Nobody publishes a list of what people ask ChatGPT. OpenAI releases no prompt volumes, and keyword tools measure Google, where the phrasing is different and the volumes for full questions are usually too small to register. When we checked, "generative engine optimization" had 4,400 US searches a month while "how to get recommended by chatgpt", the way a person actually asks, had 20, and "how to get cited by perplexity" had no data at all.
So the prompts you track have to be built, not downloaded. This guide is the method we use: collect candidate questions from five places, group them by intent, pick 10 to 50, write them in buyer language, and keep the wording fixed so your measurements stay comparable.
We make Seen42, an AI visibility tool, so read our product notes with that in mind. Everything else in this guide works without it.
Why the prompt list matters more than the tool
Every visibility number you will ever see is "share of answers to these prompts that named us". Change the prompts and the number changes. Ten branded prompts give you a flattering score and no information about new buyers. Ten vague category prompts give you a list of market leaders you cannot displace this year. The useful list sits in between: specific enough that a small company can be the right answer, common enough that buyers actually ask it.
Five places to find candidate questions
Start by asking the assistants. Ask ChatGPT, Perplexity and Gemini "What questions do buyers usually ask before choosing a [category]?" and "What is the best [category] for [audience]?" for each audience you serve. When we asked Perplexity how to do this on 2026-10-20, it described exactly this, with the caveat that assistants can map likely themes but cannot say how often each is asked. Treat the output as candidates, not data.
Mine your own conversations. Sales calls, demo questions, support tickets, onboarding emails and your site search log contain the exact words buyers use. Keep the phrasing. "Does it work with Xero" is a prompt; "integration coverage" is not.
Export long queries from Google Search Console. Filter for queries containing "best", "vs", "alternative", "how", "compare" and "for", and for anything over six words. These are the closest observable proxy for conversational prompts, and they come with real impression counts.
Read the questions on review sites and forums. The titles of G2 and Capterra comparison pages, the questions in their Q&A tabs, and recommendation threads on Reddit and industry forums are buyer prompts written by buyers. Note which competitors are named together; those pairs become comparison prompts.
Ask the assistants about your competitors. "Alternatives to [competitor]" and "[competitor] vs [competitor]" surface the comparison prompts your category already runs on, and the answers show which tools are currently named for them.
Group by intent
Sort candidates into four groups and keep some of each.
- Category: "best [category] for [audience]", "which [category] should a [type of company] use".
- Comparison: "[A] vs [B]", "alternatives to [A]", "is [A] worth it".
- Problem: the job to be done, with no product named. "How do I run payroll for contractors in three countries."
- Branded: "pros and cons of [you]", "what do customers say about [you]". Two or three, kept out of your headline number, so you can see how answers describe you once they name you.
Pick the panel
Ten prompts is enough to start; fifty is plenty for a single product. For each candidate ask three questions. Would a buyer who does not know us ask this? Could we honestly be a good answer? Is there a page we could publish that answers it? Keep the ones that pass all three. Then write each one as a person would type it, in the first person where that is natural, with the constraint included: team size, budget, country, platform. "Best helpdesk for a 10 person SaaS that needs Slack" is better than "best helpdesk software".
Keep a control group
Before you build anything, mark a few prompts as controls: prompts you will not publish pages for. AI answers drift on their own. If your targeted prompts improve and the controls do not, you have evidence your pages did something. If both move together, the assistants changed, not you. Our guide to measuring AI visibility explains the method.
Fix the wording and the schedule
Write each prompt once. If you later want different wording, retire the prompt and start a new one with its own history; editing a prompt in place quietly breaks every comparison with earlier checks. Ask the same prompts, on the same assistants, at the same interval, in fresh sessions, and record who was named, in which position, and which pages were cited.
What to do with the list
The list is a publishing plan. For each prompt where a competitor is named and you are not, open the cited sources. If they are roundups and review sites, your job is to be on them. If they are the competitor's own comparison or pricing pages, your job is to publish a better page for that question. The answer engine optimization checklist covers that page.
Where Seen42 fits
Seen42 checks the prompts your buyers ask on ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews, shows who is named and cited instead of you, and builds and hosts comparison, facts and guide pages from your own facts. Paid plans start at $39 a month for 15 prompts; the Free plan checks 10 prompts once on ChatGPT and Gemini. It suggests the first prompts from your own website, your category and the competitors you confirm, grouped by intent, with the branded and control prompts set up for you; they are suggestions you edit, not measured search volume. Every weekly check then records who was named and which sources were cited for each one.
