's Global Privacy Control and Do Not Track signals, whether your browser identifies itself as automated (so that we skip bots and link checkers), and, once the visit is reported, the events, pages, visible time, and scroll depth named above; it stores nothing on your device and sends our server only what is listed above. The page tells our server about the visit once it has been visible for about 2 seconds, without waiting for your answer to the analytics cookie banner. If by then you have declined analytics cookies on goat.ai, or if your browser sends a Global Privacy Control or Do Not Track signal, it does not, and no alert is made. Declining later, or turning on Global Privacy Control or Do Not Track later, also stops the page from telling our server what you do next in the current visit and from adding the reference number to its later PostHog events, and stops alerts for your later visits, but does not recall an alert or reply already sent; write to privacy@goat.ai and we will delete it. We do not make session recordings of a page you arrive at from our outreach domains (see "Analytics cookies" above). These choices and signals do not affect the records our email-sending service makes of our emails (a click on a link in our email is recorded by Smartlead before you reach goat.ai), or the messages and PostHog records our server makes from them; to stop those, unsubscribe through the link in our emails, or write to privacy@goat.ai.

We use these alerts, and the messages and records about our emails described above, to see which organizations and which of the people we emailed respond to our outreach emails and to decide whether to follow up with them. An alert contains your IP address and its approximate location and the device, browser, time zone, and language details described above. Apart from the names and email addresses of the people we emailed described above, it contains no email address, and no name other than the name of the organization that operates the network, the names of the city and region of that location, the short name in the address after the domain, and the domain of an organization we emailed that the visit matched; it describes the type of device and browser with labels such as "iPhone" or "Edge". On a home or mobile connection, the IP address and its location can relate to you rather than to an organization. Our email-sending service records when our emails are opened and which of their links are clicked; an alert names the people whose email was opened or clicked shortly before it (or, while our server cannot read that record, an alert that matched no organization suggests looking for our emails opened in the 30 minutes before it), and our team may also read an alert next to the names of the people we emailed at the organization the alert names. An alert can therefore show us, or let us guess, who among the people we emailed visited; that guess can be wrong (for example, when someone else used the same link, or an email security scanner opened the email). If you send one of our contact forms during a visit that caused an alert, the reply to that alert lets our team connect the alert with what you wrote in the form, including your name and email address. Likewise, if you book a call through one of our links during such a visit, the reference number in the message about the booking, and the reply to that alert saying that a call was booked, let our team connect the alert, with its IP address and approximate location, to your name and email address (see "Booking a call" below). We use this only to decide whether to follow up. We do not sell the alerts or use them for ads on other websites. Using them to decide whether to send a follow-up email is direct marketing, and if you object to it we always stop. We make them on the basis of our legitimate interests (Section 9), and you can object to them (Section 10, "Your right to object to outreach alerts"). Section 12 explains what sending the alerts to Telegram means for transfers.

Our team's outreach analytics tool: members of our team can look up the information described above about our outreach emails and the visits that come from them, and the calls booked with us (see "Booking a call" below), through our own outreach analytics tool, either themselves or through AI assistants they use for their work. The tool only reads what we already hold: it sends no email or message and changes no record. It gives counts by day, page, outreach domain, email step, and campaign, counts of page views and clicks on goat.ai and of the visits by our guess of who made them, and, for the people we emailed in the campaigns our server reads at the time, their name and email address, the domain of their email address, which of our emails they were sent, from which outreach domain and in which campaign, when they were sent, first opened, first clicked, and replied to, and how often they were opened and clicked, whether one was opened within two minutes of delivery (a sign of an email security scanner) and whether a click on its link looks like a person's or a scanner's (our guess from how soon after delivery it came), our guesses linking visits to them and the page views and clicks recorded in those visits, and the calls booked under their email address (the time of the call, where the link was, and the reference number, never the notes or other answers of the booking form). For a visit that arrives from our outreach domains, it gives what the alert and its replies said (including the totals of what the visitor did next), without the IP address or its location, for as long as our server keeps it (up to 26 hours), and after that the time, the reference number, the page, the outreach domain, and how the visit arrived from it, and, only when we linked the visit to a person we emailed, the short name in its address and the page views and clicks recorded in that visit. It never gives an IP address, a location, the text of our emails or of any reply, session recordings, or anything from the dashboard, and it never names, or links a visit to, someone who unsubscribed from our emails, whose address bounced, or whom we blocked, and never gives the records of our emails to them, or anything about anyone whose email address belongs to an organization that objected (Section 10), even while PostHog still holds it; a visit by someone who stopped our emails is still shown like any other, without naming them. The tool itself keeps no copy of what it gives, apart from holding the result of a query in our server's memory for up to a minute and its counts per campaign, which name no one, for up to an hour; where its answers can stay is described in Section 11. Only members of our team we name can use it, each with their own key, which we revoke when they leave. When a member of our team uses an AI assistant with the tool, the provider of that assistant, or of the AI model it runs on, receives the questions and the tool's answers, including the names and email addresses in them, and keeps them under its own terms in that member's conversations with the assistant. We use the tool, and these assistants, only to see which organizations and which of the people we emailed respond to our outreach emails and to decide whether to follow up with them.

