# Is the AI answer about you backed by its sources? Free answer check · EarnCitations

> Free: paste an AI answer and its cited links; each claim is checked against the pages it cites: supported, contradicted, unaddressed or unreadable. No AI calls.

Source: https://earncitations.com/claims.html

> Every page on this site: https://earncitations.com/llms.txt (the whole site as one file: https://earncitations.com/llms-full.txt)

Free tool updated 2026-11-22

## Is what AI says about you backed by its sources?

Paste an answer from ChatGPT, Perplexity, Claude or an AI Overview, with the links it cites. We read each cited page the way an AI crawler does and check the answer claim by claim.

**Send an answer here in one click (bookmarklet)**

Drag this to your bookmarks bar: [Check this AI answer](https://earncitations.com/claims.html). In ChatGPT, Perplexity, Gemini, Claude, Copilot or a Google AI Overview, select the answer (with its links), then click the bookmark. This page opens with the answer, the links inside your selection, and the engine filled in. It doesn't check anything until you press the button.

The answer travels in the part of the address after #, which browsers never send to a server, and this page removes it from the address bar once it's read. It can still be in your browser history, like any page you open. The bookmarklet reads only what you selected, leaves out links back to the assistant itself (your conversation's own address, for example), and opens this page without telling it which site you came from. If clicking it does nothing (Safari blocks bookmarklets on some sites), copy the answer and paste it above.

### The four verdicts

- Supported. A cited page states it: the claim's key words in one passage, and every price, duration and percentage exactly.

- Contradicted. The passage about the same thing gives a different figure. Either the answer or the page is out of date.

- Unaddressed. We didn't find it stated on any cited page. The assistant may have taken it from elsewhere, or from what it already knew. A page can also say it in other words, as a feature list or table does ("Unlimited links & calendars" for "unlimited booking links and calendar syncing"), so open the page before you act on it.

- Unknown. We couldn't read the page (robots.txt asks crawlers like ours not to, it didn't load, it isn't an HTML or text page, or it's an app shell with almost no text without JavaScript), or the page is short and runs scripts and doesn't state the claim in what we can read: the rest may be added by JavaScript.

The same check is earncitations claims answer.txt, the MCP tool check_answer_claims and POST /api/claims ([docs](https://earncitations.com/developers.html#api)). To fix the page that should answer a question, use the [question doctor](https://earncitations.com/question.html); to see what a crawler reads on one page, [what AI sees](https://earncitations.com/view.html).

### Questions

#### What does "supported" mean?

That a cited page states the claim: its key words appear together in one passage, and every price, duration ("14-day") and percentage in the claim appears there exactly. It does not mean the claim is true. A source can repeat an outdated fact, so check the important ones against a current page you trust, ideally your own.

#### And "contradicted"?

The passage on the cited page that best matches the claim gives a different figure of the same kind: the answer says $39 a month, the page says $29. That is the most useful finding: either the answer is wrong, or the page is out of date. Which one it is, you decide; we show both.

#### Why are so many claims "unaddressed"?

Because an assistant can combine what it read with what it already knew, and cite pages for only part of an answer. An unaddressed claim with a figure in it is worth a look: no page the answer cites states that figure. If it's about you and true, publish it on a page you own; if it's wrong, state the right figure plainly there.

#### Does this use AI to judge the answer?

No. It matches words and figures, the same way every time, so it's free and repeatable. That also means it can miss a claim phrased very differently from the source, and it can't judge opinions. Recommendations ("best for small teams") are marked as judgments: a source can back the reasons, not the choice itself. A wrong price inside one ("a good fit for small teams at $29 a month" when the page says $49) is still marked contradicted.

#### Where does the method come from?

Splitting an answer into claims and giving each claim–source pair one of four verdicts (supported, contradicted, unaddressed, unknown) follows the source-tracing method Jam (spreadjam.com) published for AI recommendations. Jam runs it with research agents in a paid plan; this is a free, deterministic version for the checks that don't need judgment.

#### What does "About you" add?

An answer can be wrong about you in three different ways, and each needs a different fix. Entity: does it name you at all? If not, accuracy doesn't matter yet. Find the page that should answer the question with the [question doctor](https://earncitations.com/question.html). Framing: are you the recommendation or a footnote, and is there a caveat attached? Accuracy: are the claims that name you supported by the pages cited? Give your brand and domain and we report the three separately. We list the claims with a caveat word ("but", "however", "lacks"), and you decide whether each is fair. We don't score sentiment. Only claims that name you count, so a follow-on sentence ("Its Business plan costs…") isn't counted as about you.

#### What if the answer is about a different company?

It happens when another company shares your name. Before anything else, the check compares the sites the answer cites, and any domains its claims about you mention, with your domain. A site that has your name but isn't yours (acme.io when you're acme.com) is flagged first in "What to do", because correcting an answer that is really about someone else would be wasted work, and a look-alike can also be a site posing as you. Your own subdomains and the competitors you list are never flagged. Checking that the answer is about the right company before judging it follows Jam (spreadjam.com).

#### Why list competitors?

A discovery question has a few slots, and whoever fills them is who you're up against for that question. Name up to 10 competitors and each gets the same reading as you: named or not, recommended (a "best for…" claim that favours them), their own page cited, and whether they come before you in the answer. In the [answer log](https://earncitations.com/track.html) each is counted per answer, over the answers that asked about it, so adding a competitor later doesn't count them as absent before. When a competitor's own page backs what the answer says about them, that page shows what yours needs to say.

#### What are the correction drafts and the ledger?

Jam's published method for correcting wrong AI descriptions separates the evidence from the request. When a claim that names you is backed by one cited page and contradicted by another, and one of them is yours, we draft a request to the other site's publisher. It quotes both passages and both URLs, and asks only for the facts to match. We can't tell which page is current, so you check that before sending, and nothing is sent for you. The ledger is a CSV with one row per claim: the verdict from each source, the passage behind the best one, and empty columns for your classification (incorrect, outdated, wrong entity, unsupported, opinion or accurate limitation), owner and status. If your page is cited but the answer never names you, we say so: that's source visibility, not a recommendation.

#### Why do two links count as one source?

Because they point at the same page. ChatGPT adds utm_source=chatgpt.com to the links it cites, and other tools add their own tracking parameters, so one page can appear under several URLs. We group links that differ only in tracking parameters, keep each link exactly as the answer gave it, and count how often the answer links the page: linking one page three times is still one source. If a cited URL now redirects somewhere else, or declares a different canonical page, we say so, because the assistant may have read an older page than the one we fetched.

#### What do you fetch, and what do you keep?

Each cited page (eight at most) and its site's robots.txt, with our identified user agent ([EarnCitationsAudit](https://earncitations.com/bot.html)), without running JavaScript. Pages whose robots.txt asks us not to read them are marked unknown, not fetched. The answer you paste is used for this check only: we don't store it or the pages.
