Get Started
Choose the win · Lesson 03

Build the 20-prompt baseline

Choose buyer questions, run the same test across engines, and record cited, mentioned or absent without cherry-picking.

22 minWorked exampleField assignment
0%course
What you will leave with

A dated 20-prompt baseline with at least three named competitors and no hand-picked omissions.

The first AEO deliverable is not an optimized page. It is a controlled test. If you choose prompts after seeing the results, change engines between runs, or count a brand mention as a citation, every later “improvement” becomes impossible to trust.

The exact prompt framework

Start with one commercial topic and write five questions in each of four buckets. This catches buyers who describe the problem differently from your marketing team.

Problem aware

How do agencies prove AI search is driving leads?

Use when the buyer knows the pain, not the category.
Category

Best AI visibility analytics for WordPress

Use for direct category competition.
Alternative

Profound alternative for a small agency

Use when a known competitor anchors the search.
Risk / proof

Does AI visibility software attribute revenue?

Use for objections that decide the purchase.

Worked example: from vague keyword to trackable questions

Too broadAI analytics

No persona, use case, constraint or decision. Results will be generic.

Track thisWhat is the best AI analytics plugin for a WordPress agency managing 20 client sites?

Category + platform + persona + operational constraint.

If you cannot think of 20 prompts

Have sales calls?Copy the exact questions prospects asked before buying.
Have Search Console data?Turn comparison, “best,” “how,” and “alternative” queries into natural questions.
Starting from zero?Use Query Fan-Out, then delete questions that would not change a buying decision.

Score without fooling yourself

Share of voiceYour cited slots ÷ all cited slots earned by you and tracked competitors × 100Example: 7 ÷ (7 + 9 + 3 + 1) = 35%

Example baseline by engine

ChatGPT25%
Perplexity40%
Gemini15%
Claude30%

Questions learners usually have

What if the answer changes every time?

That is normal. Run the same set on a schedule and read the trend. Never claim a win from one favorable answer.

What if my brand is named but not linked?

Record “mentioned.” You earned awareness, but the engine trusted another source enough to attach the citation.

What if no brands are cited?

Keep the result. It tells you the prompt may produce an informational answer rather than a commercial shortlist. Do not replace it only because the result is inconvenient.

Do the work

Field assignment

Scenario

You sell project-management software. Your team wants to track “project management,” but buyers actually ask different questions at every buying stage.

  1. Write five problem-aware, five comparison, five alternative and five purchase-risk prompts.

  2. Run all 20 against the same engines on the same day.

  3. Record your status, winning domains, position and answer framing.

Definition of done

A dated 20-prompt baseline with at least three named competitors and no hand-picked omissions.