How to Find Out What AI Actually Says About Your Brand
The method is simple: write a fixed set of questions your buyers would actually ask, put every one of them to ChatGPT, Claude, Gemini and Perplexity, and record which brands get named. Then repeat the same set next month without changing a word.
The whole thing takes an afternoon the first time. What follows is how to do it so the result means something — and, further down, what changes once you stop doing it by hand.
Why one prompt, once, tells you nothing
This matters more than most teams assume. Semrush's 2026 AI Visibility Index, built from 126 million U.S. AI search prompts, found that 45% of marketing leaders cannot accurately measure their brand's visibility inside AI-generated answers, and only 9% have tools that track all the relevant metrics across platforms. Most people who think they know how AI talks about their brand are working from a handful of screenshots, not a method.
Step 1: build the question set
This is the part people rush, and it decides whether the exercise is useful or theatre. You want twenty questions, spread across four kinds of intent.
Brand questions name you directly. "What is [company]?" "Is [company] any good?" These are easy to appear in — the model was handed your name. What they test is accuracy, not competitiveness.
Category questions ask for options without naming anyone. "What are the best invoicing tools for a small business?" This is where a shortlist gets built and where you find out whether you are on it.
Problem questions describe a situation with no category named at all. "How do I stop losing track of client invoices?" These are the hardest to win and reach people who have never heard of your category, let alone you.
Comparison questions weigh named options. "[Company] vs [competitor]" and "alternatives to [competitor]". Highest buying intent of the four.
Five of each gets you to twenty. Write them the way a person would type them — full sentences, not keyword fragments. Someone asking an assistant writes differently from someone typing into a search box, and that difference changes the answer.
Step 2: ask them properly
Three rules, and skipping any of them contaminates the result.
Start a fresh conversation for every question. Models carry context. If you ask about your company and then ask a category question in the same thread, the earlier mention influences the later answer. You will conclude you are doing better than you are.
Turn off personalisation and memory where you can. An assistant that has learned you work at a company will name that company more readily. That is your own history reflected back at you, not a signal.
Do not lead. "What are the best tools like [your company]?" guarantees a mention and proves nothing.
Step 3: record what matters
A spreadsheet with one row per question and platform. Four columns are enough, but the first two are easy to collapse into one — and shouldn't be.
- Were you named? Yes or no. No partial credit.
- Were you cited? A mention and a citation are not the same thing — a mention is your name appearing in the text, a citation is the model actually linking to or naming your site as a source. Track them separately. Semrush's Index found that on Gemini specifically, the overlap between brands that get mentioned and domains that get cited can be as low as 30%, which means most teams tracking only one of the two are missing the other entirely.
- Where in the answer? First recommendation, middle of a list, or a passing aside at the end. Position matters because readers stop.
- Which other brands were named, and which sources were cited? All of them, in order.
Eighty rows, twenty questions across four platforms. That is your baseline.
Step 4: read the competitor column first
Most people look at their own hit rate and stop. The more valuable column is the one listing everyone else.
Count how often each brand appears across all eighty answers. That ranking is how AI systems currently understand your competitive set — and it is routinely different from the list on your internal slide. Companies you have never considered rivals show up constantly. Companies you benchmark against turn out to be invisible too.
The cited sources column is nearly as useful. If the same three review sites appear beneath most answers and you are listed on none of them, you have found a concrete gap that no amount of work on your own website will close.
Why the four platforms disagree with each other
This is also why the check has to run across all four platforms, not one. The Semrush Index quantified the gap: ChatGPT cites an average of 15 sources per response and leans on community and reference sites like Reddit and Wikipedia, while Gemini cites an average of just 3 sources per response, drawing from a much narrower pool. Visibility concentration also varies enormously by category — in the study's most concentrated industries, the top three brands captured over 80% of all visibility; in the most open ones, under 42%. Only 36 global brands held top-100 visibility on every platform, every month, for the entire study. Almost every other brand, including large ones, is strong somewhere and close to invisible elsewhere. If your question set only ever runs on the assistant you personally use, you are seeing one slice and calling it the whole picture.
Why asking once tells you almost nothing
These systems are probabilistic. Ask the same question twice, five minutes apart, and you can get two different answers naming two different sets of brands. Neither is wrong.
This has two consequences worth internalising. A single answer that names you is not proof of visibility, and a single answer that omits you is not proof of a problem. What means something is the rate across many questions, and the direction that rate moves over months.
It also means a screenshot proves nothing — to you, your team, or a client.
What content actually moves the number
Once you have a baseline, the obvious next question is what to change. The clearest evidence comes from the GEO: Generative Engine Optimization research out of Princeton, Georgia Tech, and the Allen Institute for AI, which tested nine content interventions across a 10,000-query benchmark. Adding quotations from authoritative sources and adding relevant statistics produced the largest measured gains — roughly 30% to 40% relative improvement in citation visibility over an unoptimised baseline. A separate meta-analysis of 54 AI-citation studies found that whether a page is crawlable at all and whether it already ranks for the query are still the two strongest predictors of all — the fundamentals underneath AEO and GEO are still ordinary SEO.
The honest limitation, and what a tool actually adds
Done properly, the manual version above takes three to four hours. Doing it monthly, across four platforms, while keeping the question set frozen and the records comparable, is a recurring job that competes with everything else on your list — which is the entire reason tools for this exist, including ours.
Lazycats.ai is built around exactly the process described above, automated: it is an LLM Check tool that shows how ChatGPT, Gemini, Perplexity, and Claude actually talk about your brand, a visibility check across all four platforms rather than whichever one you happen to open, and a way to increase AEO visibility by turning what it finds into a prioritized list of what to fix next. It runs the fixed question set, keeps it frozen between runs, and separates mentions from citations the way Step 3 above describes — because, as the Semrush data shows, treating those as one number hides a third of the picture on some platforms.
Run the manual version once before you automate anything. You will understand what the numbers mean far better for having produced eighty rows by hand, and you will know whether the questions in your set are the ones your buyers actually ask — which is exactly what any tool, including ours, needs to get right to be useful at all.
LazyCats AI plans, compared
Which plan fits which stage
The Free Audit is the one-time version of Step 1 through 4 above, done for you: one site, one scan, and the top three issues, enough to see whether the exercise is worth doing properly. Starter is for a single site that needs the check to actually repeat — weekly, unlimited audits, the full issue breakdown, and the full citation and source map, at $19.99 a month. Agency is built for anyone running this process across client sites rather than one brand: five sites, sentiment analysis, shareable report links, and white-label reporting so the output can go out under your own name, at $49.99 a month. Scale adds the two things that only make sense past a certain size — a Refresh Tracker that watches for changes between scheduled audits instead of waiting for the next one, and team access — across up to 20 sites for $99 a month, which works out to the lowest per-site cost of any plan.
What the category is actually for
If you want to know what AI actually says about your brand, you need more than a one-time prompt test. You need a structured check across models, prompts, mentions, and citations, repeated on a schedule you don't have to remember. Whether you build that spreadsheet by hand this month or hand it to a tool next month, the method is the same — the only thing that changes is who is doing the repeating.
