How to Find the Prompts Your Customers Ask ChatGPT

On 2026-10-25 Leadee asked 56 US buyer questions about the 7 websites of Netbase JSC, the company behind Leadee, on three engines. For this guide we re-counted the answers by the shape of the question. A run counts only if the brand was named unambiguously. The numbers are pooled over Perplexity sonar, Google AI Mode and OpenAI: Perplexity and AI Mode ran each question 3 times, OpenAI 0-3 times before the account ran out of credit.

The gap is the point. When the question contained the brand name, the brand was named in about three answers in four. When it did not, about one in fourteen. Among unbranded questions, "best ..." lists and comparisons did a little better than how-to questions, but the intervals overlap, so we cannot rank them. Two cautions: the shapes come from our own text rule, and each brand has its own niche and size, so the shape effect is mixed with the brand effect.

One concrete example. OpenAI (gpt-5 with web search, through the official API) named CMSmart (cmsmart.net, part of Netbase JSC) in 2 of 3 runs of "best WooCommerce product designer plugin for print on demand", and in 1 of 1 run of "Fancy Product Designer alternatives for WooCommerce". Both are unbranded, both name the job and the platform, and the second names a market player. That is the shape of prompt worth finding: specific, in the buyer's words, and without your name in it. With 3 runs it is an indicator, not a rate.

No. OpenAI does not publish what users type into ChatGPT, and there is no ChatGPT equivalent of Google Search Console that shows you the prompts behind your visits. Any tool that shows "ChatGPT prompt volume" is estimating from other data, and Leadee is no exception: our Prompt Finder generates phrasings from a template; it does not read anyone's chats.

What you can do is reconstruct the questions from places where buyers already put them in writing. The steps below go from the most reliable source (your own customers' words) to the least (generated patterns).

Open the last three months of sales enquiries, support tickets and chat transcripts, and copy every question a prospect asked before they bought. Ask your sales team for the first three questions they hear on most calls, and write them down word for word. These are the closest thing you have to real prompts: the same buyer is likely to put a similar question to an assistant.

Note the context that comes with each question: company type, country, budget, the tool they use today. Prompts to ChatGPT often carry that context ("for a small print shop", "in the UK", "that works with WooCommerce"), and it changes which companies get named.

If your site has a search box, export what people typed into it. Then open the Queries report in Google Search Console and filter for question words and comparison words: how, what, which, best, vs, alternative, cost. These are searches that already found you, phrased by real people, and many of them are the short version of a longer ChatGPT prompt.

Do not copy keywords as they are. "woocommerce product designer" becomes "what is the best WooCommerce product designer plugin for a print-on-demand store?" when a person asks an assistant.

Buyers who are choosing ask list and comparison questions. Write them for your category: "best [category] for [type of customer]", "[market player] alternatives", "[market player A] vs [market player B]", "is [market player] worth it for [use case]". Use the names of the market players your buyers actually compare you with; your sales team knows them.

In our own data these list and comparison questions were where unbranded mentions came from, including the CMSmart example above. They are also the questions where you can see who takes the place you want. Our guide to getting products recommended by ChatGPT covers what to do once you know them.

Run your main keywords through the free ChatGPT Prompt Finder. It returns candidate prompts grouped by intent (informational, commercial, transactional) and funnel stage. Keep only the ones that sound like something your customers have actually said; delete the rest.

Generated prompts are useful for coverage, because they remind you of question types you missed. They are the weakest source on this list, so never let them outnumber the questions from steps 1-3.

Label each prompt: branded (contains your name), best/list, comparison/alternatives, how-to, other. Keep the branded ones in a separate small group; as the table above shows, they make almost any brand look visible. Your main list should be 10-20 unbranded prompts that cover the buying stages and the markets you sell in.

Merge near-duplicates and write each prompt as one complete question with its context. A list you can run every week beats a long list you run once.

Ask each prompt in ChatGPT several times and record whether your brand is named and cited, as described in how to check if ChatGPT mentions your brand. Put the results in a ChatGPT prompt tracking sheet so you can repeat the same list over time.

Prompts where market players are named and you are not are your to-do list: the pages ChatGPT cites for them show where you need to be present, which our guide on getting cited when ChatGPT prefers other websites explains. The other ChatGPT guides are on the ChatGPT SEO overview.

Show the list to two people who talk to customers every week and ask them to strike out any prompt a real buyer would never say. Then check the mix: at least half the prompts should come from steps 1-3, most should be unbranded, and each important product or service should have at least one list question and one comparison question.

Re-check the list every quarter. New market players, new features and new buyer worries change the questions.