Someone weighing your company no longer only searches for it. Increasingly, they ask about it. Is this company legitimate? What does it actually do? How does it compare to its competitors? They type the question into ChatGPT or another assistant and read a fluent, confident paragraph in reply. Pew Research found that about 44% of U.S. adults had used ChatGPT by the middle of 2026, and Semrush's study of the buyer journey put 64% of AI users on it monthly to research products and brands. Brand mentions in ChatGPT are now a reputation surface, and it is one most companies have never read.
What makes this surface different from search is that it does not show its work. A results page lays out its sources and lets the reader judge. An assistant delivers a conclusion. If that conclusion is wrong, outdated, or borrowed from a critic, the person reading it usually has no way to tell and no reason to doubt it. The answer sounds authoritative because assistants always sound authoritative.
First, the Objection Worth Answering
A fair objection comes up here. Google's AI Overviews, the generated summaries at the top of search, barely appear for a bare brand name. Ahrefs' analysis of 146 million results found AI Overviews on just 0.09% of navigational searches. If AI almost never summarizes your name, why worry about AI and your brand?
But it is important to look one line down in the same data. Branded queries overall trigger an AI Overview 13.1% of the time, more than a hundred times as often as the bare name. The difference is the modifier. Searching the name alone tells Google you want the company, so it points you at it. Add a trailing modifier and the intent flips: 'brand reviews,' 'brand versus a competitor,' 'is brand legit,' 'brand pricing' each read as a request for explanation, and Google, assuming more information is wanted, generates it. The same names that are almost never summarized get summarized constantly the moment a question is attached to them.
And that is still only Google. AI Overviews and chat assistants are different surfaces. An AI Overview is something Google decides to show. A chat assistant is something a person chooses to open and ask, and it answers about your brand every time. The bare-name search still returns your classic brand SERP; the modified query increasingly returns an AI summary; the question typed into ChatGPT returns whatever the model has assembled about you. The reassuring statistic covers one corner of one surface, not the whole picture.
TWO SURFACES, NOT ONE
An AI Overview is something Google decides to show for a query. A chat assistant is something a person opens and asks. The first rarely triggers on your name alone but shows up once a modifier is added; the second answers about you on demand. Watch them separately.
Where the Answer Comes From
When an assistant answers a question about a company, it is not reading a live, complete file. It synthesizes from whatever public material it was trained on or can retrieve: reference sites like Wikipedia, community discussion on Reddit and other forums, review platforms, mainstream editorial coverage, and, somewhere in the mix, the company's own site. The hierarchy of source types is fairly stable. The exact weighting is not.
Treat any specific figure as a snapshot with a short shelf life. In September 2025, Semrush's tracking recorded Reddit's share of ChatGPT citations falling from roughly 60% to roughly 10% within about two weeks, after a change in how Google exposed search results. Early-2026 trackers still place Wikipedia and Reddit among the largest single sources ChatGPT draws on, but the precise percentages differ from study to study and month to month. What holds is the shape: a narrow set of reference, community, review, and editorial sources does most of the talking about your brand, and it is not the same set that ranks for you on Google. Ahrefs has found that only a small fraction of the URLs assistants cite overlap with Google's top ten results, which means your search rankings do not automatically make your AI answer accurate.
Why the Answer Is So Often Wrong
An assistant produces a fluent answer whether or not it has good material to build one from. When a company is thinly or inconsistently documented in public, when it exists mostly on its own website and little else, the model fills the gaps with inference, with a competitor's framing, or with a fact it absorbed years ago and never updated. The result is the failure mode specific to this surface: confident, well-written, and wrong. A founder who left in 2022 is still named as the CEO. A product the company never shipped appears among its offerings. A single old controversy becomes the defining fact. None of it is malicious. All of it is repeated to every person who asks.
What Actually Shapes It
You cannot edit the answer. You can change what the answer is built from. Because assistants assemble their picture of a company from the public record, improving that picture is the same footprint work that governs the rest of your brand reputation: an accurate Wikipedia entry where one is warranted, consistent profiles across the platforms assistants read, a genuine review presence, and a site that states the basic facts plainly. How well Google and other systems understand your company as an entity is the foundation under all of it, and a brand present only on its own domain is the one most easily gotten wrong. This is the same public footprint that shapes your brand SERP, and that brand SEO exists to manage. The AI answer is one more thing to read.
Two things sit next to this and are deliberately left to their own homes. Systematically monitoring and optimizing how your brand appears across AI engines is a distinct discipline, with its own tooling, covered in the GEO and AEO pillar. And when an assistant repeats something false or hostile that traces to a specific source, correcting it means addressing that source, which is the work of burying negative search results. Brand mentions in ChatGPT are the symptom you can read; those are two of the levers you actually pull.
How to Read Your Own AI Answers
Checking is simple and worth doing on a schedule. Open the major assistants, ChatGPT first, then Gemini and whichever others your audience uses, and ask the plain questions a customer would: what the company is, whether it is legitimate, who is behind it, and how it compares to a named competitor. Note three things: what is wrong or out of date, what is missing that should be there, and which sources the answer leans on when it shows them. Then run it again in a month, because the answer will have moved. This is a periodic reputation check, not a one-time audit, and its systematic version belongs to the monitoring discipline above.
The Bottom Line
For most of the web's history, the question was what appeared when someone searched for your name. Now there is a second question: what an assistant says when someone asks. You do not control that answer and cannot correct it directly, but you are not powerless over it either. It is built from a public record you can improve, and it shifts as that record shifts. The companies' AI will describe accurately are the ones already doing the unglamorous work of being clearly, consistently, and verifiably themselves everywhere the machine is reading.