Paste an article or AI answer and we'll break it into claims, flagging which are well-supported, unverified, or disputed — with a credibility score and sources on Pro.
Paste text or load the example to verify its claims.
We split the text into individual factual claims and rate each one — supported, unverified, or disputed.
An overall 0–100 score tells you how trustworthy the passage is at a glance.
Clickable sources with credibility grading and multi-source verification — on Pro.
A passage of text is rarely one big "true" or "false." It's a stack of separate factual claims — a date here, a statistic there, an attributed quote, a cause-and-effect statement — and any one of them can be wrong while the rest read perfectly. This tool's job is to pull that stack apart. Instead of judging the paragraph as a whole, it isolates each checkable statement and asks the same blunt question of every one: can this be stood up, or not?
That matters more now than it used to. Language models write fluently even when they're inventing things, so the usual tell — clumsy prose around a shaky fact — is gone. A fabricated statistic can sit inside a polished, well-cited-sounding sentence and never draw your eye, because nothing looks wrong. A claim-level check is built for that gap: it ignores how confident the writing sounds and looks only at whether the underlying statements hold up.
The first pass is extraction. The tool reads your text and separates the genuinely checkable assertions — "the law passed in 2019," "the company employs 40,000 people" — from the parts that can't be fact-checked at all, like opinions, predictions and rhetorical framing. There's no point grading "this is the most important reform in a generation," so it doesn't try.
Each extracted claim is then weighed on how well it stands on its own. Specific, dated, named claims are easier to verify than vague ones; an assertion that lines up with well-established knowledge is treated differently from one that contradicts it or rests on a number nobody can trace. The tool also watches for the internal tells of a fabrication — a precise-looking figure with no source, a quote attributed to someone who's unlikely to have said it, a detail that quietly conflicts with another claim in the same passage.
On Pro, the check goes further and matches claims against graded web sources, so a flagged statement comes back with somewhere to look rather than just a rating. Either way, the output is a structured map of which sentences carry the risk — so you spend your verification time where it counts.
This isn't a tool you run once for fun — it pays off whenever a wrong fact slipping through would cost you something. A few of the moments people reach for it:
Every claim comes back as Supported, Unverified or Disputed, and the difference between the middle two matters. Supported means the statement is consistent with well-established information. Disputed means it conflicts with the known record or is likely inaccurate — that's your red flag. Unverified is the one people misread: it doesn't mean false, it means the claim is plausible but couldn't be confirmed here. A genuinely true but obscure fact often lands as Unverified simply because there's nothing public to anchor it to.
The overall credibility score is a quick read on the whole passage, not a guarantee. Treat the result as a prioritised to-do list, not a final ruling — it's very good at telling you where to look, and deliberately humble about declaring something settled. For anything high-stakes — a published claim, a legal or medical detail, a number you're about to quote — confirm a Disputed or Unverified flag against a primary source before you act on it. No automated checker, this one included, should be the last word on a fact that matters.
A clean input gets a sharper result. The most common way to get a misleading read is to feed the tool something it was never meant to grade — paste a passage that's all opinion or prediction and almost everything comes back Unverified, not because it's suspect but because there was nothing factual to check. Lead with the sentences that actually make claims.
A couple of other traps worth knowing: stripping a claim out of its surrounding context can flip a fair statement into a misleading one, so keep enough text for the meaning to survive. And don't read Unverified as a synonym for wrong — chasing every Unverified flag wastes the time the tool was meant to save you. Spend it on the Disputed ones first. Used that way, the check does the triage and you do the judgement, which is the right division of labour.
Pair this with the Hallucination Detector to catch fabricated facts.
Hallucination Detector