Booking a call: the "Book a 15-min call" links on our dataset pages (goat.ai/gatc and goat.ai/gatc-s) open our booking page on Cal.com (Cal.com, Inc., United States) in a window on the same page (or in a new tab, if you open the link in one, use the link in the window that opens it there, or click it before the page has finished loading), and the link on our contact page, when you open it about datasets, opens it in a new tab; in this Policy, "our booking page" means it in either form. Nothing from Cal.com loads on our dataset pages until you click one of these links. When you open it, the link gives Cal.com the campaign tags in the page's address (utm_source, utm_medium, utm_campaign, and utm_content, each when it is a short plain name; in place of any that is missing or is not, our own: one naming the page, two fixed ones, and one naming where the link was; for a visit that arrives from our outreach domains, utm_content can be the short name for your company in the address after the domain, described above), where on our page the link was, and, for a visit that arrives from our outreach domains and only while you have not declined analytics cookies and your browser sends no Global Privacy Control or Do Not Track signal, its reference number (described above). The link carries nothing else about you. We do not fill in the booking form for you. Like any website you open, Cal.com also receives your IP address and your browser's details, and so do the services Cal.com's page loads content from or sends reports to (for example Google's servers, for the profile picture of the member of our team you book with, and Sentry, which receives the reports Cal.com's page makes on how it performs and on its errors and can receive the address of our booking page, with the tags and the reference number it carries). In the window on our dataset pages, our booking page is still Cal.com's own page: what you enter in it goes from your browser to Cal.com, not through our servers. Cal.com's page can also tell our page what happens in it, such as that the calendar has loaded or that a booking was made, with its details; our page's own code ignores these messages: it keeps none of them and sends none of them anywhere. Cal.com may keep its own cookies and similar entries in your browser for it, such as a security cookie, __cf_bm, that expires after 30 minutes, a cookie named uid that lasts until you close your browser, and an entry Cal.com uses to hold a time while you book. You can close the window at any time; closing it does not cancel a booking. After you close it, our booking page stays loaded in it, hidden, for up to five minutes, so that opening the window again keeps what you entered; then our page removes it.

Cal.com processes what you enter in its booking form, and the booking, for us as our service provider, under the data processing agreement that is part of its terms of service (for its own website, its own privacy policy applies). The bookings, with everything you entered, including any guests you add, stay in our Cal.com account, where our team can see them, until we delete them. Cal.com also passes each booking, with your name and email address, the time of the call, what you entered in the booking form, and any guests you add, to any calendar and video-call services connected to our Cal.com account, which add the call to the calendar of the member of our team you book with and create its meeting link; the calendar entry stays there until we delete it. Unless we have turned these emails off in Cal.com, Cal.com also emails each booking, rescheduling, and cancellation, with what you entered, to the member of our team you book with, in our mailboxes (Google Workspace, Section 12), where it stays until we delete it.

When you book, cancel, or reschedule a call, Cal.com sends our server the whole booking record: your name and email address, your time zone and language, the time of the call, your notes and any reason you gave for rescheduling or cancelling, your answers to any other questions of the booking form (such as a phone number, if it asks for one), the names and email addresses of any guests you add, the meeting location and its link, details of the member of our team you book with, Cal.com's own number and status for the booking, and, as far as Cal.com passes them on, those tags, where the link was, and that reference number. Our server uses only your name, your email address, your time zone, the time of the call, your notes, Cal.com's number and status for the booking, the campaign tags, where the link was, and the reference number; it discards the rest as soon as it has read the record, without keeping, logging, or sending it.

Our server then sends our team a Telegram message with your name and email address, the domain of your email address (unless it is one of the common personal mail providers), the time of the call in UTC and in your time zone, the utm_source and utm_content tags and where the link was, the reference number with a link to that visit's events in PostHog, your notes, and, if we sent you one of our outreach emails and our server can read its record (see above), which of our emails you were sent and when it was sent, first opened, first clicked, and replied to. It sends this message to the same chats as our outreach alerts and also to our team's main Telegram group, which receives no other message from our server apart from the messages about contact-form submissions ("Support and communications data" above), even if you have stopped our outreach emails or your organization has objected to our outreach (Section 10), because you booked the call yourself (the message then leaves out the record of our emails). If our server still holds the alert of the visit with that reference number (up to 2 hours), it adds a reply to that alert, in the chats that hold it, saying that a call was booked, rescheduled, or cancelled, and for when. It also sends PostHog the booking, rescheduling, or cancellation under the email address typed in the booking (not your name; because Cal.com does not verify that address, the record says it is unverified and changes nothing else PostHog holds under that address), with the time of the call, where the link was, the utm_source and utm_content tags, and the reference number, but never your notes, and nothing at all if our server knows that you stopped our outreach emails, or if your email address belongs to an organization that objected (Section 10). So PostHog groups it with the records of our emails to you, if any, and with that visit's events, which the reference number then links to your email address. Declining analytics cookies, or a Global Privacy Control or Do Not Track signal, keeps the reference number out of a booking link you open after that, so that neither the page nor the booking adds a reply to the alert of your visit; it does not stop the Telegram message or the PostHog record of a booking, which we make on the basis of our legitimate interests (Section 9); to have them deleted, write to privacy@goat.ai (Section 10). Nothing about a booking is stored in our database: our server keeps what it uses only in its memory while it makes the message and sends PostHog its record (usually a few seconds, at most a few minutes), except Cal.com's number for the booking and whether it was booked, rescheduled, or cancelled, which it keeps for 24 hours so that a booking Cal.com sends twice makes one message, and it writes none of it to our logs. The booking messages stay in our team's chats like the alerts (Section 11).

No category pre-screening. GOAT does not pre-screen, filter, classify, or block trace data by content category before capture, and does not detect or remove special-category, health, payment-card, government-ID, biometric, precise-geolocation, or children's data prior to ingestion. The contributor is solely responsible for ensuring such data is not connected or submitted.

3. Where the Information Comes From

We collect information directly from you; from your organization's administrators and users; from capture agents you install and Sources you connect; from observability providers and code tooling you authorize; and automatically from your use of the Service and of our marketing website, including through our analytics providers (Section 2). For a visit that arrives from our outreach domains, a database from IPinfo tells us which organization operates the network your IP address belongs to, and, for an alert, a database from DB-IP gives the approximate location of your IP address (city, region, and country) (Section 2, "Outreach emails and visits that come from them"). If we sent you one of our outreach emails, our email-sending service, Smartlead, tells our server your name and email address, which of our emails you were sent and when, when they were opened, clicked, and replied to, and whether you unsubscribed, your address bounced, or we blocked you (Section 2). If you book a call on our booking page on Cal.com, Cal.com sends our server the booking (Section 2, "Booking a call"). Where we use payout, accounting, payment, or (in the future) fraud or sanctions-screening vendors, we may also receive information from them.

