New research across 7 AI platforms and 55,000 tests found that being mentioned by AI is not the same as being believed. Only 0.6% of businesses show real proof.
BLUF: Getting cited by AI and getting trusted by AI are two different things. New research across seven AI platforms and more than 55,000 tests found that a mention does not equal belief, and the gap comes down to one thing: whether a machine can check what you say about yourself. Facts get believed. Adjectives get skipped.
Plenty of owners have had this moment. You type your business into ChatGPT, and there you are. Named. Spelled right. Sitting in the answer next to two competitors.
Then nothing happens.
It feels like a win, and for about a day you treat it like one. But a mention is only the first half of the transaction. The buyer reads the whole answer, and somewhere in that reading they decide which of the three names sounds like a real business with real proof behind it, and which two sound like copy someone wrote about themselves.
That decision is the one that pays. And until recently nobody had measured it.
In June 2026, Burson, one of the largest communications firms in the world, published a study called The Credibility Paradox with the AI search company Profound. They ran 85 companies across 10 industries through 7 answer platforms and produced more than 55,000 believability forecasts.
The headline finding is short enough to fit on a bumper sticker. Being cited by an AI engine does not mean being believed by it.
Three things fell out of the data:
Here is the part that should stop you. Nearly every tool sold in this market right now counts how often you get mentioned. The research says the count is necessary and nowhere near sufficient.
AI Trust Signals has scored more than 3,000 businesses on the signals AI checks before it recommends anyone. One of those signals is simple: can a machine confirm that real, credentialed humans run this company?
Out of 3,245 businesses in the AI Trust Signals database, only 21 score High on that signal. Twenty-one. That is 0.6% of the market.
The more interesting number is the middle. 72.2% score Medium. We call it the Medium Trap, and it is exactly what it sounds like. These companies have a team page. They have headshots. They have first names and job titles. What they do not have is the depth a machine needs to verify a single claim: full bios, credentials that can be checked, links to real profiles elsewhere on the web, and named authors on the content they publish.
Medium feels like done. To an AI engine, Medium reads as unverified.
That is the Credibility Paradox showing up in a scan of real businesses. Almost everybody has something. Almost nobody has proof.
You do not need a content department. You need to convert the claims you already make into claims a machine can check.
Step one: name the humans. Put full names, real bios, and credentials on your team page. Link each person to a profile that exists somewhere else on the internet. This moves you out of the 72.2% faster than anything else on the list.
Step two: replace adjectives with facts. Go through your homepage and your top two service pages. Every time you find a word like best, trusted, or leading, either delete it or swap it for something with a number, a date, or a licence attached.
Step three: get corroborated somewhere you don't control. Reviews on more than one platform. A quote in a local publication. A trade association listing. The proof that counts most is the proof you cannot edit.
None of this is a marketing campaign. It is a Tuesday afternoon, and most owners can do the first step themselves.
We see the same pattern almost every week. A contractor with 200 five-star reviews, twenty years in business, and a team page listing "Mike," "Dave," and "the crew." Great company. Genuinely great. And an AI engine reading that page cannot confirm that a single qualified person works there.
Meanwhile a smaller competitor down the road has four named people, each with a bio, a licence number, and a profile that matches. Fewer reviews. Less history. More proof.
When a buyer asks AI who to call, the smaller shop keeps getting recommended, and the bigger one keeps wondering why the mentions never turn into calls.
The market spent the last two years learning to ask whether AI can see them. That was the right question in 2024. The better question now is whether AI has any reason to believe what it sees, because a mention with nothing behind it loses to a mention with proof behind it every single time.
People still buy from businesses they trust. They ask AI first now, so the trust has to be legible to a machine before it ever reaches a human.
Find out what AI actually says about your business. Run the free check at aitrustsignals.com/recommends. No signup, and it takes a couple of minutes.
A citation means an AI engine named your business in an answer. Belief means the engine treats your claims as credible enough to recommend you. Research from Burson and Profound found the two do not move together. You can be mentioned often and still be passed over for a competitor with more verifiable proof.
Factual, checkable claims perform best: products, licences, named people with credentials, published policies, and specific commitments like a response time. Subjective claims about being the best or most trusted score poorly. If a person could not confirm the statement in about a minute, an AI engine will discount it too.
Anything you publish about yourself is a claim. Anything published about you elsewhere is corroboration. AI engines weigh independent sources like reviews, press coverage, and public conversation more heavily because those sources have no reason to flatter you. A business with proof in both places is far harder to skip.
If your team page has photos and first names but no full bios, credentials, or links to outside profiles, you are in it. So is roughly 72% of the market. The fix is to add depth to what already exists, not to build something new from scratch.
See what AI says about your business.
