FrontStat guides

Does ChatGPT recommend my product? How to check properly, and what to do if it does not

By Alex, maker of FrontStat · Updated 8 October 2026 · 7 min read

You can find out in two minutes, and most founders do it wrong the first time. They open ChatGPT, type the product's name, get a flattering paragraph, and conclude that ChatGPT recommends them. It does not. It described them, because they asked about themselves. A buyer never asks about you by name. A buyer asks for a product like yours and gets three names back, and the only question that matters is whether yours is one of them.

Here is how to check it properly, what the three possible answers mean, and what to do about each.

Key takeaways

• Ask the buyer's question, not your name, with ChatGPT Search (web search) on, three times, on ChatGPT, Claude and Gemini.
• Write down who it names instead and which pages it cites; the citations are why.
• To get ChatGPT to recommend you: be describable, publish the page that answers the question, and get onto the pages it already cites. Schema and structured data make you parseable; they do not make you recommended.
• Three in four new products are named by none of the three assistants in launch week. Zero is the normal starting point.

Ask the question a buyer would ask

Write the question the way a customer who has never heard of you would type it. It names the job, not the product. A few real examples from products FrontStat checked this month:

• "What service lets my coding agent deploy a full-stack app online in minutes?"
• "What AI app builder creates native iOS and Android apps, not just web wrappers?"
• "What app tracks the time I spend on a goal so I can improve my productive time?"

If your question only makes sense with your name in it, it is not a buyer question yet. Rewrite it until a stranger could ask it.

Turn web search on, and ask three times

Two settings change the answer completely. First, web search: ChatGPT with search on reads current pages and names current products; without it, it answers from training data, which for a product launched this year is nothing. Buyers increasingly get the searched answer, so that is the one to test. Second, repetition: a single answer varies from one run to the next. Ask three times, and treat "named in two of three" as recommended. FrontStat confirms a change on ChatGPT or Gemini with two extra samples for the same reason.

Then do the same on Claude and Gemini. They disagree with ChatGPT more than you would expect: in launch-week checks across 347 products, Claude recommended 19%, ChatGPT 12% and Gemini 6%, and only 2% of products were named by all three. Each one is a separate audience.

Read the whole answer, not just your name

Three things in each answer are worth writing down.

1. Who it named. If you are absent, these are the products the buyer will try. In most answers the list is short and stable: across the same 347 products, the three assistants agreed on at least one rival for a quarter of them.
2. What it cited. ChatGPT with search shows its sources. These pages are why it named who it named. For new products the hosts cited most were GitHub, the Apple App Store, Reddit, Google Play, Wikipedia and dev.to. Products whose own site was cited in at least one answer were recommended 51% of the time; products never cited, 8%.
3. What it thinks you are. Ask "What is your product?" as a separate question. ChatGPT got it wrong, thin or could not find it for one new product in five. Products all three assistants described correctly were recommended almost twice as often as the rest.

The three answers, and what each one means

It recommends you. Good. Note which page it cited, because that page is doing the work, and check again next week: answers move. If it names you third of three, the two above you are the comparison you should publish.

It names others and not you. The normal case for a new product, and the fixable one. Look at the cited pages. Either they do not mention you, or there is no page about you that answers this question. Both have the same fix, below.

It does not know you exist, or describes a different company. This comes first. Until the assistant can describe you correctly, no page you publish will get you named. Fix the homepage first sentence, the title tag, the markup, and check that your robots.txt lets OAI-SearchBot, GPTBot, ClaudeBot and Google-Extended read the site.

The fix, in three steps

1. Be describable. One plain sentence on the homepage saying what you are and for whom, used identically on every listing and profile, with SoftwareApplication markup. Rank first for your own name on Google; assistants that search start there.
2. Publish the page that answers the question. Title it with the buyer question. Answer it in the first two sentences. Add a short, factual comparison with the products currently named instead of you. This is the page the assistant is looking for and not finding.
3. Get onto the pages already cited. Your report lists them. A GitHub README, an App Store listing, an AlternativeTo entry, a dev.to post, a mention on the comparison page the assistant quoted. Each one is a second source that says what you are.

The full list, with the data behind each step, is in How to improve AI visibility.

A note for local businesses

If buyers ask "best plumber near me" rather than "what tool does X", the same logic applies with different sources. ChatGPT and Google's AI answers lean on Google Business Profile, review sites and local directories for local questions, so a complete profile with the same name, address and description everywhere, recent reviews, and a page per service that answers the local question are the equivalent of steps 1 to 3 above. The check is the same: ask the question a customer would ask, with search on, three times.

Check it in 30 seconds instead

The free checker does the above for you: it reads your site, writes the buyer question, asks ChatGPT, Claude and Gemini with web search on, and shows whether each recommends you, who it names instead and which pages it cited. FrontStat Pro repeats it every day on up to 10 of your questions and emails you the day an answer changes, so you find out when a fix worked, or when a rival took your place.

Questions