The Difference Between Facts and Claims
If you have ever wondered why ChatGPT, Claude, or Gemini repeats some of what your website says and quietly drops the rest, here is the direct answer. AI sorts every sentence it reads about your business into one of two piles. Statements it can check against other sources go into one pile. Statements only you could have made go into the other. The first pile becomes what AI says about you. The second pile mostly disappears.
That is the difference between a fact and a claim, and it is not the difference most people assume. It has nothing to do with whether the statement is true. A fact, in the way AI handles information, is a statement about a property of your business that other sources could confirm or contradict: what you do, where you are, who you serve, how long you have operated. A claim is a statement that cannot be checked from the outside because no source other than you has any standing to make it: that you are the best, the most trusted, the highest quality, the leader. Facts can be wrong. Claims cannot even be wrong, because there is nothing to test them against. And a system that decides what to believe by comparing sources has no use for a statement that only one source can make.
Understanding that sorting explains a great deal about why AI descriptions of businesses look the way they do. It also explains why a website can be honest, well written, and accurate, and still produce an answer that says almost nothing.
Why Truth Is Not the Dividing Line
The instinct is to think of facts as true things and claims as exaggerations. That is a reasonable definition for a courtroom. It is the wrong one for AI, because AI has no way to determine truth. There is no verification step inside these systems, no moment where a statement is held up against reality before it reaches a prospect. What the system does instead is compare, and the mechanics of that comparison are covered in What Makes AI Confident in an Answer? When many sources describe your business the same way, the description hardens into something AI states plainly. When sources disagree, it hedges or picks a side. When only one source says something, that something sits in limbo.
Look at what that means for a sentence like “we were founded in 2004 and serve commercial clients in Maricopa County.” Every part of it can be checked. A state filing, a directory, an old news mention, a review that says “they did our office building” all either agree with it or do not. The sentence is a fact in the sense that matters: it makes a testable assertion, and the web can vote on it.
Now look at “we are the most trusted commercial contractor in the Valley.” Nothing on the web can vote. No directory records trust. No review can confirm a superlative. The statement is not false. It is simply unverifiable by design, and unverifiable statements have no weight in a system built on agreement between sources.
There is a hard consequence hiding in this. A wrong fact that many sources repeat outranks a true claim that only you make. If a discontinued service still appears on six old listings, that is six sources agreeing, and AI may trust old information about your business precisely because it looks corroborated. Truth was never the test. Checkability was.
Everything on Your Own Website Starts as a Claim
Here is the part that catches most owners. From where AI sits, a statement is not a fact because of what it says. It is a fact because of who else says it. Which means that on your own website, every sentence starts life as a claim, including the ones that read like plain fact.
“We specialize in restaurant bookkeeping” is a property statement, written in the right register, with your business as the subject. It is also a statement made by the party with the strongest motive to make it. AI reads your site, but as AI Reads My Website, So It Knows What My Business Does. Not Necessarily. explains, it weighs your site as one voice among many. The specialty becomes a fact when your directory listings say restaurants, when reviews mention restaurants, when a local article describes the restaurant work. Until then it is a claim that happens to be phrased like a fact, and it carries the confidence of a single source.
This is why schema stops at the edge of your website. Organization schema lets you state your properties in the clearest possible format, and that has real value, but it is still you stating them. Structured data cannot manufacture a second source. As What Schema Actually Does puts it, markup labels what your site already says. It does not turn a self reported property into a corroborated one.
Recent guidance from the AI visibility field lands on the same point from the other direction. Analysts writing about law firm visibility in 2026 note that AI assistants weigh third party recognitions and consistent directory data far more heavily than a firm’s own site, because those sources can be verified and self published content cannot. The principle is not specific to law. It is how source comparison works.
What Happens to Each Pile
Once you see the two piles, AI answers stop being mysterious.
Corroborated facts come out clean and specific. AI states your services, your location, your founding date, and your specialty in the confident tone it uses for anything the sources agree on. This is the version of your business that shows up when things are working.
Uncorroborated facts come out hedged. Your site says you handle commercial work, nothing else does, and the answer says you “appear to focus on” or “may also offer” commercial work. What AI Uncertainty Looks Like When It Describes a Business catalogs those tells. A prospect reads the hedge as doubt, and doubt at the research stage functions as a no.
Claims come out as nothing. They are not contradicted or softened. They are simply absent, because there was never anything in them a comparing system could use. The superlatives that fill so many homepages leave no trace in the answer, a pattern I described from a different angle in Why AI Can Misread Marketing Language. That post is about register. This one is about verifiability. A sentence can be written in flat, unpromotional language and still be a claim if no outside source could ever confirm it.
And the empty space left by discarded claims does not stay empty. As explained in What Missing Information Can Reveal About AI’s Understanding of a Business, when a model lacks stated properties it fills the gap with the defaults for your category. A page that is mostly claims does not produce a bad description. It produces a generic one, indistinguishable from the average business in your industry.
Three Questions That Sort Your Own Sentences
You can run this test on your own site in an afternoon. Take any sentence about your business and ask three things.
Could a stranger confirm this from a source other than you? If yes, it is a candidate fact. If the only possible source is your own opinion of yourself, it is a claim.
Could this sentence be wrong? A fact can be. “We have four locations” is wrong if you have three. “We deliver exceptional service” cannot be wrong in any checkable way, which is exactly why it carries no weight.
Who is the subject? Facts about your business have your business as the subject and a property as the predicate. Sentences about how the customer will feel, or about the industry in general, are not facts about you no matter how true they are.
Run enough sentences through that filter and a pattern usually emerges. The pages a business is proudest of tend to be the ones with the highest ratio of claims to facts, because those are the pages written to persuade. The facts are often buried in a footer, an old about page, or a directory profile someone filled out years ago. That is the material AI is actually working from.
Why This Costs Money Without Showing Up Anywhere
The moment this matters is the one you never witness. Your prospect may ask AI about you before contacting you, and what they receive is assembled entirely from the fact pile. If that pile is thin, the answer is thin. If it contains old facts that many sources still repeat, the answer is confidently outdated. If it contains nothing but the category, the prospect hears a description that fits you and forty competitors equally, and picks the one AI could describe with more specificity.
None of that generates a signal. The prospect does not report that the answer was vague. Your analytics never see them. The claims you invested in writing did their job on the human readers who reached your site, and did nothing for the readers who were filtered out before they got there.
Finding Out Which Pile You Are In
You cannot tell from rereading your website which of your statements AI treated as facts and which it discarded, because you already know all of them are true. The sorting happened somewhere else, across sources you do not control, and the only visible evidence is in the answers themselves.
That is what the AI Business Understanding Report documents. I question ChatGPT, Claude, and Gemini about your business from multiple angles and record what each model states with confidence, what it hedges, and what it leaves out entirely. Where the three models agree, that usually marks a fact the wider web has confirmed, which is why agreement between ChatGPT, Claude, and Gemini is worth reading closely. Where they hedge or go silent, that usually marks a property your business believes about itself that nobody else has yet confirmed. The report shows you that line, drawn by the systems your prospects are actually asking.
Your website is full of statements. Some of them are facts to AI and some are claims, and the difference was decided without consulting you. Ordering a report is how you find out where that line currently falls, and what it is costing you on the wrong side of it.