4. How We Use Information

We use personal information and trace data to: provide, maintain, secure, and support the Service; authenticate users and manage organizations and access; calculate program metrics and Cashback, process payouts, and review for suspected abuse or fraud (which we do manually and on a best-effort basis; we do not represent that we operate automated fraud, sanctions, or abuse-detection systems); apply automated, best-effort redaction and deidentification techniques to trace data (which are automated and probabilistic, may be partial, are not guaranteed to run on or remove sensitive content from any particular trace, and do not include any human-review step); create analytics, benchmarks, reports, and product features; comply with legal obligations and enforce our agreements; communicate with you; measure and improve our marketing website and, if you turn on Product analytics, the dashboard, and find and fix errors in them (Section 2); send emails about our datasets to people at businesses and research organizations (Section 2, "Outreach emails and visits that come from them"); see which organizations and which of the people we emailed respond to our outreach emails and decide whether to follow up with them (Section 2, "Outreach emails and visits that come from them"); prepare for and hold the calls you book with us (Section 2, "Booking a call"); and create and license De-Identified Data and Corpus Products from the trace data you connect.

5. How Our Trace-Data Processing Works

When trace data may contain personal information, GOAT generally processes that raw data on behalf of the contributor that connected the Source, under that contributor's instructions and the DPA. Connecting a Source, installing a capture agent, configuring a project, committing capture configuration, or sending traffic to a project slug you control authorizes and instructs GOAT to ingest and durably capture trace data (including in encrypted, un-redacted raw form) immediately and irreversibly. That capture begins on connection, is not conditioned on any further opt-in, and cannot be reversed for data already captured. You are responsible for the lawful basis and for all required notices and consents for everyone whose personal information appears in trace data you connect — including teammates and end users — and for all traffic sent to any slug you control (including anonymous ingest, which is enabled by default).

Raw trace content is captured and stored in encrypted, access-segregated form at ingestion, before and independently of any redaction or deidentification. GOAT applies an automated, best-effort redaction and deidentification process before content is promoted to its queryable store, but does not warrant that it runs on, detects, or removes data from any particular trace or field, or that it renders any data non-personal; the process is probabilistic, includes no human-review step, and raw content is retained in encrypted, access-segregated storage independently of redaction. Authorized GOAT processing systems and a limited number of authorized personnel may access decrypted Raw Trace Data on a need-to-know, access-logged basis for processing, redaction, validation, support, security, fraud-investigation, and legal-compliance purposes. If and to the extent GOAT transforms trace data into De-Identified Data that it designates as no longer constituting personal information, GOAT may use that data as part of its own commercial corpus and products as described here and in the Terms.

6. De-Identified Data and Licensing of Datasets

A core part of GOAT's business is creating datasets, benchmarks, and research products from De-Identified Data and licensing them to recipients such as frontier AI labs, investment-research and alt-data teams, market researchers, academics, and other approved commercial or research recipients.

We design and operate our pipeline with the goal that outbound products comprise De-Identified Data, but we do not warrant that every record in any dataset meets the deidentification standard, and you acknowledge that residual linkability may remain. De-Identified Data may retain coarse technical metadata — including model identifiers, token counts, timing, cost, content fingerprints, session identifiers, provider response identifiers, observability end-user identifiers, and repository or source labels — that may remain linkable to a contributor or account. We maintain internal mappings linking pseudonymous contribution identifiers to contributor accounts (and resolve traces to the responsible account via organization membership) for attribution, team visibility, payout, audit, and fraud-prevention; these mappings are retained and controlled by us as a controller and are not provided to dataset recipients. For information we treat as deidentified under California law, GOAT (i) maintains reasonable technical and organizational measures intended to prevent reidentification, (ii) commits to maintaining and using such information only in deidentified form and not to attempt to reidentify it except as permitted by law, and (iii) contractually requires recipients of Corpus Products not to attempt to reidentify any individual. We do not warrant that recipients will comply, and we are not responsible for a recipient's breach. You assume the risk of, and agree not to hold GOAT liable for, any reidentification arising from metadata or residual identifiers GOAT was authorized to retain or process.

7. How We Disclose Information

We may disclose personal information to: service providers and subprocessors that host, secure, store, transmit, or support the Service (Section 12); analytics providers that process analytics data for us — PostHog (website visit counting without cookies; with your consent, website analytics cookies, session recordings and error reports, and dashboard Product analytics, including dashboard error reports; and the records of our outreach emails and our guesses linking visits to the people we emailed, and the calls booked on Cal.com) and Google (website analytics, only with your consent) (Sections 2 and 12); Netlify, which hosts our outreach domains, Smartlead, which sends our outreach emails, Cal.com, which runs our call booking page and sends our server the calls booked on it, any calendar and video-call services connected to our Cal.com account, to which Cal.com passes the bookings, and Telegram, on which our team receives outreach alerts, the messages about clicks on links in our emails, about calls booked on Cal.com, and about contact-form submissions, and the daily outreach file (Sections 2 and 12); providers of the AI assistants our team uses with our outreach analytics tool, and of the AI models those assistants run on, which receive what our team looks up with it (Section 2, "Our team's outreach analytics tool"); payout, wallet, accounting, and payment vendors, and — where we elect to use them in the future — fraud, sanctions, identity, or tax vendors; professional advisors, auditors, insurers, and financing counterparties; law enforcement and regulators where required or to protect rights or safety; acquirers or counterparties in a merger, financing, acquisition, restructuring, or asset transaction; the contributor organization that controls a workspace (if you are a user under that organization); and approved recipients of Corpus Products, which we intend to comprise De-Identified Data.

