Every prompt someone types into an LLM like ChatGPT has essentially a search volume of one. No two buyers word it the same way, so you can never be fully sure whether you show up for a specific prompt. For most of a 17-month study, Semrush found that 65% to 85% of ChatGPT prompts didn't match any keyword in its 27-billion keyword database.
The prompt also isn't the whole input. ChatGPT and Claude both remember context from past chats. For example, when a founder asks for a software recommendation, the model probably already knows their team size, their tools, and what they're trying to fix. It maps where they are now to where they want to be.
So tracking "What's the best project management tool?" tells you very little about which prompts to track. This guide shows you how to choose prompts to track for AEO from what your buyers say on sales calls. Then you track several per page to get a real read on your visibility.
Step 1: List your category and competitor-alternative keywords
Start with the keywords that define your market. They tell you which topics your prompt set needs to cover.
Pull together four groups:
- Category terms: What your product is, such as "project management software" or "B2B SEO agency."
- Use-case terms: The jobs people hire you for, such as "client project tracking for agencies."
- Competitor alternatives: "[Competitor] alternatives" for each competitor you lose deals to.
- Comparisons: "[You] vs [competitor]" and "[competitor] vs [competitor]."
If you already have an SEO keyword list, most of this exists. Keep it to the topics you sell into, and drop anything informational that a buyer wouldn't search close to a decision.
Step 2: Connect your sales call recorder to your LLM
Your sales calls hold the language buyers use when they describe their problem to an AI tool. Francesco Montesanto of Optimizely describes it well. "The questions a prospect asks in the first fifteen minutes of discovery are almost exactly what they were typing into ChatGPT the night before."
Most B2B companies already record their sales calls. Connect your call recorder to the LLM you already use through MCP (Model Context Protocol). MCP lets Claude, ChatGPT, and other AI tools read data from other apps, so the model can search your transcripts directly.
Test the connection with a simple request, such as "Summarize my most recent sales call." If the model returns the right call, you're ready for prompt research.
Step 3: Extract buyer language from your calls
This is where prompt research for AEO actually starts. You pull buyers' exact words from dozens of calls at once.
Pick the right calls first. Use discovery calls with prospects, and skip check-ins with existing customers.
Then run this prompt in your LLM. Change the time frame and the recorder name to match your setup.
You have access to my sales call transcripts through [Grain / Fathom / Fireflies].
Find discovery calls with prospects from the last 6 months. Skip calls with existing customers.
For each call, pull out:
1. Problem: how the buyer describes their problem, in their exact words.
2. Trigger: what made them start looking now (for example, "we outgrew X," a new hire, or a tool that failed).
3. Current setup: the tools, competitors, or workarounds they use today.
4. Category language: what they call the type of product they're looking for.
5. Constraints: team size, budget, required integrations, timeline, or anything else that rules options out.
6. Desired outcome: what they want to be able to do once the problem is solved.
7. Questions: what they asked us about fit, pricing, or alternatives.
Then give me:
* A table with one row per call and one column for each item above.
* The 10 phrases buyers repeat most often, with the number of calls each appears in.
* Every competitor mentioned, ranked by how many calls it came up in.
Use the buyers' exact words. Don't summarize or tidy up their language. If something wasn't discussed on a call, write "not mentioned."Read the output for patterns. A trigger that shows up in 8 of 30 calls is a prompt worth tracking. A phrase one buyer used once probably isn't.
The prompt above gives you a broad view of how buyers talk. You can also run it for one topic at a time. Pick a category, product, or feature you want to be recommended for, and pull the buyer language for that topic alone. This feeds the topic-level and page-level prompts you'll write in Steps 4 and 5.
Using my sales call transcripts, find every discovery call where prospects discussed [topic, for example "client reporting" or "HubSpot integration"].
For this topic only, pull out:
1. How buyers describe the problem, in their exact words.
2. What they used before and why it stopped working.
3. The constraints and requirements they mentioned.
4. The questions they asked about it.
List each quote with the call it came from. Use the buyers' exact words.Step 4: Write prompts that carry your buyer's context
Combine your Step 1 topics with the language from Step 3. A short keyword becomes a prompt that sounds like a real buyer describing a real situation.
A simple pattern works well. Start with the situation, add a constraint and the pain, then make the ask. The two prompts below cover the same topic.
- Keyword-style prompt: "What's the best project management tool?"
