AI prompt research
People don't type keywords into AI assistants. They ask questions. This tool writes the questions your potential customers are asking ChatGPT, Claude, and Perplexity right now, from a plain description of what you offer.
Use them to check the answers yourself, to plan the pages that should answer them, or as the tracked questions in an Everfound audit.
What this tool does
Keyword research assumes someone types three words into a search box and picks from ten blue links. That is not how an assistant is used. People describe a situation and ask for a recommendation, in whole sentences, and the assistant answers with a handful of named companies. The unit of demand is a question, not a keyword, so the first thing you need is the list of questions.
Give this tool a plain description of what you sell, or paste the keywords you already rank for, and it returns the questions a buyer in your category would actually put to an assistant. They come grouped by intent: questions that name your brand, questions about the category as a whole, questions describing the problem you solve, and questions about a specific use case. That grouping is the useful part, because the four behave very differently.
How to read the output
A question that names your company is the easy one. An assistant asked “what is Acme?” has been handed the answer, so being mentioned proves very little. The questions that matter are the open ones, where nobody said your name and the assistant picks whoever it knows. A company can be named in every branded question and appear in none of the open ones, and that gap is the whole problem worth measuring.
So take the open questions first. Paste one into ChatGPT and Grok and read the answer. If you are named, note which sources it cited, because those are the pages doing the work. If you are not, note who was: those companies are your real competitors in the only ranking that now matters, whatever the search results say.
Who this is for
Founders and marketers who suspect buyers are asking an assistant before they ever reach a search engine, and agencies who need to show a client what those answers currently say. It is free, it needs no account, and nothing is stored against you. If you only ever use this page and check the answers by hand, that is a legitimate way to use it.
What comes after
Checking by hand stops scaling at about a dozen questions, and it tells you where you stand today without telling you why. A full Everfound audit asks your questions across 6 assistants, records every answer verbatim with the pages each one cited, crawls your site to score what an AI reader can and cannot use, and generates the schema, answer box, FAQ and llms.txt to fix what is missing. A second run then measures which checks moved.
You can see the whole thing before paying: the sample report is a real audit in full, the method is documented step by step, and the comparison page covers the other tools in this category, including where they beat us.