8. California Privacy Disclosures

If you are a California resident, you may have rights to know, access, correct, and — subject to the significant limits in Section 10, including that deletion is manual, best-effort, and may be technically infeasible for encrypted raw, immutable, or already-distributed data — delete personal information, and to opt out of its "sale" or "sharing." You have a right not to be discriminated against for exercising these rights.

GOAT licenses De-Identified Data and Corpus Products and does not treat data that satisfies the legal standard for deidentified or aggregate information as personal information. To the extent any disclosure of personal information, or of information that remains reasonably linkable to you, qualifies as a "sale" or "sharing" under California law, you may opt out by emailing privacy@goat.ai; we will honor verified opt-out requests, which will at least stop future inclusion of your information in Corpus Products. We do not currently process Global Privacy Control or other browser-based opt-out preference signals automatically as an opt-out of such a sale or sharing. On goat.ai, we treat a Global Privacy Control or Do Not Track signal as declining analytics cookies (Section 2). We use website and dashboard analytics to measure and improve our website and dashboard and, for visits that arrive from our outreach domains and for the people we emailed, to see which organizations and which of the people we emailed respond to our outreach emails (Section 2); we do not use them for cross-context behavioral advertising.

Notice of Financial Incentive. The cashback program is a financial incentive tied to your contribution of trace data. Material terms: the personal information implicated is the trace data you connect (Section 2); you opt in by connecting Sources and contributing eligible usage, and you may withdraw at any time by disconnecting Sources or emailing privacy@goat.ai. Good-faith value estimate and method: GOAT calculates cashback from the measured output-token volume of your eligible contributions, multiplied by the model's published output price and an applicable program rate and multiplier, as described in the Terms; GOAT uses this same measure as its good-faith estimate of the value of the data to GOAT, which varies by contribution and is not a guaranteed amount. The incentive is reasonably related to the value GOAT derives because both are computed from the same usage measure. GOAT may revise this methodology. If you opt out of the commercialization-linked processing the program depends on, you may be unable to continue participating or to receive future cashback.

Categories of California personal information we may collect and disclose include identifiers, professional or employment information, internet or network activity, commercial information, account credentials (hashed/encrypted), payout instruments, support records, and inferences for program administration, anti-fraud, or analytics. We do not knowingly sell or share the personal information of children under the applicable age threshold (Section 14). For website and dashboard analytics (Section 2), we collect identifiers: your IP address, which we pass on in transit and store only on the audit event of a change of your dashboard analytics choice; daily visitor codes; cookie and browser-storage identifiers; and, for the dashboard, your GOAT user ID and project ID. We also collect internet or other electronic network activity information (pages viewed and how long, referring site, clicks, setup steps, error reports, browser and device details, and, if you accept analytics cookies on goat.ai, recordings of how you use its pages) and approximate geolocation (country and region, derived by Google Analytics only if you accept analytics cookies). We collect them from your browser and our servers, to measure and improve our website and dashboard and to find and fix errors in them. We disclose them only to PostHog and Google, which process them to provide analytics services to us, and we retain them as described in Section 11. For outreach alerts (Section 2), we collect identifiers (your IP address, which our server reads to look up the organization that operates your network and the approximate location of the address, keeps in its memory for at most 10 minutes, sends to our team in the alert, and does not keep in our database or application logs; a random reference number; and, if we emailed you, your name and email address), internet or other electronic network activity information (which of our outreach domains you came from and how, the short name in the address after the domain, the page, the type of device, operating system, and browser named from your user-agent string, your time zone and browser language, and, later in that visit, the clicks and contact-form submissions the page records for analytics, with when they happened, the goat.ai pages you open, how long the page was visible, and how far you scrolled; and, if we emailed you, which of our emails you were sent, from which outreach domain and when, and when it was first opened, first clicked, and last replied to, and how often it was opened and clicked), and approximate geolocation (the city, region, and country that a database from DB-IP gives for your IP address), and we draw inferences from your IP address and the short name in the address (the name and domain of the organization that operates your network, whether it looks like an internet, cloud, or VPN provider, and whether the visit probably came from an organization we emailed; and, from your time zone, device, browser, network, location, the short name, and whether a second copy of the visit arrived, whether the visit came from a person, an email security scanner, our team, or one of our test links; and, if we emailed you, our guess that a visit was yours and, for a click on a link in one of our emails, whether a person or an email security scanner made it, and, for an open of one, whether it was probably automatic). We collect them from your browser, our servers, databases from IPinfo and DB-IP that our server queries itself, and our email-sending service, Smartlead, to send our outreach emails, to see which organizations and which of the people we emailed respond to them, and to decide whether to follow up with them. We disclose them to Telegram (your IP address and its approximate location, the outreach domain, the short name in the address after the domain, the name and domain of the organization that operates your network and whether it looks like an internet, cloud, or VPN provider, whether the visit probably came from an organization we emailed and which one, the page, the type of device, operating system, and browser, your time zone, local time, and language, our guess whether the visit came from a person, an email security scanner, our team, or one of our test links, the reference number, and what you did next on goat.ai during that visit and when; and, if we emailed you, your name and email address, in an alert or a reply about a visit shortly after you opened or clicked one of our emails, in a message when you click a link in one of our emails (with which email it was, when it was sent and clicked, and our guess whether a person or an email security scanner clicked it), and in the daily file, with that day's email events and our guesses, and, once you stop our emails, your email address); to PostHog (the outreach domain, how you came, and the reference number, with the outreach event and with that visit's later events, our guess whether the visit came from a person, an email security scanner, our team, or one of our test links, and, as with other events, the page address with its campaign tags and, in transit, your IP address; and, if we emailed you, your name, your email address and its domain, our campaign, which of our emails you were sent and when they were sent, first opened, first clicked, and replied to, and how often they were opened and clicked, with our guess whether an open or a click was automatic, and our guess linking a visit's reference number to you); and, if you accept analytics cookies, to Google (the page address with its campaign tags), as described in Section 2. Netlify, which hosts our outreach domains, receives your IP address, your browser's user-agent string, and the address you open. Smartlead, which sends our outreach emails, receives your IP address and your browser's or mail program's user-agent string when you open one of our emails or click a link in it (Section 2). For calls you book with us (Section 2, "Booking a call"), we collect identifiers (your name, your email address, Cal.com's number for the booking, and the reference number of the visit, if the booking carries one), the time of the call, the time zone and language you booked from, your notes and your other answers in the booking form, the names and email addresses of any guests you add, and internet or other electronic network activity information (the campaign tags of the link and where on our page it was), from Cal.com, to prepare for and hold the call. Cal.com holds the bookings for us as our service provider, and our server discards what it does not use (Section 2). We disclose to Cal.com, from our link, the campaign tags (for a visit that arrives from our outreach domains, utm_content can be the short name for your company in the address after the domain), where the link was, and the reference number; Cal.com passes the booking to any calendar and video-call services connected to our Cal.com account and can email it to our mailboxes (Section 2). We disclose to Telegram your name, your email address and its domain, the time of the call and your time zone, the utm_source and utm_content tags, where the link was, the reference number, your notes, and, if we emailed you, the record of our emails to you; and to PostHog the email address typed in the booking, the time of the call, where the link was, the utm_source and utm_content tags, and the reference number. When our team looks up this outreach and booking information through our outreach analytics tool with an AI assistant, we disclose to the provider of that assistant, or of the AI model it runs on, what the tool gives (Section 2, "Our team's outreach analytics tool"), never an IP address or a location. We retain them as described in Section 11. Do Not Track: on goat.ai we treat it as declining analytics cookies (Section 2); the dashboard does not respond to it, because dashboard analytics run only if you turn them on. If you accept analytics cookies on goat.ai, Google collects information about your use of goat.ai only; we do not allow third parties to track you across other websites through our website or dashboard.

