This post gives you a manual method you can run in an afternoon with no special software. It also covers why a single answer proves nothing, which mistakes make the results meaningless, what tends to influence whether a brand gets mentioned, and what to do with what you find. Where it helps, it points to the AI Search Visibility Check, which automates part of the same process.
How people use assistants to find brands
A growing number of customers skip the list of links and ask a direct question: which project tool suits a ten-person agency, who does emergency plumbing in my area, what is a good accountant for freelancers. The assistant answers in a few sentences and usually names a small number of options. Where a results page shows ten listings, an answer might name two or three.
That changes what visibility means. On a search page a lower position still gets some clicks. In a written answer, a brand is either named or it is absent, and the way it is described shapes how the reader sees it. An assistant that calls your premium service a budget option, or lists a product you discontinued, does damage that never appears in your analytics.
You will see terms such as ChatGPT SEO, LLM SEO and generative engine optimization for the work of influencing those answers. The labels vary and the practice is young. Nobody outside the companies that build these assistants can say exactly how they choose which brands to mention, so treat any claim of a guaranteed method with suspicion. What you can do is measure what is happening and improve the information available about your business.
Why one answer proves nothing
If you ask an assistant a question twice, you may get two different answers. These systems generate text with some built-in variation, so the order of brands, the wording and sometimes the list itself can change from one run to the next. A single answer that includes you is a pleasant moment. A single answer that leaves you out is not a verdict either.
The wording of the question matters as much. "Best accounting software for small businesses" and "accounting software for a two-person design studio that invoices in euros" are different questions, and they can produce different names. Customers phrase things in dozens of ways, so any test built on one phrasing tells you about that phrasing only.
The assistant you use matters too, and so does whether it is searching the web at the moment you ask. Different assistants draw on different material, and the same assistant can behave differently depending on its settings. Record which assistant and which mode you used, so you can repeat the test under the same conditions later.
The practical conclusion is simple. You need several realistic questions, each asked more than once, and you need to report results as a share (for example, named in 6 of 20 answers) instead of a yes or a no.
A repeatable method to check AI search visibility
Start by writing 8 to 12 questions a real customer might ask. Mix three kinds. Category questions name no brand: "what is a good CRM for a small recruitment firm". Comparison questions name two options: "is X better than Y for a small team". Problem questions describe a need without a product: "how do I stop losing track of leads from my website". Write them in plain, casual language, the way a customer would type them, and include your location or audience where it matters.
Open a fresh conversation for each run so earlier answers do not influence later ones. Ask each question at least three times, ideally on different days. Do not include your brand name in category and problem questions, because that is the one thing that would make the result meaningless.
Record the results in a spreadsheet with one row per answer. Useful columns are: the question, the run number, whether your brand was named, its position in the list, how it was described in a few words, which competitors were named, and whether anything about you was wrong. Copy the description exactly. The wording around a mention often matters more than the mention.
When the runs are done, count. For each question, how many of the runs named you? Which competitors appear most often? Which descriptions repeat? Patterns across runs are real. A single odd answer is noise.
What tends to influence whether a brand is mentioned
No one can promise what moves an assistant's answers, but a few things tend to matter, and all of them are within your reach. The first is a clear description of what you do on your own site. If your service page says in plain words what you offer, for whom and where, there is something accurate to draw on. Pages full of slogans give a summarizer very little to work with.
Consistency helps. Use the same business name, the same description of your services and the same basic facts on your site, your profiles and your listings. If you appear under three slightly different names, an assistant has a harder time treating them as one business.
Presence in places an assistant may learn from also tends to matter: reputable directories in your field, industry articles that mention you, review sites, and your own structured data, which labels facts such as your organization name and services in a form software can read. The technical SEO audit checklist covers that markup and the crawlability checks that sit underneath it.
Reviews and third-party mentions give an assistant descriptions of you that you did not write yourself. These are tendencies, not rules. You can do all of the above and still go unmentioned for some questions, and a competitor with less careful pages may appear anyway. Plan the work as improving the quality of public information about you, and measure the effect over time without expecting a particular outcome.
