Being named by ChatGPT is no longer a win. Since GPT-6, it can mean “named and doubted”

Key findings

• 22 times more warnings since GPT-6. Before 10 October, 0.5% of ChatGPT answers said “I couldn’t verify” or “this is not a verified ranking”. After it, 10.1%.
• Only ChatGPT does it. The other four AI apps got the same questions on the same days. They said it in 0 of 7,688 answers.
• The less ChatGPT knows a brand, the more it doubts it. A coffee-discovery app that ChatGPT never names on its own got the warning in 48% of its mentions. A consultancy it names in 7.6% of answers got it in 2%.
• ChatGPT tells the buyer, and the buyer starts to doubt too. The warning is written into the answer the person reads in the chat. If you believe AI knows every company, “even AI can’t vouch for them” sounds like a reason not to buy.
• For a lesser-known brand, it comes when the buyer checks it. ChatGPT rarely names such brands on its own. So the warning shows up in the answer to “is it reliable?” or “X or Y?”, the questions a buyer asks right before deciding.
• Your own numbers are now “company claims”. ChatGPT marks success rates and testimonials from a brand’s own website as unverified.

The three warnings

ChatGPT adds three kinds of warnings about brands. Only the first one existed before October.

1. “This is the company’s own claim.” ChatGPT points out that a number comes from the brand itself. On a Dutch R&D tax-credit consultancy’s success rate, it wrote: “those are company claims, not independently checked figures” (translated from Dutch).
2. “My list is not a verified ranking.” ChatGPT says its own shortlist hasn’t been checked. On cafés: “These are candidates, not verified winners.”
3. “I couldn’t verify this brand.” ChatGPT says it found too little about the company, or can’t tell which company with that name you mean. On a coffee-discovery app: “My verdict: not enough verified evidence to recommend [the app] confidently yet.”

“That doesn’t mean [the product] is unreliable; it means there isn’t enough verified review evidence here to confidently recommend it.”ChatGPT, about a B2B AI app builder, October 2026

A step, not a drift

Warnings 2 and 3 almost didn’t exist until October. Then they appeared within a week.

Not all brands were audited both before and after. So we also compared the same questions for the same brand:

• R&D tax-credit consultancy (Netherlands), same 50 questions: “not a verified ranking” went from 0.8% to 4.9% of answers.
• Solar-panel recycling (Germany), same 50 questions: from 0.4% to 2.1%.
• B2B AI app builder: 0 of 270 answers in September, 41 of 900 on 18 October.

Only ChatGPT does it

In one audit, every AI app gets the same questions on the same day. So the other four show what ChatGPT would look like without the change.

The first warning, “the company’s own claim”, isn’t new. It isn’t only ChatGPT’s either: Perplexity and Copilot also attribute some claims to the provider. And it isn’t aimed at one kind of brand. In ChatGPT answers since 10 October, 75 of 117 of these labels were about competitors, not the brand the question was about.

Who gets the warning: brands ChatGPT barely knows

We counted every mention of a brand in ChatGPT answers since 10 October. Then we checked how often “I couldn’t verify” was in the same sentence or the one before. Next to it is how often ChatGPT names the brand on its own, in answers to questions without the brand’s name. That shows how well ChatGPT knows it.

The brands ChatGPT never names on its own get the warning most. The consultancy it names in 7.6% of answers gets it in 2%. The competitors next to them are mostly well-known names, and they get it in 0.3% of their mentions. So the warning isn’t about who is asking. It is about how well ChatGPT knows the brand. With four brands, this is a pattern, not a law.

ChatGPT doubts out loud, so the buyer doubts too

Search engines judge sources quietly. A buyer never sees why a page ranks lower. ChatGPT now writes its doubt into the answer, in the chat, in plain words: “I couldn’t verify”, “not enough verified evidence to recommend it confidently”.

My view: many people treat AI as if it has read everything and knows every company. So when the AI says it doesn’t know a brand, the buyer doesn’t blame the AI. They start to doubt the brand. The AI’s doubt becomes the buyer’s doubt.

For a lesser-known brand, it also comes at the worst moment. ChatGPT rarely names such a brand on its own, so the brand shows up when the buyer asks about it by name: “is it reliable”, “how much is it”, “X or Y”. That is where the warning lands, right before the buyer decides.

