It depends far more on what kind of business you run than on anything else I can see. Across 78 free checks, five local service businesses were named in 166 of 375 AI answers (44%). The other 73, small software products, online tools and two services with no town in the question, were named in 172 of 5,336 (3.2%).
Same five AIs, same method, same two weeks. A very different game.
How I ran it
Between October 31 and November 14 2026 I ran my free check on 78 small businesses. For each one I wrote five questions a buyer would type: a “best X for this need” question, one with a price cap, a “what do people like me use” question, an “alternatives to the big name” question and a plain problem with no product words in it. Each question went to ChatGPT, Gemini, Claude, Perplexity and Meta’s Muse three times, with web search on.
That is 75 answers per business. The first 9 checks, from October 31, used four AIs (Muse joined the next day), so 60 answers each. Two checks lost an answer or three to errors. Total: 5,711 answers.
I made no new AI calls for this post. It is the same data I already had, sorted by industry. The businesses stay anonymous.
The table
| Industry | Checks | Named at least once | Answers naming the business | Rate |
|---|---|---|---|---|
| Local service businesses (a town in every question) | 5 | 5 | 166 of 375 | 44.3% |
| Ecommerce and Etsy seller tools | 5 | 3 | 52 of 345 | 15.1% |
| Mac and phone apps | 24 | 12 | 57 of 1,740 | 3.3% |
| Developer tools | 10 | 4 | 21 of 717 | 2.9% |
| Consumer web products | 11 | 5 | 23 of 810 | 2.8% |
| Business and creator software | 21 | 7 | 19 of 1,574 | 1.2% |
| Services with no town in the question | 2 | 0 | 0 of 150 | 0% |
| All | 78 | 36 | 338 of 5,711 | 5.9% |
The local five were a CPA firm, a plumber, a Pilates studio, a family dentist and a web design agency, in cities of roughly 130,000 to 1.7 million people. Every one of them was named, between 19 and 48 times out of 75.
Of the 73 others, 42 were never named in any answer. Business and creator software was the hardest: 14 of 21 never named, median zero.
The ecommerce row is mostly one product. One Etsy seller tool was named in 39 of 75 answers. Without it, the row is 13 of 270 (4.8%).
Why local is a different game
The question narrows the field. “Best plumber for a burst pipe in [your town]” has a short list of possible answers, and the AI has to name some of them. “Best invoicing app” has a few famous answers and the AI reaches for those.
You can see it in who else gets named. In the software checks, the most-named rival showed up in a median of 39% to 72% of answers, depending on the industry. That is one brand owning the question. In the local checks the top rival’s median was 24%, and each local check named between 102 and 200 different businesses across its 75 answers. Local answers are a crowd. Software answers are a monopoly with a few guests.
Claude behaves completely differently. Claude named small software in 13 of 1,095 answers (1.2%). It named the local businesses in 41 of 75. Muse did something similar: 55 of 957 for the 73 others, 45 of 75 for the local five. If you sell software and only test Claude, you will think nobody can be named. If you run a dental office and only test Claude, you will think it is easy.
| AI | Local five | The other 73 |
|---|---|---|
| Muse | 45 of 75 | 55 of 957 |
| Claude | 41 of 75 | 13 of 1,095 |
| ChatGPT | 33 of 75 | 54 of 1,095 |
| Gemini | 24 of 75 | 28 of 1,095 |
| Perplexity | 23 of 75 | 22 of 1,094 |
The part that holds in every industry: your own site
When an AI named a business, it had usually read that business’s own website. Local: 118 of the 166 naming answers cited the business’s own site. Everyone else: 131 of 172.
And the reverse. Of the 42 businesses that were never named, 38 never had their own site cited either, not once in 60 or 75 answers. The AI did not get there at all. That is not a wording problem. It is a reading problem.
When an AI did name one of these businesses, it recommended it in 319 of 338 answers. Being described badly was rare. Being absent was normal.
What this does not show
- The local five are biased upward. I found them through web searches, and I wrote their questions from their own sites, so the niche words match. They are established businesses with real websites, not a random sample of a town. A fair local test needs businesses picked without searching, from something like a chamber of commerce list. I have not run that yet.
- When I asked the generic question instead (“who is the best dentist in Fort Lauderdale”), one or two businesses took most of the answers and most of the town was never named. That data is in who AI recommends in Fort Lauderdale.
- 78 hand-picked businesses, mostly new software. Small groups: five local checks, five ecommerce. Read the rates as a direction, not a market average.
- Answers came through the APIs with web search on and a US location. The app on your phone can answer differently.
What I would do with this, by industry
- Local service business: you are probably already in the crowd. The question is which questions. Ask the specific ones your customers ask (the emergency, the price, the language, the injury), not just “best [trade] in [town]”. The next post has the question-by-question numbers.
- Etsy and Shopify seller tool: the one tool that did well was cited from its own pricing and feature pages. Plain facts on a page an AI can read come first.
- Software, apps, developer tools: the most likely result is zero. Start by checking whether the AIs ever cite your site. If they don’t, nothing you write on it can help yet. Then look at who they name instead, and which pages they read about that rival.
Or let me ask for you. The free check is the same 75 answers as every row in the table above, with the list of who was named instead and which pages each AI read.
How I counted. Source: the Heard of You runner database, the v_answer_final view (one label per answer, a human label beating the judge’s when both exist), for every free-check cycle with at least 60 answers between October 31 and November 14 2026. Excluded: two calibration checks on well-known products, my family’s own dumpling business, one founder’s second pass with different questions, and runs that never finished. Where a business was checked twice, the first complete check counts. “Named” is the mentioned label. “Own site cited” means the answer cited a page on the business’s own domain (answer_citations.is_client_domain). Rival shares come from the products_named list on each answer. Industry groups are mine, from each business’s category. Count date: November 15 2026.