AI assistants name a brand when the question already contains the brand's category, and almost never when it doesn't. In our test of one client across 7 AI assistants, Young Fashion Show was named in 47% of answers to questions like "Is there a kids fashion show in Chicago?", in every answer to questions that used its name, and in only 7% of answers to questions that described the need without the category, such as "confidence-building activities for a shy 10-year-old". The assistants also disagreed with each other by more than three times: ChatGPT named the show in 34 of 41 answers, Claude in 10.

What we tested

Young Fashion Show is a client of ours: a fashion show for kids, teens and adults in Chicago, New York, Miami and Los Angeles. We rebuilt its website and set it up for Google and AI search; it went live on 16 October 2026. We then wrote 41 questions the way its real buyers ask them, and asked every one in 7 AI assistants.

The assistants were ChatGPT, Google AI Mode, Google AI Overviews, Gemini, Claude, Perplexity and Microsoft Copilot. For every answer we recorded whether the show was named, whether it was recommended, whether its website was linked, and whether what the answer said about it was correct. We ran the full set on 17 and 18 October, the first two days after launch, and again on 26 October. The figures below are from 26 October: 276 answers, with 11 Copilot answers still to come because Copilot's daily limit stopped the run.

1. The category word decides most of it

When a parent asks for "a kids fashion show in Chicago", the assistant searches for kids fashion shows in Chicago, and a business that clearly is one has a real chance of being named. When the same parent asks for "something more than a photoshoot for my 9-year-old who loves dressing up", the assistant has to decide on its own that a runway show is an answer. It rarely does. It suggests photography studios, acting classes and theme parks instead.

What to do about it: write pages that connect your buyers' needs to your category, in their words. For Young Fashion Show that means pages such as a runway show as a birthday gift, or as a confidence builder for a shy child, with the real details of how to take part. Most businesses only describe what they are. The buyer who doesn't know your category yet is the one no competitor is answering either.

2. The assistants disagree with each other, a lot

* In the first run, Perplexity's free plan limit blocked 31 of the 41 questions, and 23 Copilot questions could not be asked, so those two are not compared.

The same business, the same questions and the same week produced anything from 10 to 34 mentions out of 41. Being named in ChatGPT tells you nothing about Claude or Gemini, and Google's two AI features did not even agree with each other in the first run. If you only test one assistant, you are measuring one assistant.

3. Being named and being linked go together

Of the 123 answers that named Young Fashion Show, 104 also linked to its website. Of the 153 answers that did not name it, only 4 showed the website as a source. So the website itself is what the assistants rely on when they mention the show: if a page answers the question clearly, the assistant names the business and sends the reader there. ChatGPT linked the site in 31 of its 34 mentions.

What to do about it: make sure every page a buyer's question could land on says, in plain words near the top, who it is for, where, when, how much and how to start. That is what an assistant can quote.

4. Directories and listing sites shape the answers

In the first run, 205 answers showed their sources. The show's own website appeared in 90 of them. Right behind it, a single children's casting directory, kidscasting.com, appeared in 67. Other runway shows' own sites appeared in up to 30 answers each, Eventbrite in 28, Facebook in 26, Instagram in 23 and Backstage in 20.

Assistants build answers from the pages they trust for a topic, and for many topics those are directories, marketplaces, event listings and social profiles, not your website. If your business is missing from them, or described differently on each, the assistant either leaves you out or repeats the mistake.

What to do about it: find the three to five directories and listing sites that appear in the answers for your category, get listed, and use the same name, description, location and website everywhere.

5. Results move, even when nothing changes

Between the two runs, eight days apart, ChatGPT went from 29 to 34 and Google AI Overviews from 9 to 16, while Google AI Mode went from 18 to 16 and Claude from 13 to 10. Some of that is the new website being read and indexed; some of it is simply how these systems work. The same question can get a different answer tomorrow.

What to do about it: measure on a schedule, with the same questions, and look at the trend across all the assistants, not at a single answer.

6. Check what they say, not only whether they name you

When the assistants did name the show, they were mostly right: 105 of the 123 mentions were fully correct when checked against the website, 17 were partly right, for example an old date or a missing city, and 1 was wrong. A mention with the wrong price or the wrong date can cost you more than no mention at all, so every test should check the details, not just count names.

What this means for your business

1. Write down 30 to 50 questions your buyers really ask, including the ones that describe the need without naming your category.
2. Ask them in all the main assistants, not one, and record named, linked and correct for each answer.
3. Answer the questions on your own pages, in plain words, with the real details near the top.
4. Get listed, consistently, on the directories the answers already use.
5. Repeat the test every month and fix the pages behind the wrong or missing answers.

The steps we took for Young Fashion Show are in How to get your business recommended by ChatGPT, and the full project is in the Young Fashion Show case study. To check your own site in 20 minutes, use our 10-point check.

The limits of this test

This is one business, in one category, in four US cities. It is a careful field test, not a statistical study, and the numbers will differ for your business and your buyers. Some answers were read before the assistant had finished writing; we marked those. One ChatGPT answer was shaped by the signed-in account's memory, so we left it out of the judgement. Nobody can guarantee an AI recommendation, and we don't.

Want this test for your business?

We run the same test for clients: your buyers' questions, every main assistant, and a report on where you are named, where you are missing and what to fix first. It is part of our Google and AI search setup, from $490 once for one of SEO, AEO or GEO, or $990 for all three. Add the free AI check to your plan or start a project.

Get the free AI check Start a project

Questions people ask

Which AI assistant is most likely to name a small business?

In our test, ChatGPT named the client in 34 of 41 answers, Perplexity in 22, Google AI Mode and AI Overviews in 16 each, Gemini in 14 and Claude in 10. That is one business in one category, so test your own questions in every assistant rather than relying on one.

Why does ChatGPT name my competitor but not me?

Usually because the competitor's pages answer the buyer's question more directly, or because the directories and listing sites the assistant trusts mention the competitor and not you. Check which sources the answer cites, and make sure your pages and listings say clearly what you do, for whom and where.

Do AI assistants use my website or other sites?

Both. In our test, the client's own website was the most common source, but a single directory appeared in 67 of 205 answers that showed sources, and event listings and social profiles appeared often too.

How often should I test AI answers about my business?

Monthly is a good rhythm, with the same questions each time. Answers moved by several mentions per assistant in eight days in our test, so one check tells you little; the trend across assistants is what matters.

Can AI answers about my business be wrong?

Yes. In our test 105 of 123 mentions were fully correct, 17 were partly right and 1 was wrong. Keep dates, prices and locations current on your site and listings, and check the details in every test.

Read next

• Is your website ready for ChatGPT? A 10-point check
• How to get your business recommended by ChatGPT, Google AI and other assistants
• Website redesign checklist for 2026: what to keep, what to drop

Sources

1. Project 66 Studio, Young Fashion Show AI test: 41 questions across 7 AI assistants, first run 17 to 18 Oct 2026, second run 26 Oct 2026.
2. Young Fashion Show: from 1 in 74 to 1 in 12 Google visitors applying, Project 66 Studio (read 26 Oct 2026).