Guide
Why Won't ChatGPT Say My Business Is Legit?
A shopper types “is Bright Path Dental legit” into ChatGPT before booking, and the answer comes back soft: “I can’t independently verify this,” followed by a suggestion to check reviews. A ChatGPT hedge on a legitimacy question is the model declining to vouch for a business it cannot corroborate against evidence outside that business’s own website. The gap sits in missing evidence, not abstract trust: no verifiable address, no stated refund policy, no proof written by anyone other than the business itself. Publish that evidence and the hedge usually closes.
Why does ChatGPT hedge instead of vouching for a real business?
ChatGPT is not weighing the business’s character. It is weighing evidence. Ask it to confirm a fact about a company and the model treats that as a claim needing support, not an opinion it can offer freely. If the only source is the business’s own homepage, that is one witness testifying about itself. A model built to avoid overclaiming will not stake a recommendation on a single self-interested source.
The check we run for a business puts real buyer-intent questions, including “is this legit” style prompts, to the main assistants and records whether and how the business gets mentioned in the answer. Three separate model providers check the same company against whatever else exists about it beyond its own copy, then answer only as far as that evidence allows.
For the separate case where a company’s published facts are simply wrong rather than absent, see Why does ChatGPT describe my company incorrectly? A legitimacy hedge and a factual error are different faults with different fixes.
How many people are actually asking AI if a business is legit?
This is not a fringe question. OpenAI’s threat intelligence team, in a piece called “Scam-Spotting and Stopping” published on OpenAI Global Affairs in September 2025, put a figure on it: threat investigator Jack Stubbs says legitimacy and scam checks drive around 15 million ChatGPT messages a month, globally. Stubbs also estimates that for every attempt to misuse ChatGPT for a scam, three people use it to protect themselves from one.
Fifteen million monthly messages is not a curiosity. It is a standing queue of buyers typing a business name into ChatGPT moments before they hand over a card number. A hedge in that moment does not just fail to help. It sits directly between the shopper and the sale.
Which trust signals actually close the gap?
Three signals do most of the work, and each is checkable, not aspirational.
| Signal | What it proves to a model | Where it needs to live |
|---|---|---|
| Verifiable address | The business is a real, locatable entity, not just a domain | Google Business Profile, a Companies House filing, a footer with a street address |
| Refund or returns policy | Buyers have recourse if the purchase goes wrong | A standalone, linkable policy page, not a line inside the general terms |
| Third-party proof | Someone other than the business vouches for it | Trustpilot, an industry directory, a press mention with a byline |
None of these is exotic. A ten-year-old local shop usually has all three scattered across the web already; the problem is that they are scattered, not written down anywhere a model can find in one pass. Consolidating them, in plain extractable text, is most of the fix.
What does measuring it actually change?
A first pass shows whether the hedge is happening at all: how often the business is mentioned when its buyers ask, and what the assistants actually say about it. The work that follows lists the buyer questions where the business never appears, names the pages cited instead, and produces the fixes themselves: an llms.txt, structured data, rewritten homepage copy and an FAQ written from the questions that were being lost. The output does more than diagnose the problem. It puts the missing evidence on the record, formatted the way a model reads it.
Common questions
Does rephrasing the question to ChatGPT fix the hedge? No. Asking more directly does not add evidence. The model still has nothing new to check the claim against, so a firmer prompt on the same underlying facts usually returns the same caveat, sometimes with different wording.
Does one Trustpilot page solve it on its own? Rarely by itself. One review platform is one third-party source, and OpenAI’s own scam-checking guidance points shoppers toward weighing several independent signals, not one. A review page combined with a stated refund policy and a verifiable address moves the needle further than any one of the three alone.
How do I find out which specific signals are missing for my own site? The check names exactly which buyer questions a business is missing from and which of these trust signals are absent, so the fixes are written from the gaps rather than from a template.
A hedge is not a verdict. It is a model telling you, plainly, what it still needs to see.
Sources
- OpenAI Global Affairs, “Scam-Spotting and Stopping,” 2025
Courage Horizon sets up AI agents for UK small businesses: software that does one job, such as chasing unpaid invoices, and asks you before it sends anything. Being found when customers ask ChatGPT is part of that work. Get a free report on your own business.