How to Find Out What ChatGPT Says About Your Brand (Free Method)

You can audit what ChatGPT says about your brand today, for free, in an afternoon

More of your buyers are asking an assistant before they ask Google. They type "best <your category> for <their situation>" into ChatGPT, read the three or four brands it names, and quietly rule out everyone it didn't. If your brand isn't in that answer, you never find out. There's no impression count, no ranking report, no lost-click line in a dashboard. You just don't get considered, and you have no idea it happened.

McKinsey projects roughly $750 billion in US revenue will flow through AI-powered search by 2028, while only about 16% of brands systematically track how they appear in AI answers. That gap is the opportunity: most of your competitors have never checked. You can, and you don't need a tool or a budget to start. Below is the exact method we use when we open an AI search visibility audit — the same steps, done by hand, that you can run this afternoon for nothing. At the end I'll be honest about where the free version stops being enough.

Step 1: Write down the real questions your customers ask

Not your keywords. Your customers' questions, in their words. The assistants answer questions, so that's what you have to feed them.

Sit down and write 15 to 30 real buying questions. Three shapes cover most of them:

• Category-plus-situation. "Best <category> for <use case>." Think "best chargeback app for a Shopify store on Stripe" or "best running shoes for flat feet." This is where a stranger with a problem starts, and it's the highest-value slot to appear in.
• Competitor alternative. "<Competitor> alternative," "something like <competitor> but cheaper," "<competitor> vs <competitor>." People shopping around ask these constantly, and the assistant's answer decides who gets added to the shortlist.
• Direct reputation. "Is <your brand> any good," "is <your brand> legit," "<your brand> reviews." This is the question a buyer asks right before they decide, and the answer here can quietly sink a sale.

Pull these from real sources, not your imagination — support tickets, sales-call notes, the searches in your own site search bar, the questions people actually email you. The closer these match how a confused human phrases things, the more honest your audit will be.

Step 2: Ask as a stranger, not as yourself

This is the step most people get wrong, and it invalidates everything after it. If you ask from your normal account, the assistant may already know who you are — your past chats, your location, your history — and it will lean toward answers shaped by that. You'll see a flattering, personalized result and conclude you're doing fine. A first-time buyer sees something else entirely.

Strip out the personalization so you see what a stranger sees:

• Use a temporary or incognito chat. ChatGPT has a temporary chat mode; use it, or log out entirely. In your browser, open a private/incognito window so there's no session carrying your identity.
• Turn off memory and personalization where the tool offers it, so past conversations don't steer the answer.
• Check from a neutral location if you can. Answers can vary by country, and you want the version your actual market gets, not the one tuned to your office.

Run every question across the four surfaces that matter: ChatGPT, Claude, Perplexity, and Google AI Overviews (the AI answer that appears above the blue links). They pull from different places and cite differently, so a brand can be invisible in one and named in another. You need all four to know where you actually stand.

Step 3: Ask the same question several ways

Here's the part that surprises people: the answer changes with the wording. "Best <category>" and "most affordable <category>" and "<category> for beginners" can return three different sets of brands. Ask a question once and you've sampled a single roll of the dice.

For each question that matters, ask two or three phrasings:

• The plain version — "best <category> for <use case>."
• A price-shaded version — "affordable" or "cheapest" or "best value."
• A trust-shaded version — "most reliable," "most trusted," or "safest."

You're not trying to game a single phrasing. You're mapping the range of answers a real spread of buyers will get, because your buyers don't all ask the same way. If you appear only under one narrow phrasing, that's a real finding — it tells you your presence is fragile.

Step 4: Log what you see — brands, order, and sources

Do this in a plain spreadsheet. One row per question-and-phrasing, one set of columns per assistant. For every answer, record four things:

• Which brands got named. Write down every one, not just whether you appeared.
• In what order. First-named carries more weight than a brand buried in a closing sentence. Position is a signal, so capture it.
• Which sources or citations the assistant used. Perplexity and Google AI Overviews show their sources directly; ChatGPT and Claude will often name them if you ask "what sources are you drawing on for that?" Write down the actual sites — a review roundup, a Reddit thread, a comparison article, a specific publication. These are the pages doing the recommending, and they matter more than the answer itself.
• The exact date. Answers move over time. A dated log is the difference between a guess and a trend.