- Buyer-style prompt: "We're a team of 25 and we've outgrown Trello. Our integrations keep breaking. What should we switch to?"
The second prompt reflects how buyers actually use AI tools. They add context, and memory adds more. The answer to the second prompt can name different tools than the answer to the first.
Play to your strengths here. Write prompts around the situations where your product is the best answer, using the triggers and constraints that came up most on your calls. Then cut any prompt where being recommended wouldn't help you sell.
If a prompt doesn't tie back to what you offer, what you want to be known for, or a problem you solve, it doesn't belong in your set.
Step 5: Build a prompt set for each page, from different angles
Track 5 to 8 prompts for each page. Because every prompt is a search volume of one, a single prompt can't tell you how visible a page is. Several angles on the same topic give you a far more reliable picture.
For each page, cover a mix of these angles:
- Head term: The short version, such as "best project management tool for a 25-person team."
- Buyer situation: The context-rich version from Step 4.
- Comparison: "[Competitor A] vs [Competitor B] for a team that lives in Slack."
- Alternative: "Good Trello alternatives for a growing team."
- Integration or constraint: "Project management tools that integrate with HubSpot and Slack natively."
- Outcome: "How can I see every client project's status without weekly check-in meetings?"
Each angle should map to one of your offerings and to something you keep hearing on sales calls. That's how you select prompts to track without drifting into topics that never lead to a deal.
Start with pages you've already published, especially your bottom-of-funnel content, where buyers are closest to a decision. Tracking prompts for topics you haven't written about tells you little about your own visibility. They're still useful in the next step, though.
Step 6: Check which sources get cited for each prompt
Run your prompts in an AI visibility tracking tool and look at what each answer cites. This tells you where LLMs get their information for your topics, and where you need to show up.
Most tools now show this per prompt. Promptwatch shows every URL cited for each tracked prompt, and Ahrefs Brand Radar shows the citations behind each prompt's answer. Peec breaks cited sources down at the domain or URL level.
Run this check for topics you haven't published yet too. The cited sources show you what it takes to compete before you write anything.
Then act on what you find:
- Third-party listicles: These are what LLMs cite most often for B2B buyer prompts, in our experience. Publish a highly specific page on your own site that answers the prompt, then work on third-party listicle placements on the exact pages being cited. Our guide on how to rank in ChatGPT search covers the on-site side.
- LinkedIn posts: Start publishing from your founder's or executives' accounts on the topics behind those prompts.
- YouTube videos: Make video content that answers the prompt directly.
Step 7: Track recommendations and review your set every month or two
Measure whether the LLM recommends you when your dream client describes their problem. A citation means a model used your page as a source. A recommendation means it named you as the answer, and that's the outcome that turns into pipeline.
Keep your prompt wording stable once it's set. Changing prompts every month breaks trend comparisons, a point Overthink Group makes about building the set before you buy a tool. SE Ranking suggests reviewing your prompts every 30 to 60 days.
Use each review to add prompts and leave existing wording as it is. When new triggers, competitors, or constraints start showing up on sales calls, run the Step 3 prompt again and add prompts for them. Retire prompts only when you stop selling into that situation.
We run AI visibility tracking this way for our clients, with prompt sets built from their sales calls and tied to the pages that drive pipeline. Book a strategy call to see which buyer prompts recommend your competitors instead of you, and what it would take to change that.
FAQs about choosing prompts to track for AEO
These are the questions teams ask most when they work out how to choose prompts to track for AEO.
What prompts should I track?
Track prompts written in your buyers' own words, tied to pages you've published and products you sell. The best source is your sales calls, because they show how buyers describe their problem, their constraints, and the competitors they're comparing.
How many prompts should I track for AEO?
Track 5 to 8 prompts per page. That's enough angles to see a page's real visibility without diluting your budget across topics that don't sell. Your total depends on how many priority pages you have and on your tool's prompt limit.
Is prompt research for GEO different from AEO?
No, the process is the same. AEO and GEO both describe getting your brand into AI-generated answers, so the same buyer-language prompts work for both. How to research prompts for GEO comes down to the steps above.
Should I track branded or non-branded prompts?
Track mostly non-branded prompts, because those show whether AI tools recommend you to buyers who don't know you yet. Conductor recommends keeping branded prompts to 25% or less of your total. Use the branded ones to check that AI tools describe your product accurately.
