Paste a passage and this free AI fact-checker breaks it into separate factual claims and searches the web for a source for each one. Three checks a day, up to 200 words, ten-word minimum, no account and no card.
The claim-splitting is the useful part and it is worth seeing before you decide whether to use it. We gave it five sentences about the Eiffel Tower and it came back with seven claims, each with a source attached. A paragraph usually contains more checkable assertions than you would count by reading it.
Run against our live tool on 4 October 2026, so you can see the shape of the result.
We pasted a short passage about the Eiffel Tower containing a completion date, a height, a designer, an intention to dismantle it, a story about a con artist selling it for scrap, and a very specific parts count.
The response came back with seven separate claims, an assessment for each, at least one source attached to each, a count of how many web searches were run, an overall score for the passage, and a disclaimer about the limits of the method.
Two things to notice. Seven claims from five sentences, because a single sentence frequently contains two assertions and the split is per claim rather than per sentence. And sources on the free tier: the free result carries at least one source per claim, which is the part of this category most often held back behind payment. More sources per claim is what the paid tiers add.
A verdict on a whole paragraph is nearly useless, because paragraphs are rarely uniformly right or wrong. The common real case is four accurate sentences and one confidently wrong one, and a paragraph-level score averages that into "mostly fine", which is the least actionable possible answer. Splitting is what makes the output something you can act on line by line.
Both matter more than accuracy rates, which is why they go before any numbers.
It cannot settle a contested question. Where sources genuinely disagree, finding a source is not the same as establishing a fact. An automated check reports what it found, and what it found depends on what ranked that day. For anything politically or scientifically contested, treat the result as a reading list rather than a ruling.
It cannot tell you what was left out. This is the bigger one and almost nothing in the category addresses it. Every claim in a passage can be individually true while the passage as a whole misleads, through selection, framing or omission. A fact-checker reads the claims that are present. The most effective misinformation is made of true statements, and this method is structurally blind to it.
A source attached to a claim means a search returned something that appears to support it. It does not mean the source is authoritative, current, or that the claim is correctly summarised from it. Open the sources on anything that matters. Our own response carries a disclaimer to that effect rather than presenting the result as settled, which we would rather you saw here than discovered afterwards.
Three good uses and one bad one.
Checking AI output before you publish it. The strongest use. Generated text states things with uniform confidence whether or not they are true, and a claim-level pass finds the specific sentence that is wrong rather than leaving you to re-read for a feeling of doubt.
Finding the checkable claims in your own draft. Often the result is simply a list of the assertions you have made, which is a useful thing to look at. Any claim with no source found is a claim you need to support or soften.
A first pass before real verification. It orders the work. It does not do the work.
Not for settling an argument. A screenshot of a tool output is not a source, and if you use one that way the first thing a reasonable person does is ask what the tool checked.
If what you are checking is an AI-generated reference list rather than prose claims, the hallucination checker is the better fit: it resolves DOIs against CrossRef, arXiv IDs against arXiv, PubMed IDs against PubMed and ISBNs against Open Library, so a fabricated citation comes back as a lookup rather than a judgement. That procedure is here. Fact-checking is for claims; citation checking is for identifiers, and they fail in different ways.
The daily count gates text scans and rewrites; fact-checking has its own count.
| Free | Starter | Pro | |
|---|---|---|---|
| Price | $0 | $9.99/mo · $7.49 yearly | $19.99/mo · $14.99 yearly |
| Actions a day (shared) | 3 | 20 | Unlimited |
| Words per scan | 1,500 | 5,000 | 7,500 |
| Words per rewrite | 260 | 5,000 | 10,000 |
| Detection a month | No cap | No cap | No cap |
| Humanizer a month | No cap | 40,000 words | 80,000 words |
One action is one scan, one file upload or one rewrite, from a single shared daily count. That daily number is the gate that matters: there is no monthly ceiling on any plan, so what separates the tiers is how often you can work and how long a single piece can be, not how much you get through in a month.
Read the free row together rather than line by line. No monthly cap sounds unlimited and is not: three actions a day at up to 260 words a rewrite is roughly 780 rewritten words in a day. It is a real allowance for checking and fixing specific passages, and it is not a plan for processing a backlog.
Current prices, the Business and Enterprise tiers and regional pricing are on the plans page, which is the page we keep up to date.
Fact-checking has its own count. 3 checks a day with no account and 5 on a free account, at up to 200 words a check with a ten-word minimum; Starter raises the length to 1,000 words. The free tier returns at least one source per claim, and more sources per claim is what the paid tiers add. It does not spend from the shared daily actions above.
Yes: three checks a day at up to 200 words, with a ten-word minimum, no account and no card. Each check splits the passage into separate factual claims, attaches an assessment and at least one source to each, and returns an overall score for the passage. Sources on the free tier are worth noting, because that is the part of this category most often held back behind payment.
More than you would count by reading. We gave it five sentences about the Eiffel Tower and it returned seven claims, because a single sentence often carries two assertions and the split is per claim rather than per sentence. That granularity is the point: a verdict on a whole paragraph averages four accurate sentences and one wrong one into "mostly fine", which is the least useful answer available.
Yes, at least one per claim on the free tier, and more sources per claim is what the paid tiers add. A source means a search returned something that appears to support the claim. It does not mean the source is authoritative, current, or that the claim was correctly summarised from it, so open the sources on anything that matters. The response carries a disclaimer to that effect rather than presenting the result as settled.
Not in the sense of settling it. Where sources genuinely disagree, finding a source is not establishing a fact, and what an automated check finds depends on what ranked that day. For anything politically or scientifically contested, treat the output as a reading list rather than a ruling, and expect to do the reading.
Omission. Every claim in a passage can be individually true while the passage as a whole misleads, through selection or framing or by leaving something out. A fact-checker reads the claims that are present and has no way to see the ones that are absent. The most effective misinformation is usually built from true statements, and this method is structurally blind to that.
Depends what you are checking. For prose claims such as statistics, dates and events, use the fact-checker, which searches for a source per claim. For an AI-generated reference list, use the hallucination checker, which resolves DOIs against CrossRef, arXiv IDs against arXiv, PubMed IDs against PubMed and ISBNs against Open Library, so a fabricated citation comes back as a database lookup rather than a judgement. Claims and identifiers fail in different ways.
Three checks a day, 200 words, no account.
Open the tool →For reference lists, where identifiers resolve against real databases.
See the procedure →Six ways to catch a fabrication, ranked by what is verified.
See the comparison →Twenty-five tools, and what each one actually does.
Browse the tools →Three checks a day, no account. You will usually find more checkable assertions than you expected.