Why ChatGPT changed

OpenAI hasn’t said it added these warnings. Here is what the evidence points to.

1. The timing matches GPT-6. Paid users got GPT-6 on 7 October 2026, Free and Go users from 8 October. It came with “Intelligent UI”: maps, buttons and small tools inside answers. Those show up in the ChatGPT answers we collected that same week. Our last audits before the change ran on 6 and 7 October. The next ones, on 13 and 14 October, show the new warnings. We ask logged out, like a visitor without a paid plan. More on GPT-6 in ChatGPT traffic dropped on 9 October 2026.
2. ChatGPT now separates what it checked from what it repeats. The new warnings are about its own confidence: “my list”, “I couldn’t verify”. They are not about other people’s rankings.
3. It calls out self-promotion in front of the buyer. Search engines push self-published “best X” lists down quietly. ChatGPT says it on the screen. It flagged company claims before GPT-6 too.
4. It protects OpenAI. If ChatGPT recommends a weak product, it has already said “I couldn’t verify it”.
5. The others don’t say it at all. Gemini, Google AI Mode and Copilot add no such warning. Perplexity shows its sources inside the answer, which does part of the same job.

What it means for brands

Being named is no longer a win. Now “named” can mean “named and doubted”, in an app with 900 million weekly users. The words ChatGPT puts next to your name are part of your brand now. If ChatGPT barely knows you, it says so to the buyer, and the buyer’s doubt follows. “I couldn’t verify” is the phrase that makes a buyer pick the name it does vouch for.

That is why AEO has to be done as a whole, not as a reaction to each update. Knowing that GPT-6 shipped won’t change what it says about you. The evidence it reads will:

1. Make your company easy to verify. Use one name, written the same way everywhere, and the same facts on every page and profile. Link your profiles from your site. If another company or a common phrase shares your name, say clearly who you are and where you are.
2. Put proof outside your own website. ChatGPT now calls your success rate and your testimonials “company claims”. Reviews on independent sites, registry entries, certificates, press and case studies on other sites are things it can check.
3. Ask ChatGPT about yourself by name. A buyer’s last question is often “is [you] reliable?”. Ask it that way, several times, and note what it says next to your name.
4. Track the verdict, not just the mention. Does it recommend you, recommend you with conditions, pick a rival, or mix you up with someone else? More in what AI visibility tracking covers and brand entity optimization for AI.

Methodology

• Data: 28,145 answers from Far & Wide audits, 2 September to 19 October 2026. 30 audits of 9 brands. ChatGPT 10,504 answers, Gemini 10,209, Microsoft Copilot 3,915, Google AI Mode 1,994, Perplexity 1,523.
• How we asked: in the real web apps, logged out, with web search on, from the buyer’s country and in their language (English, Dutch, German or Russian). Each question was asked several times.
• Brands, by industry: solar-panel recycling (Germany), savings and cashback app (Netherlands), R&D tax-credit consultancy (Netherlands), coffee-discovery app (UK), B2B AI app builder, browser games, grocery retail (Netherlands), car repair (Moldova), group gifting (India). Our internal test sites are left out.
• Before and after: answers before 10 October 2026 against answers from 10 October. In each audit, all AI apps get the same questions on the same day. For two brands we also compared the same 50 questions.
• Finding the warnings: lists of phrases in four languages for each kind of warning, for example “I couldn’t verify”, “not a verified ranking”, “geen onafhankelijk geverifieerde”, “nicht unabhängig geprüft”, “bedrijfsclaim”, “nach eigenen Angaben”. Each answer counts once.
• Who a warning is about: the brand named in the same sentence or the one before. Competitors are the other brands named in that answer.
• How well ChatGPT knows a brand: the share of ChatGPT answers that name the brand when the question doesn’t.
• General vs sample findings: we state as general only what holds across all answers, including answers about well-known brands. Findings that depend on how well ChatGPT knows a brand are marked as such.
• Collection check: our collection method didn’t change between 6 and 14 October. The answers from 18 and 19 October, which still show the warnings, all came through our main collector.