9. GDPR and UK GDPR Legal Bases

Where GOAT acts as a controller, we rely on one or more of the following legal bases.

Contract: to create and administer accounts, provide the Service, and process payouts.

Legitimate interests:

  • to secure the Service, prevent and investigate fraud, improve reliability, support users, learn how people who contact us found goat.ai, enforce agreements, and defend claims, including IP logging on security events;
  • to count visits to our marketing website without cookies, so that we know how many people visit, which pages they read, and how they found us, and can improve the site (you have the right to object, Section 10);
  • to email people at businesses and research organizations about our datasets (direct marketing; you can object at any time, and we always stop, Section 10);
  • to see which organizations and which of the people we emailed respond to our outreach emails and decide whether to follow up with them. For a visit that arrives from our outreach domains, we look up the network it comes from and, for an alert, the approximate location of its IP address, record an outreach event and our automatic guess of who made the visit in PostHog (and add its reference number to that visit's later PostHog events), and alert our team, with the visitor's IP address and its approximate location, the visitor's device, browser, time zone, and language, our automatic guess whether the visit came from a person, an email security scanner, our team, or one of our test links, what the visitor did next on goat.ai, the organization that operates the visitor's network and, where that network or the short name in the address matches the list of organizations we emailed, that organization, and the people we emailed whose email was opened or clicked shortly before the visit. For the people we email, we read from our email-sending service when our emails to them were sent, first opened, first clicked, and replied to, record those events, our guess whether an open or a click was automatic, and our guesses linking visits to them in PostHog under their email address, and send our team a message when they click a link in one of our emails and a daily file that names them. Our team may also read an alert next to the names of the people we emailed at that organization, which can point to the person we emailed. Our team may also look up this information through our outreach analytics tool, including with AI assistants (Section 2, "Our team's outreach analytics tool") (you have the right to object, Section 10);
  • to prepare for and hold the calls people book with us on Cal.com: we send our team the booking and record it in PostHog under the email address of the person who booked (you have the right to object, Section 10);
  • to keep records of your analytics choices so that we can demonstrate them.

Consent: where we rely on it for specific optional processing, including analytics cookies, session recordings and error reports on goat.ai, and dashboard Product analytics, including its error reports. You can withdraw it at any time, as easily as you gave it, without affecting processing before you withdrew it.

Legal obligation: tax, sanctions, accounting, regulatory, or law-enforcement obligations.

Where GOAT processes raw trace personal data on behalf of a contributor under the DPA, the contributor is responsible for identifying the lawful basis.

10. Your Rights and Their Limits

Depending on your location and our role, you may have rights to access, correct, delete, port, object to, or restrict processing, or to withdraw consent. Submit requests by emailing privacy@goat.ai. We may need to verify your identity and authority before acting, and we will acknowledge and respond within the timeframes required by applicable law (for example, 45 days under California law and one month under the GDPR/UK GDPR, each extendable as permitted).

Your right to object to website visit counting. We count visits to goat.ai without cookies on the basis of our legitimate interests (Section 9). You have the right to object to it. Because the count is recorded under a code that changes every day and is not linked to you, we cannot single out your visits in order to stop counting them, and goat.ai has no separate setting for it: declining analytics cookies, or sending a Global Privacy Control or Do Not Track signal, does not stop it. To stop your browser from sending these events, you can block requests to goat.ai/gx/ (for example, with a content-blocking browser extension). You can also write to us at privacy@goat.ai.

