Guide
A short guide to prompt research
What is prompt research?
Prompt research is the practice of finding out which questions people ask AI assistants about a category, a problem or a brand, and how they phrase them. The output is a list of realistic prompts, grouped by topic, audience and buying stage, that a team can then track in ChatGPT, Gemini, Perplexity, Claude, Copilot and Google's AI surfaces to see whether its brand is named, recommended or cited.
It matters because an AI answer is not a list of ten links. A buyer who asks an assistant for a shortlist usually gets a handful of names and a reason for each. If you do not know which questions lead to those shortlists, you cannot measure whether you are on them, and you cannot write the content that would put you there.
Think of it as keyword research for conversations: same goal, different unit of analysis.
How is prompt research different from keyword research?
Keywords are compressed. People type two to four words into a search box and refine by clicking. Prompts are expanded: they carry context such as company size, budget, existing tools, location or a deadline, and they are often followed by a second and third message that narrows the answer. Two people with the same keyword intent can send prompts that produce entirely different shortlists.
That changes the unit of research. Instead of ranking terms by monthly searches, prompt research maps intents and the modifiers around them: the persona asking, the job to be done, the comparison being made and the stage of the decision. Keyword data still helps, but as evidence that a topic has demand, not as a count of prompts.
A keyword says what. A prompt usually says what, for whom, under which constraint, compared with what.
Where does prompt evidence come from?
Because AI platforms do not publish per-prompt query reports the way search engines publish keyword planners, every prompt research method relies on proxies. The useful ones each reveal a different part of the picture, and each has a blind spot worth knowing before you trust it.
| Signal | What it tells you | Watch out for |
|---|---|---|
| Keyword data | Which topics have demand in classic search | Search volume is not prompt volume; use it as a topic signal only |
| People Also Ask & related questions | How people phrase questions in full sentences | Phrased for search results pages, often shorter than real prompts |
| Search Console long queries | Your own audience's conversational, multi-word questions | Only covers queries where you already appear |
| Sales calls, support tickets, reviews | The constraints, comparisons and objections buyers actually raise | Small samples; needs cleaning and grouping |
| Forums and communities | Unprompted wording, problems and recommendations | Skews toward enthusiasts and edge cases |
| Fan-out queries from AI engines | The web searches an engine runs while answering a prompt | Only visible where an engine exposes its searches |
No single source is enough. Good research triangulates at least three of these.
Can you trust a prompt volume number?
Some tools show a volume or a demand score next to each prompt. Those figures are estimates. They are typically modelled from a panel of users, from clickstream data, or derived from keyword volumes, and the method determines how much weight they can carry. A figure derived from search volume describes interest in a topic; it does not tell you how many people typed a given sentence into an assistant.
Treat volume figures as a ranking aid within one tool, not as an absolute count you can compare across vendors or report as audience size. Our notebook entries on data and methodology go through the main estimation methods and the questions to ask before relying on any of them.
Ask every vendor one question: what exactly was counted, and from which sample?
What does a good prompt set look like?
A prompt set is the list of prompts you track over time. A useful one is balanced rather than large. It covers the main question types for your category and is reviewed regularly as products, competitors and language change.
- Category prompts, where a buyer asks for the best options without naming a brand.
- Problem prompts, where the buyer describes a situation and asks how to solve it.
- Comparison prompts, which put two or more named options side by side.
- Branded prompts, which ask about your company directly: pricing, reviews, alternatives.
- Persona and market variants, so a prompt is asked the way different audiences and countries would ask it.
Rule of thumb: every prompt should be something a real buyer could plausibly send, word for word.
How should you choose a prompt research tool?
Start with transparency. A tool should state where its prompt suggestions come from, how any volume figure is produced and what it cannot see. Next, check fit: whether suggestions reflect your market, language and personas, whether near-duplicates are filtered, and whether you can scope research to your own keyword lists or Search Console data. Finally, look at what happens after research. Prompts are only valuable if they flow into tracking, so check how easily a researched set becomes a monitored one, and how often the tool refreshes it.
Our comparison of prompt research tools applies these criteria to each vendor, with prices checked on the date shown, and our methodology page lists every source we use.
Run a pilot on your own category before you compare dashboards. The data source matters more than the interface.
