Fact Checker

Paste a claim or an article, get a sourced verdict: confirmed, likely, unverified, disputed, or false.

The Fact Checker reads what you give it, pulls out the checkable claims, searches for primary sources, and scores every claim on a five-tier scale. The verdict shows its workings: the sources behind each claim, tiered by quality, and the evidence gaps it could not close. Primary sources only; no speculation.

What you might use it for:

• The forwarded article: paste the text and see which of its claims stand up, which are disputed, and which nobody can actually verify.
• A number that sounds too good: check the stat in a pitch deck or a post before you repeat it in a meeting.

Who it is for: anyone who passes information on: readers, writers, analysts, and the family group-chat sceptic.

Where to find it: the sidebar: Fact Checker (/fact-checker).

Watch first: fact-checker.mp4 (captions toggleable in your player, or load fact-checker.srt). The screen is a real recording of the product; the narration is an AI-generated voice.

Check the facts

On STEP 1 · Paste a claim or article:

• Paste the claim or the whole article: up to 2,000 words and 12 analysed claims per run.
• Add files takes documents and images too; documents are extracted and images go through OCR.
• Click Check the facts. The pipeline extracts the checkable claims, searches for primary sources, and scores each claim; it takes about a minute, and the page says so while it works.

The verdict

The Verdict panel returns:

• An overall assessment with a confidence score.
• Every analysed claim with its own five-tier pill: CONFIRMED, LIKELY ACCURATE, UNVERIFIED, DISPUTED, or FALSE OR MISLEADING. Confirmed is earned, not given: it needs multiple independent sources or it is downgraded.
• Source notes, commentary tags, and the evidence gaps it could not close.
• Copy with sources to take the verdict elsewhere.

The inspection

Below the verdict, the Inspection panel shows the full workings: a claim card for every claim with its reasoning, supporting and contradicting sources tiered by quality (with provenance and bias flags), the non-verifiable statements it set aside, and the methodology. This is the panel to read before you argue with anyone.

Go deeper

After the first check, four follow-ups unlock: find primary sources, rate my confidence, flag misleading framing, and write a sourced summary. The Talk To My Agent chat keeps the claim and the report in context for anything else.

Limits and FAQ

Is the verdict guaranteed? No: it is an automated fact-checking aid. It cites what it found and labels its certainty; UNVERIFIED means "no adequate source found", not "false".

How much can I check at once? Up to 2,000 words, 12 analysed claims, and a 50,000-character cap per run; longer texts are truncated.

Where do sources come from? Live web search during the run; each source is tiered by quality and linked so you can open it yourself.

Which model does the checking? Whichever model the lane is configured to use (visible in the top bar).