Your right to object to outreach alerts. We make the outreach alerts described in Section 2 on the basis of our legitimate interests (Section 9), and you have the right to object to them at any time; where we use them for direct marketing (to decide whether to send a follow-up email), we always stop when you object. A Global Privacy Control or Do Not Track signal, or declining analytics cookies on goat.ai, stops the alerts as described in Section 2 ("Outreach emails and visits that come from them"); it does not recall an alert or reply already sent. You can also write to privacy@goat.ai, naming your organization: we then stop the alerts about it (for its network, and for an address after our outreach domains that names it), delete the alerts about it and their replies from the Telegram chats of our current team (and ask anyone who received them and has since left our team to delete their chat with our alert bot), no longer use its visits to decide whether to contact it, stop keeping, showing, and sending PostHog the records of our emails to anyone whose email address belongs to it, and stop sending our outreach emails to the address you write from (tell us any other address of yours we emailed). Stopping our emails means adding your address to the block list of our email-sending service and unsubscribing it from our campaigns, which has the same effect as unsubscribing through the link in our emails: from then on (within a few minutes) our server no longer names you in new alerts or messages, other than in the daily file, in the messages about a call you book with us yourself (Section 2, "Booking a call"), and in the message about a contact form you send us yourself (Section 2, "Support and communications data"), and no longer sends PostHog new records of our emails to you, and we delete your person in PostHog, with the records of our emails and our guesses it holds, by hand, from that daily file (if it does not list you, for example because you stopped our emails in a campaign our server no longer reads, write to privacy@goat.ai and we will delete it). Once our server has recorded that you stopped our emails or that your organization objected, our outreach analytics tool no longer names you, links a visit to you, or gives the records of our emails to you, and gives nothing about an organization that objected; your visits are still shown like any other visit (Section 2, "Our team's outreach analytics tool"). Messages that already name you stay in our team's chats unless you ask us to delete them (write to privacy@goat.ai); answers our tool already gave an AI assistant stay in the conversations of the member of our team who asked. If you stop our emails, your visits to goat.ai are still handled like any other visit (Section 2). After an organization objects, our server still answers its visits like any other, so they are still counted in PostHog (Section 2), and where we know only its network, our server still looks that network up in order to recognize it; it makes no alert. If you booked a call with us, you can ask us at privacy@goat.ai to delete the messages about your booking from our current team's chats (our team's main Telegram group included), the booking from PostHog, the booking from our Cal.com account and any entry for it in the calendar it was added to, and Cal.com's emails about it from our mailboxes; to change or cancel the call itself, use the links in Cal.com's confirmation email.

Withdrawing analytics consent. On goat.ai, use "Analytics preferences" in the footer; in the dashboard, use the "Product analytics" switch in the account menu. Withdrawal takes effect as described in Section 2.

Analytics data and your requests. Website analytics data is recorded under a daily code or a random browser identifier that we do not link to you, so we generally cannot find it to answer an access, correction, or deletion request (GDPR Article 11). Outreach alerts are different: if you tell us your organization and the date of the visit, we can find its outreach alert and, with the alert's reference number, the PostHog events of that visit (the outreach event, our server's guess of who made the visit, and every later event of that page that carries the number). We will tell you what they contain, delete the alert and its replies from our current team's chats, and stop the alerts as described above. If we emailed you, we can also find, by your email address, the person PostHog holds for you (the records of our emails and our guesses linking visits to you), tell you what it contains, and delete it; so that our server does not create it again from Smartlead's records, we also stop our outreach emails to you (as described above). If you booked a call, we can find by your email address the booking PostHog holds and, with its reference number, the events of the visit it came from; we will tell you what they contain and delete the booking. PostHog does not let us delete individual events of this kind, so those events stay in PostHog as described in Section 11 (including any short name for your company in their page address), without the alert that linked them to your organization's network. Dashboard analytics data is linked to your GOAT user ID, and we handle requests about it like requests about your other account data, subject to the limits below.

Complaints. You also have the right to lodge a complaint with a data protection supervisory authority, in particular in the EU or EEA country where you live or work, or with the UK Information Commissioner's Office.

These rights are subject to important limits that reflect how the Service works:

  • No self-service erasure/export. We do not provide self-service deletion or data export. Requests are handled manually, on a best-effort basis, subject to technical feasibility.
  • Immutable records. Certain records (consent, rights, provenance, audit, and payout-ledger records, including the actor IP recorded on audited events) are maintained as immutable, append-only ledgers for security, fraud-prevention, payout integrity, and legal compliance. Where we cannot delete or correct such a record without undermining those purposes, we may restrict processing instead.
  • Raw and derived data. Raw trace data is retained encrypted and may not be individually or promptly deletable; De-Identified Data and Corpus Products already derived, and datasets already exported or licensed, are excluded from deletion and return obligations.
  • Forward-only revocation. Withdrawing consent or changing settings applies only to data captured afterward and does not recall data already captured or used.
  • Where an exception applies, we will delete what we can and tell you what we cannot delete and why.
  • Processor role. Where we hold data only as a processor on behalf of a contributor, we may direct you to that contributor, though we may assist where required.
  • Third-party data subjects. Where you appear in trace data a contributor connected (for example, as a teammate, employee, contractor, or end user of a contributor's product), the contributor — not GOAT — is the controller and is responsible for receiving and responding to your request; we will direct you to that contributor and our role is limited to commercially reasonable, technically feasible assistance to the contributor.