The sources column is the one people skip and later regret. It's the map. The assistants don't invent their shortlist — they assemble it from pages they trust, so the fastest way to change what they say about you is to change what those pages say. You can't get cited by ChatGPT until you know which sources ChatGPT is reading.

Step 5: Note whether you appear at all — and who appears instead

Now read your log honestly. For each question, mark one of three outcomes:

• You're named, and placed well. Good. Note the phrasings and surfaces where it holds, and the sources that put you there.
• You're named, but late or hedged. You appear fourth, or with a qualifier ("a smaller option is…"). You're in the room but not on the shortlist. Worth improving.
• You're absent. The question is squarely yours and your brand doesn't come up at all.

When you're absent, the important move is to study who's there instead. Those are the brands winning the recommendation you're not getting. Look at what they have in common: the same three review sites keep citing them, they have a clear comparison page, they're described in consistent language everywhere the assistant looked. That pattern is your to-do list. And check the reputation questions with special care — if "is <your brand> any good" pulls up one angry thread as its main source, that single page is shaping every buyer who asks, and you'd never have known.

Step 6: Repeat weekly and track the movement

A one-time snapshot tells you where you stand. Weekly snapshots tell you which direction you're moving and whether anything you changed actually worked — which is the whole point.

• Keep the same questions and phrasings each week so the comparison is honest. Same inputs, different dates.
• Track two or three named competitors alongside yourself in the same sheet. AI search visibility is relative — being named matters less than being named ahead of the specific rivals a buyer is weighing you against.
• Watch the sources column for change. A new site starting to cite the category is an opening. A source that's gone cold is a lead drying up.
• Give it time. These answers shift on the order of weeks as the assistants re-read the web, not minutes.

A month of weekly logs turns a vague worry — "I don't think we show up in ChatGPT" — into a specific, trackable position you can actually work on.

What to do with what you find

The audit is diagnosis. Here's the treatment, roughly in order of impact. None of it is exotic; it's the unglamorous work of making your brand legible to a machine reading the web.

• Fix entity clarity. The assistant has to understand what you are, who you're for, and what you're not, in plain and consistent language. If your homepage, your review profiles, and your comparison pages all describe you slightly differently, the model can't form a confident picture — and it recommends brands it's confident about. Say the same clear thing everywhere.
• Get cited by the sources the assistants already trust. Your audit named them in the sources column. Getting onto those pages — the review roundups, the comparison articles, the community threads, the "best <category>" listicles the models keep quoting — moves the answer more than anything you publish on your own site. This is the real work behind getting cited by ChatGPT: not writing more, but earning a place on the pages it reads.
• Add structured data. Clear, machine-readable markup on your product and organization pages helps the assistants parse who you are and pull the right facts. It won't win the recommendation on its own, but it removes reasons to get you wrong.
• Publish the comparison and reputation pages that answer the actual questions. If "is <your brand> any good" and "<competitor> alternative" are the questions buyers ask, have honest, specific pages that answer them — the kind a model can quote a straight sentence from.

Each fix should show up, weeks later, in the log you're already keeping. That loop — audit, fix, re-check — is the whole discipline.

Do it yourself, or we do it for you

Everything above is genuinely free and worth doing with your own hands at least once, because nothing teaches you your position faster than watching an assistant recommend a competitor for a question that should be yours. Plenty of brands run their own weekly log and act on it, and that's a good outcome.

Where the free method runs out is scale and consistency: checking four assistants across dozens of phrasings, every week, staying logged-out each time, tracking competitors, and turning the sources column into actual placements — that's a real job, and it's easy to let slip after week two. That's the part we run as a service. Our AI search visibility work is the same method, done continuously and at scale, with the follow-through of getting you cited by the sources your buyers' assistants actually read — or we do this for you if you'd rather keep your team on the store.

But start with the afternoon version. Write the questions, ask them as a stranger, log who gets named, and look honestly at whether it's you. Most brands have never checked. The ones who do get a head start on a channel their competitors can't yet see.