Does ChatGPT name your product? How to check AI answers yourself

Write them the way a buyer types them into a search box or a chat, without your product’s name: what they want (“free alternative to [competitor] for Mac”), the problem (“how do I … without …”), a complaint about the tool they use now, and one or two that name competitors (“[A] vs [B] vs [C]”). A question only an existing customer would ask tells you nothing.

Read each question and ask whether someone who has never heard of you would type it. Search two of them in a private window: if the results are about a different kind of product, reword the question.

Our 10 questions for Upright, Basalt Studio’s free Mac app for Logitech mice, included “free alternative to Logi Options+ for Mac”, “Logi Options+ uses too much CPU on my Mac, what can I use instead” and “Mac Mouse Fix vs SteerMouse vs BetterMouse”. The questions that named a competitor (5 and 7) or Logi Options+ itself (3 and 9) got answers about those products. They are worth asking, and hard to win.

Ask every question in Perplexity, ChatGPT and Google, with a new chat or a new search for each. Use a private window and stay logged out of ChatGPT and Perplexity. For Google, use a browser profile that is not signed in, so your own history does not shape the AI Overview. Note the date and time. Two phrasings of each question give a fairer picture than one.

Before you count an answer, check that it searched the web: Perplexity and ChatGPT list their sources with the answer, and an AI Overview shows its links. An answer with no sources came from the model alone and tells you less about what to fix.

Asked on 20 Oct 2026, between about 01:37 and 01:52 Türkiye time: 17 minutes of asking. Perplexity and ChatGPT were logged out and searched the web for every answer, and an AI Overview appeared on all 10 Google searches. Two things fell short of the method above. Google was read in a browser signed in to a personal Google account, so its answers may be lightly personalised, and each question was asked in one phrasing only. Gemini was not run, so we have 30 answers, not 40.

One row per answer: the date, the assistant, the question number, whether you were named, every product named in the order it appears, and the domains of the sources cited. Count a product as named only when the answer offers it as an option, not when it comes up as the tool being replaced. A tab-separated file, ai-log.tsv, with these columns is enough:

date	engine	question	named	products	domains

Once the log has rows, the first command counts the answers that named you and the second the products named most often. Change -f5 to -f6 to count the cited domains the same way. The commands work the same in zsh and bash.

cut -f4 ai-log.tsv | sort | uniq -c
cut -f5 ai-log.tsv | tr ',' '\n' | sed 's/^ *//' | sort | uniq -c | sort -rn | head -8

Run on the 30 answers of our check, as published in the sample report, the commands printed this. “no” 30 times: Upright was named in 0 of 30 answers.

  30 no
  15 BetterMouse
  14 Mac Mouse Fix
  12 SteerMouse
  12 LinearMouse
  11 Logi Options+
   8 OpenLogi
   5 macOS built-in settings
   5 Mouser

Sort the cited domains into four piles: your own site, competitors’ sites, places anyone can be listed (GitHub, AlternativeTo, Reddit, app stores, directories) and press. The third pile is the one you can act on this month. A domain cited in many answers is worth more than one cited once.

For each listing site in the log, search it for your product and for your main competitor. Where they have a page and you do not, you have found a gap.

reddit.com and github.com were each cited in 12 of the 30 answers, alternativeto.net in 5, and upright.hellobasalt.dev in none. Every open-source Logitech tool the assistants named was cited through its GitHub repo. Upright’s repo described it as a tool for the MX Vertical only, and Upright had no AlternativeTo page.

Rank what the log shows by what it is likely to change. First, whether search engines can find your pages at all: the assistants answer from web search. Then the listings the answers cite, then the products they name that your own pages leave out, then the questions no page of yours answers.

Next to each fix, write the question or the domain that points to it. The next check can then tell you whether that fix changed anything.

What the 20 Oct log told us to fix, and where each stood on 21 Oct:

• Get indexed. No page from upright.hellobasalt.dev appeared in web search for any of the 10 questions; the site was days old. A page that is not indexed cannot be cited.
• Name the apps the assistants name. OpenLogi was named in 8 of the 30 answers and Mouser in 5, and Upright’s alternatives page mentioned neither. On 21 Oct it covers both, under the title “Logi Options+ alternatives for Mac (2026): 7 apps compared”.
• Fix the GitHub repo. By 21 Oct its description begins “Free, open-source Logi Options+ alternative for Mac” and it carries the logi-options-alternative topic.
• Answer the MX Master questions. A page for the MX Master 3S went live on 20 Oct.
• Get listed. A Homebrew tap was not there yet on 21 Oct. An AlternativeTo listing is on the same list.

Same questions, same wording, same assistants, same kind of session, around the same day each month. Compare who is named and which domains are cited with the month before. When you ship a page meant to answer a question, add that question.

Answers change from one run to the next, even minutes apart, so one answer that changed is noise. Look for a pattern across several questions or several months: a site you got listed on starting to be cited, or your product named by more than one assistant.

The next run asks the same 10 questions in the same three assistants. Questions 3 and 9 ask about the MX Master and will show whether the new page is picked up. Until then we cannot tell you whether any of these fixes changed an answer.