How to find the exact prompts your buyers ask AI
You cannot get recommended for a question you have never seen. The first real step in any GEO effort is not writing content. It is building the honest list of prompts your buyers use to decide, and finding out who the engines already trust to answer them.
Start from the decision, not the keyword
A buyer does not ask an engine about your category in the abstract. They ask it to choose. The prompts that matter are the ones attached to a decision with money behind it: which option is best for their situation, their city, their specific goal. Write those down first, in the buyer's words, not your product's.
A quick way to generate them is to paste your main pages into an assistant and ask it to produce fifty bottom-of-funnel queries a buyer might use before choosing in your category. Then cut the list down to the ones that carry real intent.
Mine the engines themselves
The engines will tell you what they trust if you look. Ask one of your decision prompts, then read the sources it cites at the bottom of the answer. Those pages are the current shortlist. Some you will want to be listed on. Others you will want to outrank with better content of your own.
On engines that show their reasoning, open it. You can watch which pages the model consults before it answers, which tells you exactly where you need to appear to be pulled into future responses. This is the closest thing GEO has to a keyword tool, and it is free.
Cluster the prompts by intent
Once you have a working list, group it. Clusters make it obvious where you are strong, where you are absent, and which content to write first.
| Cluster | What the buyer is doing | Priority |
|---|---|---|
| Comparisons | Weighing you against named alternatives | High |
| Best-of / location | Looking for a shortlist in their city or market | High |
| Use-case / credential | Solving one specific outcome or requirement | High |
| Alternatives | Actively shopping away from a competitor | Medium |
| Buying guides | Learning how to choose before choosing | Medium |
Test clean, or the data lies
Personalisation quietly ruins this exercise. If you test on your own logged-in account, the engine leans toward showing you the brands you already know, including your own. Run every prompt in a temporary or incognito session so the result reflects a buyer with no history. Test the same prompts on more than one engine, because they cite different sources and will not agree.
Log it in a spreadsheet. For each prompt, record whether you were named, whether you were cited by link, which competitors appeared, and which sources the engine used. That single sheet is your baseline, your content plan and your outreach list in one place.
What you do with the list
The absent clusters become your content plan. The cited sources you do not control become your outreach list. The prompts where a competitor is named and you are not become your priority. You now know the exact questions to win, instead of guessing at topics and hoping an engine notices.
Frequently asked questions
Why not just use our existing SEO keyword list?
Because AI prompts are longer, more specific and more decision-shaped than search keywords. Someone types French classes into Google, but asks an engine for the best French course in their city for passing a specific exam. Your keyword list captures the topic. It usually misses the phrasing and the intent that actually triggers a recommendation, which is what you need for GEO.
Why test in incognito or temporary mode?
Because the engines personalise answers based on your history. If you have visited your own site fifty times, the model is far more likely to surface you for you, which tells you nothing about what a real prospect sees. Always test in a temporary or incognito session so the result reflects a cold buyer, not your own footprint.
How many prompts do we actually need?
Fewer than most people expect. Twenty to thirty prompts that map to real buying decisions will tell you more than two hundred generic ones. Focus on the comparison, alternatives, local and use-case queries where a shortlist gets formed. Depth on the decisions that carry money beats breadth across questions nobody buys on.
How often should we refresh the prompt set?
Re-run the same prompts monthly, on the same engines, in a clean session. Answers shift as models update and as competitors publish, so a fixed set run on a fixed cadence is the only way to see whether your visibility is actually moving. Change the prompts and you lose the ability to compare month to month.
See how you show up today.
The free Snapshot checks how your brand appears across ChatGPT, Perplexity, Claude and Google AI in your market, then hands you a ranked list of fixes. No card, no commitment.