ChatGPT processes 2.5 billion prompts every day. Visitors arriving from those conversations convert at 15.9%, nearly 9x higher than Google organic. The brands getting cited in those answers aren't just winning visibility. They're winning revenue.
But most AEO advice stops at "optimize your content for AI search" and never answers the harder question: which specific prompts should your brand be targeting, and how do you actually choose them?
That's what this guide covers. By the end, you'll have a systematic method for discovering the right prompts and a clear approach for prioritizing them.
Why Prompts Are Fundamentally Different from Keywords
A keyword is static and universal. Type "project management software" into Google and everyone gets roughly the same results. AI engines don't work that way.
Modern AI assistants like ChatGPT and Perplexity have memory. They know what industry you work in, what tools you've mentioned before, what problems you've been trying to solve. Two people asking the exact same question can receive completely different answers, because the engine is personalizing its response based on everything it knows about each person. A founder at a 10-person startup and a VP of Engineering at a 2,000-person enterprise might ask "what's the best project management tool?" and get entirely different recommendations, tailored to their context.
This makes prompt targeting a moving target in a way keywords never were. You're not just optimizing for a query, you're optimizing for a query as filtered through a user's identity, history, and context. The implication: cast a wide net across the prompts relevant to your audience, because the same underlying question can surface in dozens of personalized variations.
Understanding Query Fanout
Before you can select the right prompts, it helps to understand what actually happens when someone submits one.
When a user enters a prompt into an AI search assistant, the engine rarely treats it as a single query. Instead, it performs query fanout: it breaks the original prompt into multiple sub-queries and runs searches for each in parallel. The results from those sub-queries are what get synthesized into the final answer, and those individual sub-queries look a lot more like traditional search queries than the original prompt did.
For example, a user asking "What's the best way for a SaaS company to reduce churn in the first 90 days?" might fan out into sub-queries like "SaaS churn benchmarks," "onboarding best practices for SaaS," "customer success in first 90 days," and "reducing early churn strategies." Each of those sub-queries surfaces content independently, meaning your content needs to be relevant and well-structured at the sub-query level, not just the top-level prompt.
This is one of the most actionable insights in AEO: the prompts users type are the entry point, but the fanout queries are where ranking actually happens. When you're building content to target a high-priority prompt, think about the sub-queries that prompt is likely to generate, and make sure your content answers each of them clearly. Lumen helps you discover query fanouts for your target prompts, surfacing the sub-queries AI engines are running in parallel.
Six Methods for Discovering the Right Prompts
Cast a wide net first, aim for 100-200 candidate prompts before prioritizing. These methods work at different funnel stages and produce different types of prompt intelligence. Use them in combination.
1. Top-down product, persona, use-case, and funnel-stage mapping
Start here if you're launching a new AEO program without historical data. Build a four-axis matrix:
- Products/services × Target personas × Use cases and pain points × Funnel stage
For each intersection, generate 3–5 prompts that the persona would ask an AI engine at that stage of their journey. A founder at awareness stage asks very different questions than a procurement manager at decision stage, even about the same product. Mapping by funnel stage ensures your prompt set covers the full arc:
- Awareness: "What is X? Why do companies use X? What are the main approaches to solving Y?"
- Consideration: "Best X for [specific situation]? X vs. Y for [use case]? How does X work in practice?"
- Decision: "Is [Brand] worth it? How does [Brand] compare to [Competitor]? What do [Brand] customers say?"
This matrix is your starting point, not your endpoint. As behavioral data accumulates from Methods 2–4, it refines and replaces the top-down assumptions. But for a clean-sheet start, this ensures you don't inadvertently ignore entire segments of your audience.
2. Community forums
Reddit and Quora contain the raw, unfiltered questions real people ask before making decisions, phrased in exactly the natural language that shows up in AI prompts. Search your core topics on Reddit and read how people frame their questions in threads. The phrasing they use, the specific constraints they mention, the comparisons they ask for, these are your prompts.
Community forums are often a better signal than keyword tools for AEO specifically, because the questions aren't compressed into search-bar shorthand. They read the way people actually talk, which is the same way they talk to AI engines.
3. Sales call transcript analysis
Your sales team hears the same questions on repeat, "how does this integrate with HubSpot?", "how does your pricing compare to [Competitor]?", "what does implementation look like?" Those aren't just sales objections. They're prompts. The same questions prospects ask your reps are almost certainly being asked to ChatGPT by buyers who haven't contacted you yet.
To mine them manually: pull your last 20 call recordings, listen for recurring questions, and cluster them by theme. Pay attention to how prospects phrase the question, not how you'd phrase the answer. The phrasing matters for AEO.
4. Google Search Console
Your GSC performance report contains thousands of questions users are already asking Google, many structurally identical to the sub-queries generated by AI fanout. The challenge is finding them buried deep in the query report.
Filter for natural-language patterns: questions starting with "how," "what," "which," "best," and "should I." These map directly to the prompt structures AI engines decompose into.
5. Internal team interviews
Customer success, sales, and support teams answer the same questions every week. Those questions are prompts.
Schedule structured prompt-mining sessions: ask CS managers to pull the top 10 support ticket questions from the past 90 days; ask sales reps to list their most common discovery call objections; ask product teams to document the use cases customers request most. The prompts that emerge are expressed in your customers' actual language, covering edge cases and nuances that generic content never addresses, and niche prompts are often the easiest to win in AI search.
6. Traditional SEO keyword data
Long-tail, question-based keywords are a useful starting point, especially if you already have a mature SEO program with accumulated data. Filter your keyword reports for queries containing "how," "what," "which," "best X for," and "difference between." These question-based structures map closely to how users prompt AI engines.
The gap to keep in mind: keyword volume reflects how often people search Google, not how often they ask ChatGPT or Perplexity. A query with 50 monthly searches may generate far more AI prompts than its volume suggests, and vice versa. Use keyword data as a discovery signal to surface topics and question patterns, not as a proxy for AI prompt frequency or prioritization.
Maintaining Visibility: AEO Is a Program, Not a Campaign
AI citation visibility isn't something you set and forget. The brands cited in response to a given prompt this month may not be the brands cited next month, AI engines continuously reweight sources as new content is published, models are updated, and user behavior shifts. You need to treat your prompt strategy as a living program.
The biggest driver of visibility loss is content staleness. Every time a competitor publishes a fresher, better-structured piece on a prompt you're targeting, your citation probability decreases. Run a 60–90 day refresh cycle on every page targeting high-value prompts.
Lumen's dashboard tracks citation performance across all major answer engines automatically, showing you which prompts you're winning, which you're losing, and what's changed between cycles, so you know exactly where to focus each refresh.
Conclusion
Prompt discovery isn't a one-time exercise. The questions your buyers ask AI engines shift as your market evolves, competitors publish new content, and models update. The brands that win in AI search treat prompt selection as an ongoing program, not a launch checklist.
Start broad. Use the six methods above to build a wide candidate list, then prioritize the prompts where you have the strongest credibility and the clearest content opportunity. Build depth before breadth, one genuinely great piece will outperform ten surface-level ones. Revisit your prompt list every quarter, and refresh your highest-value content every 60–90 days.
Book a call with us to learn more about how Lumen can help you win AI search.