How to Check If AI Mentions Your Brand

To check if AI mentions your brand, ask ChatGPT, Claude, and Gemini the questions your buyers actually ask — "best [your category] for [your customer]" — in fresh sessions, and write down which brands each answer names. Below is the exact manual method: free, about thirty minutes, and a way to score the results that survives the fact that answers change between runs.

A word on what this measures. Not whether an assistant recognises your name — that tests memory. Whether it puts you in front of someone who has not heard of you yet, and is asking which product to buy.

Why AI mentions are worth measuring

Two measurements are worth more than the industry's estimates. First, adoption: 49% of US adults say they use AI chatbots, according to a Pew Research Center survey of 5,119 adults conducted in February 2026 — roughly double the share from 2023.

Second, what happens to the click. Pew also tracked the actual browsing of 900 US adults across 68,879 Google searches in March 2025: when an AI summary appeared, users clicked a regular search result in 8% of visits, against 15% when no summary appeared. They clicked a link inside the summary in 1% of visits.

Read honestly, that is the whole argument. The answer increasingly replaces the visit, so what the answer says about you is the thing worth measuring. What nobody can tell you is a conversion rate for AI mentions: no provider publishes how citations are chosen, and any precise "brands cited get X% more traffic" figure is an extrapolation from someone's sample, not a law. Your own answers are the only data that describes your market.

How do you check manually — free, in thirty minutes

1. Write down ten buyer questions in your customers' words — "best invoicing tool for freelancers in Canada", "which CRM should a five-person agency use" — not your brand name and not industry jargon.
2. Open a fresh chat with each assistant — ChatGPT, Claude, Gemini — with history and memory off where the product allows it. A contaminated session tells you what the model knows about you, not what a stranger hears.
3. Ask each question and record every brand named, in the order it appears.
4. Rerun each question two or three times. Answers vary between runs, and a brand that appears once in three runs has weak presence, not visibility.
5. Put the results in a spreadsheet and repeat monthly, because models and the sources behind them keep changing.

The arithmetic sets the ceiling on this: three assistants times ten questions times three runs is ninety answers to collect and read every month, for one market. As a diagnostic it works well. As a monthly routine it is the first thing to fall off the list in a busy week, which is the honest reason automated monitoring exists — ours included.

How do you score what comes back?

Raw answers are not a measurement until you decide what counts. Five columns are enough, one row per run:

• Named or not. The binary question, and the only one that matters when the answer is "not".
• Position in the list. First and second are recommendations; eighth is a footnote in a list nobody reads to the end.
• Framing. "The standard choice for X" and "there is also Y" are not the same mention, even though both name you.
• Which competitors appear, and how consistently. The names that show up in every run are the ones the model treats as the category.
• Whether your own site was cited as a source, or you were named from the model's memory with someone else's page linked underneath.

Then score across runs rather than screenshots: out of the thirty answers for one assistant, how many named you? Three out of thirty is noise; twenty out of thirty is a position worth defending. Track that share month over month and treat single answers the way you treat a single search ranking check — direction, not truth.

What should you do if AI never names you?

Zero visibility is a normal starting point, not a verdict. Work in this order.

First, close the content gap. Assistants recommend products whose pages answer the exact question that was asked, so the ten questions you just tested are also your editorial list — one page per question, written to answer it rather than to rank for it.

Second, fix the technical floor, so the facts you do publish are readable: pages that carry their content in the HTML the server returns, Schema markup on prices and products, and an llms.txt map of what matters on your site. This is one-off work.

Third, earn mentions somewhere other than your own domain. Models lean on sources they consider trustworthy, and one substantial mention in a publication your industry actually reads does more than ten posts on your own blog.

Then re-check monthly, with the same questions, and watch the share of runs that name you.

Frequently asked questions

Which AI assistant should I check first?

Start where your buyers are. ChatGPT has the largest consumer audience, so it is the first stop for most B2C and small-business products; Claude and Gemini turn up more in B2B workflows, and Perplexity users tend to be researching in depth. Checking those four covers most commercial questions being asked today.

How often should I re-check AI answers?

Monthly is the floor: models, search indexes, and the pages they draw on all change, so your position moves even when your site does not. While you are actively shipping changes, weekly checks tell you sooner whether they landed.

Can I just ask the AI directly whether it knows my brand?

You can, but it answers a different question. "Do you know [brand]?" tests recall, and recall does not sell anything. What matters is whether the model brings you up unprompted when someone describes their problem — so test unbranded questions, the ones your buyers actually type.

Find out whether AI sends buyers to you — or past you

The manual grid above gives you the answer for free, in an afternoon. If you would rather have it in three minutes and repeated for you, that is what we built: our free audit runs 3 buyer questions through the AI panel, records which brands get named instead of you, and turns the gaps into fixes you can ship — llms.txt, Schema markup, titles, FAQ blocks. Either way, start with the measurement.