11. Retention

We retain personal information and trace data for as long as needed for the purposes in this Policy, and in some cases on a long-term or indefinite basis. Encrypted raw trace data is retained for as long as necessary for re-processing, provenance, security, audit, and fraud-prevention, is access-segregated, and is reviewed for deletion when the contributor relationship ends and no legal-hold, fraud-investigation, or audit need remains, or when the data can no longer support re-derivation or provenance. Audited security and access events, including the actor IP recorded on those events, are stored in append-only logs and retained on a long-term basis for security, integrity, and fraud-prevention. Apart from the analytics data and cookies described below, and the in-memory limits on our servers described in Section 2, we do not operate a fixed deletion timer; only short-lived authentication grants expire automatically. Analytics retention: the goat_consent cookie expires after 180 days, and the Google Analytics cookies 13 months after your last visit. PostHog's storage entries on goat.ai have no expiry date of their own: we delete them when you decline, or the next time you visit goat.ai after your choice has expired, and you can delete them yourself in your browser at any time. PostHog keeps analytics events, and the profile that groups a dashboard user's events, for as long as we keep the PostHog project that holds them; our current PostHog plan lets us query events for 1 year after collection. Error reports are events and are kept the same way. Our current PostHog plan lets us view a session recording for up to 30 days after it is made. Google Analytics deletes event-level and user-level data about 14 months after collection; aggregated reports that identify no one, such as daily visits per page, are kept as long as our Google Analytics property exists. Records of your dashboard analytics choices and their audit events are append-only and are retained with your account's other legal records. Outreach alerts: alerts, with the IP address and approximate location in them, stay in our team's Telegram chats with no fixed deletion time, as our record of how organizations responded to our outreach; we delete the alerts about an organization that objects, and their replies, from our current team's chats (Section 10; someone who leaves our team is asked to delete their chat with our alert bot, Section 2), and our servers keep no copy of them beyond the 26 hours described in Section 2. Our outreach analytics tool keeps no copy of what it gives, apart from holding the result of a query in our server's memory for up to a minute and its counts per campaign, which name no one, for up to an hour; what it gave an AI assistant stays in the conversations of the member of our team who asked, and with the provider of that assistant or of its AI model under that provider's terms, with no fixed deletion time set by us. The messages about clicks on links in our emails, the messages about calls booked on Cal.com and about contact-form submissions (in the same chats and in our team's main Telegram group) and the daily files stay in the same chats in the same way. Our server keeps nothing about a booking beyond the 24 hours described in Section 2, PostHog keeps the bookings like other analytics events, and the bookings stay in our Cal.com account with no fixed deletion time (we delete one when you ask, Section 10); the calendar entries made for a booking stay in the connected calendar, and Cal.com's emails about a booking in our mailboxes, with no fixed deletion time either. Our server keeps what it reads from our email-sending service only in its memory (the record of each email for as long as that email is part of a campaign we read, until our server restarts, and the email events it makes from them for up to 26 hours), and PostHog keeps those events, with our guesses linking visits to the people we emailed, like other analytics events (above), except that we delete by hand the person of anyone who stops our emails or objects (Section 10). PostHog keeps the person it holds under the email address of someone we emailed (name, email address, the domain of that address, and our campaign) or who booked a call (the email address alone) for as long as we keep the PostHog project, unless we delete it as described in Section 10. Smartlead keeps our emails, the replies to them, and its records of them in our Smartlead account, with no fixed deletion time, until we delete them. The outreach event in PostHog, our server's guess about the visit, and the visit's later events that carry its reference number, are kept like other analytics events (above). Netlify, which hosts our outreach domains, keeps its records of requests to them under its own retention practices; we do not copy them. Our subprocessors may maintain their own backups and disaster-recovery copies under their standard practices; GOAT does not operate an independent backup or point-in-time-recovery system, and we may be unable to remove personal information from any subprocessor-maintained copies.

12. Subprocessors and International Transfers

We use the following subprocessors to operate the Service: Supabase (managed Postgres, authentication, and authentication-email delivery; AWS-hosted); Cloudflare (R2 encrypted object storage and dataset exports; storage region "auto"); Upstash (Redis queues, cache, and rate-limiting); Resend (transactional email); Google (OAuth sign-in identity provider); our compute host (currently DigitalOcean, with servers in New York, United States), on which we also run our own trace/span data store (self-hosted ClickHouse); and Coinbase (used manually, off-platform, to send payouts; GOAT performs no automated integration and no sanctions/KYC screening). For website and dashboard analytics we also use PostHog (PostHog, Inc.; PostHog Cloud US, which stores data in the United States) and Google Analytics 4 (Google LLC) (Section 2), and PostHog also receives the records of our outreach emails and the calls booked on Cal.com described in Section 2; they receive no trace data and are not Subprocessors under the DPA. For the outreach alerts described in Section 2 we also use Netlify (Netlify, Inc., United States), which hosts our outreach domains and receives the IP address, user-agent string, and requested address of each visit to them; Smartlead (521 Products Pty Ltd, Australia), which sends our outreach emails, records when they are opened, clicked, and replied to (it learns of an open from a small image in the email and of a click from a link that passes through it, and receives the IP address and user-agent string of the device or service that loads the image or follows the link), and from which our server reads those records; Cal.com (Cal.com, Inc., United States), which runs our call booking page (the "Book a 15-min call" links, including the window on our dataset pages, Section 2), holds the bookings for us as our service provider, passes them to any calendar and video-call services connected to our Cal.com account, and sends our server the calls booked, rescheduled, or cancelled on it; Telegram (Telegram Messenger Inc.), on which our team receives the alerts, the messages about clicks on links in our emails, about calls booked on Cal.com, and about contact-form submissions, and the daily outreach file; and databases from IPinfo and DB-IP that our server downloads and queries itself, so neither receives information about our visitors (IP address data is powered by IPinfo; IP geolocation by DB-IP, licensed under CC BY 4.0); and the AI assistants our team may use with our outreach analytics tool, whose providers, and the providers of the AI models they run on, receive what our team looks up with it (Section 2). None of them receives trace data, and none is a Subprocessor under the DPA. We receive email, including contact-form messages, in Google Workspace (Google LLC), which hosts our mailboxes. We do not currently use a payment processor or a third-party fraud/sanctions-screening provider. A current subprocessor list is available on request at privacy@goat.ai, and we provide notice of material changes as described in the DPA.

