How to Check Whether ChatGPT Knows Your Brand (and What to Do If It Doesn't)

Three different questions hiding in one

"Does ChatGPT know us?" sounds like one question, but it is really three, and each has a different fix. First: does the model know your brand from its training data, without looking anything up? Second: when an AI assistant searches the web, does it find your site and cite it? Third: when a buyer asks a category question such as "best invoice reminder software for freelancers", does your name come up at all?

A brand can pass one and fail the others. Plenty of small companies are found instantly by search-backed assistants but are completely unknown to a model answering from memory. Knowing which case you are in tells you where to spend your time.

Test 1: memory (two minutes)

Open ChatGPT and turn off web search if your plan shows the option. Ask: "What is [your brand] ([your domain])? Answer from what you already know. If you don't recognise this specific company, say so."

There are three outcomes. The model describes you accurately: you are in its training data. It says it doesn't know you: you aren't, which is normal for young or niche companies. Or it describes a different company with a similar name, or invents plausible-sounding details: the worst case, because buyers who ask will get the wrong answer.

When we ran this on AutoChase, an invoice-chasing tool, in September 2026, the model's answer was simply that it didn't recognise the company.

Test 2: search (two minutes)

Now ask the same question in an assistant that searches the web and shows its sources, such as Perplexity or ChatGPT with search turned on. Look at two things: whether the answer is correct, and whether your own website appears among the sources.

This is where young brands often do much better. The same AutoChase check that drew a blank from memory was answered correctly by Perplexity, citing autochase.app directly. Search-backed assistants read the live web, so a clear, crawlable homepage can be enough to be found — as long as AI crawlers are allowed in (see /blog/robots-txt-front-door).

Test 3: the buyer's question (one minute)

The first two tests ask about you by name. Real buyers don't. They ask the category question, and that is where visibility actually matters. Ask two or three of the questions your customers really type — "best [category] for [audience]", "[competitor] alternatives" — and note who gets named.

Do this more than once. AI answers vary from one run to the next, so a single answer is an anecdote, not a measurement. Our post on measuring GEO work explains why five or six answers per question is the minimum before drawing conclusions (/blog/is-your-geo-working).

Reading your results

If you pass test 1, most of the work is protecting accuracy and winning more category answers. If you fail test 1 but pass test 2, you exist on the web but not in the models' memory — keep feeding the sources models learn from, and be patient, because memory only updates when models are retrained. If you fail both, start with the basics: crawlable pages, clear Organization schema, and independent profiles that describe you the same way (/blog/organization-schema-sameas-for-ai).

Doing it properly, without the guesswork

Running these checks by hand once is useful. Running them consistently — the same questions, several engines, repeated over weeks — is what tells you whether anything is changing. Anovox's AI Awareness audit runs the memory and search tests for you, checks the sources AI learns from (Wikidata, Google's business data, review sites, community discussion, your own schema), and turns every gap into a ranked plan. The free check at /check runs a quick version with no account.