Brand FactCheck
Verify what AI answers claim about your brand against your own approved facts.
AI answers do more than mention your brand: they make statements about it. Prices, founding dates, policies, product capabilities. Some of those statements are wrong. FactCheck extracts the factual claims about your brand from every tracked answer, verifies each one against a fact base you control, and scores how accurately the engines describe you, measurement window by measurement window.
The fact base
Verification is only as good as the facts behind it, so FactCheck starts with a knowledge base you curate. There are three ways to fill it:
- Add facts manually. A fact you type in is live immediately.
- Guided crawl. FactCheck suggests the pages on your site worth extracting from (pricing, about, support and similar), crawls the ones you pick, and proposes one atomic fact per statement it finds.
- Paste a document. Paste the text of a brochure, FAQ, or policy document and proposals are extracted the same way.
Crawled and pasted facts always arrive as proposals: nothing is verified against them until you approve each one. Every fact carries a category (product, pricing, company, people, policy, or other) and an optional effective date for statements that change over time.
Facts age. An approved fact left untouched for 180 days gets a review chip, so an old price or a retired policy does not silently keep passing verification. A coverage meter shows which categories your approved facts cover, so you can see the gaps before you read the scores.
How claims are verified
Each measurement window, FactCheck reads your tracked answers, extracts the checkable statements about your brand, and compares each one with your approved facts:
- Accurate: the claim matches an approved fact.
- Inaccurate: the claim contradicts an approved fact. Each inaccurate claim is also classified as a hallucination (invented outright), outdated (true once, not anymore), or incorrect (wrong about a current fact), so you can tell fabrication from stale information.
- Unverifiable: no approved fact covers the claim, so it cannot be judged either way.
Accuracy Score
The Accuracy Score for a window and engine is the share of decided claims that were accurate: accurate divided by accurate plus inaccurate. Unverifiable claims never count against you; they are reported separately, as a signal that your fact base has a gap.
Always read the score against knowledge-base coverage. A high score over a thin fact base means little was checked, not that everything is right. Expanding the fact base turns unverifiable claims into decided ones and makes the score both stricter and more trustworthy.
Review queue and feedback
Verdicts the verifier is not confident about wait in a review queue instead of moving the score; confirm or correct them and they count from then on. On any claim you can also push back:
- Not an error suppresses the claim: it leaves the score, and the same claim will not be flagged again in later windows.
- Wrong verdict replaces the verdict with yours, and the score recomputes.
Themes and sources
Recurring claims are clustered into themes, so twelve variations of the same wrong price read as one problem instead of twelve rows. Themes are ranked by how often they appear and how often they are wrong, which makes the worst recurring inaccuracy the first thing you see.
Each claim is also attributed to the sources the answer cited, split between your own pages and third-party sites. That shows you which external pages keep feeding an inaccuracy back into the engines.
Fixing inaccuracies
When inaccurate claims trace back to your own pages, FactCheck feeds the recommendations engine: you get a concrete recommendation to fix the page so it states the facts, and coverage gaps turn into suggestions for content worth publishing. For third-party sites, generate an outreach pack: a neutral correction request listing what the site gets wrong, the pages involved, and your correct facts with sources, ready to paste into an email to the site owner.
Availability
FactCheck is available from the Pro plan up. Pro verifies up to 500 claims per measurement window and Scale up to 2,000; the claim pack add-on adds 500 more to either, and Enterprise plans have no cap. When a window reaches its allowance, verification pauses and picks up again with the next window.