Common mistakes when testing AI search visibility
The first mistake is asking one question once. It feels like a test and gives you a coin flip. The second is leading questions that contain your brand name, such as "why is Acme the best option for payroll". The assistant will happily talk about Acme, and you learn nothing about whether it would have suggested Acme unprompted.
The third is treating one result as a score. A figure like "we appear 30 percent of the time" is only meaningful when you know how many questions and runs it came from, and it will move a little each time you repeat it. Differences of a few points between two rounds are often just variation. Look for larger and repeated changes before you conclude anything.
A fourth is testing only questions you already do well on. It is natural to write questions that match your best pages. Ask at least a few that you expect to lose, because those show where a competitor owns the conversation.
The last is ignoring accuracy. Being named with the wrong location, an old product line or an incorrect claim can be worse than not being named. Check every statement an answer makes about you against your own site, and note the ones that are wrong.
Automating the check with the AI Search Visibility Check
The manual method works, and it is slow. Writing questions, running each several times and logging every answer takes hours. The AI Search Visibility Check from Dynamic Growth runs a version of this for you. You enter your brand, website and a few topics. The software writes realistic customer questions, puts them to an assistant, and reports how often your brand and your competitors appear and how they are described, with the full answers included so you can verify them.
It is honest to say what it does not do. It asks Claude only. ChatGPT, Gemini and Perplexity are left out, and the report names which assistant answered. The result is a measured share across the questions and repeats it ran. Assistants do not publish a ranking to report. It reads public information only and has no access to your accounts. You can compare it with a Competitor Comparison if you want to see how the sites of the brands that keep appearing differ from yours.
Software produces the report, so read it as a guide and check the answers yourself before acting. A free trial lets you look at the report format, and the results are most useful as a baseline you can compare with a later run.
What to do with the results
Sort what you found into three piles. The first is errors: wrong facts, old services, wrong locations. Run an SEO Audit if you suspect the cause is unclear titles or headings. Correct the source on your own site first, then check any profiles or listings that repeat the mistake. The second is absence: questions where you were never named. Open the pages that should answer those questions and ask whether they state the answer plainly in the first few sentences. The third is competitors who appear often. Look at how they are described and at what their pages say that yours do not. The competitor website analysis method works well for this step.
Pick a small number of changes, make them, and leave the rest alone for a while. Rewriting everything at once makes it impossible to tell what had any effect. A clearer service page, a short FAQ that answers the questions customers really ask, consistent business details and a few well-chosen listings are reasonable first moves, and each costs little.
Because answers vary and assistants change, repeat the test on a schedule. Quarterly suits most small businesses. Between rounds, make sure your own pages stay intact, since a template edit can remove the very text an assistant was using. Weekly Monitoring re-runs the Marketing Audit on your site every seven days and alerts you if something gets worse.
Your next step: write your 8 to 12 questions today, run each one three times in fresh conversations, and fill in the spreadsheet. Once you have those first numbers, choose the single question where you were missing and rewrite the page that should answer it.
Frequently asked questions
How do I check if ChatGPT mentions my brand?
Write 8 to 12 realistic customer questions that do not contain your brand name. Ask each one several times in fresh conversations and log whether you are named, in what position and how you are described. Report a share across all runs, because a single answer varies and proves little.
Can I pay to be mentioned by AI assistants?
Nobody can reliably sell you a mention, and you should be wary of anyone who claims to. What you can do is publish clear, accurate information about your business, keep your details consistent across the web, and earn reviews and third-party coverage. Then measure whether your mentions change over time.
Why does the assistant give different answers each time I ask?
These systems generate text with some variation, so the wording and the list of brands can shift between runs. The way you phrase the question also changes the result. That is why a useful test asks many questions several times and counts how often your brand appears.
Is AI search optimization different from SEO?
They overlap. Clear pages, accurate details, crawlable sites and trustworthy mentions help in both. The difference is the output: search pages show ranked links, while assistants write an answer that names a few brands. Keep doing sound SEO and add the question testing described here.
How often should I test AI search visibility?
Quarterly is enough for most small businesses, with a re-test after you change key pages or when a competitor launches something new. Use the same questions each time so the rounds are comparable, and look for larger, repeated changes before drawing conclusions.