Is my product recommended by ChatGPT?

Open ChatGPT in a private window, or signed out, so your own history doesn’t colour the answer. Make sure web search is on. Then ask the questions a buyer would type — without your brand name in them. If you name yourself, the engine will happily talk about you, and you learn nothing.

• What are the best [your category] tools for [your buyer]?
• What’s a good alternative to [the category leader]?
• I need [the job your product does] on a small budget. What should I use?
• [Competitor A] vs [Competitor B] — which is better for [use case]?
• Which [category] tool is easiest to set up for a team of five?

Write down, for each answer: were you named, in what position, who was named instead, and which sources the answer linked to. Then repeat the same questions in Perplexity and Gemini. Different engines read different indexes and often disagree.

These systems are non-deterministic. Ask the same question twice and you can get different products in a different order — and with search on, each run may read a different set of pages. So one “yes” or one “no” is a coin flip, not a finding.

• Ask each question several times. Being named in 1 of 3 runs is a very different position from 3 of 3.
• Ask across engines. Being a regular on Perplexity says nothing about ChatGPT.
• Track position, not just presence. Named first and named eighth are not the same result.
• Date it. The answers move as the web moves. A result is a snapshot.

Done properly, that is a lot of typing: five questions, three runs, three engines is already 45 answers to read and tally.

With search on, an engine answers from pages it retrieves for the question. Your product gets named when those pages name it. In practice that means:

• Third-party lists and comparisons. “Best X for Y” articles, review sites and roundups are read constantly. If the pages the engines cite for your category don’t mention you, the answer won’t either.
• Community threads. Forums and Q&A threads where real users compare tools show up often as sources. Take part honestly and disclose who you are — seeded or fake threads get removed and cost you the trust you were trying to earn.
• Your own pages that answer the question. A clear statement of what you are, who it’s for, what it costs, and how you compare with the incumbents gives the engine something quotable.
• Consistent category language. If your site calls you a “workflow OS” and buyers ask for “project management software”, you are hard to match to the question.

None of the above helps if the engines’ fetchers are turned away. Check your robots.txt for the crawlers that feed answers:

These are separate from the training crawlers (GPTBot, ClaudeBot, Google-Extended). Blocking training is a legitimate choice; blocking the search and user fetchers usually removes you from the answers without anyone noticing.

• Text without JavaScript. Fetchers often read the raw HTML. A page that is empty until a script runs can look blank to them.
• Identity structured data. Organization or product markup that states your name, site and what you do removes guesswork.
• llms.txt. Cheap to add, harmless, and not a lever on its own.

The manual check is worth doing once. It gets expensive in time when you want an answer you can act on: enough runs to trust the rate, every engine your buyers use, and the sources behind each answer so you know where to show up.

That is what a recomention report does. Questions are written for your category with your name left out, each is asked several times on ChatGPT, Claude, Perplexity, Gemini, Grok with live search on — up to 600 graded answers in one report — and you get the competitors named instead of you, the sources the engines read, whether your site is readable to their crawlers, and a prioritised fix list. Every answer is included verbatim so you can check the reading yourself. How the number is produced.