Personal information may be processed in the European Union, the United States, and other countries, as determined by our hosting and subprocessor configuration. Apart from analytics data sent to PostHog (including the records of our outreach emails and the calls booked on Cal.com), the names and email addresses of the people we email held by Smartlead, the outreach alerts and the other messages our server sends to Telegram, the bookings Cal.com passes to the calendar and video-call services connected to our Cal.com account, and what our outreach analytics tool gives the AI assistants our team uses (all described below), where a transfer requires a transfer mechanism, we rely on, or will put in place, an appropriate mechanism such as the EU Standard Contractual Clauses or the UK International Data Transfer Addendum; a copy of the relevant safeguards is available on request at privacy@goat.ai. Analytics data passes through our servers in the United States on its way to PostHog, which stores it in its US region; analytics data collected before we moved to that region on October 14, 2026 remains in PostHog's EU region (Frankfurt, Germany). PostHog, Inc. and Google LLC are U.S. companies, and they and their own subprocessors may access or process analytics data in the United States and other countries. The records of our outreach emails that our server sends to PostHog, which include the names and email addresses of the people we emailed, wherever they live, and the calls booked on Cal.com, under the email address typed by the person who booked, reach PostHog's US region in the same way, and we name no transfer safeguard for them. Smartlead (521 Products Pty Ltd) is an Australian company; it holds the names and email addresses of the people we email, our emails to them and their replies, and the records of our emails (including the IP addresses and user-agent strings it receives when our emails are opened or their links clicked), and we name no transfer safeguard for them either. Cal.com, Inc. is a U.S. company; it holds what you enter on our booking page, including the names and email addresses of any guests you add, and the bookings for us under the data processing agreement that is part of its terms of service, which, as Cal.com states, relies on the EU–U.S. Data Privacy Framework and its UK Extension (Cal.com, Inc. is certified under them) and on the Standard Contractual Clauses with the UK Addendum and the Swiss amendments. Cal.com passes each booking to any calendar and video-call services connected to our Cal.com account; we name no transfer safeguard for them. The providers of the AI assistants our team uses with our outreach analytics tool, and of the AI models they run on, can process what it gives them, including the names and email addresses of people we emailed wherever they live, in the United States and other countries, and we name no transfer safeguard for them. Google processes Google Analytics data under its data processing terms, which rely on the EU–U.S. Data Privacy Framework and the Standard Contractual Clauses. Telegram stores messages on its own servers in several countries and does not offer us a data processing agreement or a transfer mechanism. We send our outreach alerts, which contain the visitor's IP address and its approximate location, the type of device, operating system, and browser, the time zone and the local time there, the browser language, and our automatic guess whether the visit came from a person, an email security scanner, our team, or one of our test links, and the names and email addresses of up to three people we emailed, their replies, which list what the visitor did next on goat.ai and when (the clicks and form submissions, the pages opened, how long the page was visible, and how far it was scrolled) or name one more person we emailed, our messages about clicks on links in our emails, our messages about calls booked, rescheduled, or cancelled on Cal.com (with the name, email address, time zone, and notes of the person who booked), our messages about contact-form submissions (with what was entered in the form, except for a request sent with the reason "Privacy or data request"), and the daily outreach file, there, including those about visits from, and people in, the European Economic Area, the United Kingdom, and Switzerland, without relying on an adequacy decision and without a safeguard under Article 46 of the GDPR or the UK GDPR, or under Swiss data protection law. Apart from the names and email addresses of the people we emailed, an alert contains no email address (a message about a booked call names the person who booked, and a message about a contact-form submission the person who sent it), and no name other than those that Section 2 lists ("Outreach emails and visits that come from them"). Where we emailed only one or a few people at an organization, an alert can relate to one of them even when it names none of them (Section 2). PostHog still receives the page address of every visit, including any short name for your company in it, like other analytics data.

13. Security

We use administrative, technical, and organizational measures designed to protect personal information and raw trace data, including: application-level authenticated encryption of raw trace content at rest (XChaCha20-Poly1305 with per-project key derivation) and TLS in transit; strict database access-segregation and least-privilege roles that prevent the dashboard and analytics systems from reading raw content; hashing of stored credentials; access controls; need-to-know restrictions; and append-only audit logging. Application-level encryption is applied to raw trace content and connected-Source credentials; certain other data (for example, account profile fields, the payout destination, and dataset exports) is stored without additional application-level encryption and is protected by access controls and our subprocessors' at-rest encryption. The redacted trace data in the trace store we run on our compute host is likewise stored without additional application-level encryption and is protected by access controls. Encryption keys are held as environment-level secrets within our deployment environment; we do not currently use a hardware security module or managed key-management service, and key rotation is performed manually.

We do not currently operate independent continuous breach-detection, intrusion-monitoring, security-monitoring, or backup/point-in-time-recovery systems of our own, and we rely in part on our subprocessors' managed infrastructure; these capabilities continue to evolve. No system is perfectly secure, and we do not promise absolute security. Where we confirm a security incident affecting personal information, we will provide notice without undue delay after we become aware, to the extent and in the manner required by applicable law and the DPA. No notice is an admission of liability or fault.

14. Children

The Service is directed to businesses and to individuals of legal age in their jurisdiction, and we do not knowingly collect children's personal information. GOAT does not verify the age of any individual and does not have actual knowledge of the age of data subjects appearing in trace data. Because trace data you connect may contain information about your teammates or your application's end users, you are solely responsible for ensuring you do not connect Sources that submit the personal information of children below the applicable age threshold, and for any affirmative consent required for consumers under 16 under California law. GOAT will not have actual knowledge sufficient to trigger child-specific obligations absent your notice.

15. Changes and Contact

We may update this Policy from time to time. For material changes, we will post the revised version and provide additional notice where required. For privacy questions or rights requests, contact privacy@goat.ai.

GOAT labs

Get paid for your production LLM traces. Research built in the open.

Privacy Policy
Research
StudiesCorpus methodologyResearch accessFAQ
Datasets
GATCGATC-SDataset access
PolymarketBench
OverviewHow it worksAll bets
Contribute
Become a contributorModels & ratesGet started
Company
AboutCareersContact
Legal
Terms of ServicePrivacy PolicyDPA
© 2026 GOAT labs · San Francisco