Your keyword tool says nobody searches for your category. Meanwhile, a buyer just asked ChatGPT for the best vendor in it and got three names. None were yours.
That gap exists because keyword research measures what people type into Google. It says nothing about what they ask an AI. A prompt research tool fills that hole. It shows which questions buyers put to ChatGPT, Perplexity, Gemini and Claude, and which of those questions name your competitors instead of you.
What a prompt research tool does
Think of it as keyword research for AI. You start with a topic, and the tool returns the prompts people use around it. Short ones, long ones, comparisons, "best of" lists, problem statements. Then it runs those prompts against the major engines and records what comes back.
The output is not a ranking. It is a map of answers. Which brands are named, which pages are cited, how the answer is framed. That map is the raw material for LLM brand visibility work.
Why keywords stop working here
Search queries are compressed. "crm small business" is three words. The same buyer asks an AI, "What CRM works for a ten-person agency that already uses HubSpot for email?" Same intent, forty words of context.
Those long prompts are where the shortlist gets built. The model reads every constraint and filters. Brands that fit the constraints get named. Brands that only match the head term do not.
Two more differences matter.
Volume is hidden. No engine publishes how often a prompt is asked.
Answers drift. The same prompt can return different brands next week.
So you cannot pull a list once and call it done. Prompt research has to repeat.
How to run prompt research yourself
Start from buyer language
Pull the phrases from sales calls, support tickets and G2 reviews. Buyers describe problems in their own words, and those words are what they paste into an AI. Write down 20 of them before you open any tool.
Expand into four prompt types
For each topic, draft one prompt in each group.
Category prompts. "What are the best tools for X?"
Comparison prompts. "X vs Y for a mid-size team."
Problem prompts. "How do I stop losing deals to Z?"
Brand prompts. "Is [your brand] any good?"
Fifty to one hundred prompts is enough to see patterns.
Run them across engines
Run every prompt on at least four engines. ChatGPT and Perplexity often disagree. Gemini leans on Google's index. Claude favors cleaner, better-structured sources. A brand that wins on one and vanishes on another has an AI visibility problem you would never see in a single-engine check.
Log three things per prompt
Record whether you were mentioned, whether you were cited, and who else appeared. Those three fields give you a mention rate, a citation rate and a share of voice against competitors.
What to look for in a tool
Skip the demo gloss and test for these.
Multi-engine coverage
Scheduled re-runs
Competitor tracking
Source-level citation data
Prompt grouping by intent
The last one is easy to overlook. A flat list of 300 prompts is noise. Grouped by intent, it tells you which funnel stage you are losing.
Also ask where the prompt list comes from. If the tool only tracks prompts you type in, you are measuring your own assumptions. A good one suggests prompts you did not think of.
Turn the data into work
Sort prompts by one rule. Competitors appear, you do not. Those are your gap prompts.
For each gap, open the sources the engines cite. That list is your to-do list. Get your brand onto those pages, or publish something that answers the prompt better than they do, in the first 150 words, in the buyer's own phrasing.
Then re-run the set in two weeks. If the mention rate moved, keep going. If it did not, change the page, not the prompt.
Where to start today
Write ten prompts your best customer would ask. Run them in ChatGPT, Perplexity and Gemini. Count how many name you.
That number is your baseline, and most teams have never seen it. If you want it tracked automatically across engines every week, LLM Search Console does exactly that.
New prompt research tactics land in the Substack newsletter every week. Subscribe below and the next one arrives in your inbox.