Start with the sentence you want an engine to say
Before tracking anything, write down the answer you would want. Not a tagline: the plain description an assistant would give to a buyer who described their situation. If you cannot write it, no engine will write it for you.
Every part of that sentence becomes something to test. The category is a word buyers use. The audience and the situation are what a question will contain. The difference is what you need sources to repeat.
Say who you are not for
Engines recommend by fit. A brand that describes itself as right for everyone gives an engine nothing to match, and ends up an acceptable answer to many questions and the first answer to none.
Naming the buyers you are not built for sharpens the ones you are. It also makes the description believable, which matters when a model is weighing your page against a review that says the opposite.
Use what customers already say
The words that belong in your claims are mostly already written down, by customers, in places your company already has access to.
- Sales call notes and recordings: the problem the buyer described before they had heard of you.
- Support tickets and onboarding questions: what people expected, and where that was wrong.
- Win and loss notes: the reasons a deal went your way, and the reasons it did not.
- Review sites and community threads: how customers describe you to each other.
- Competitors' pages: the claims your category already takes for granted.
Where you sit changes the mix. A business selling to other businesses usually has more in its own call notes and tickets. A consumer brand usually finds more in public reviews and communities.
From themes to a list of claims
Read across the sources for themes that repeat. Trust the ones that turn up in more than one place: a phrase that appears in call notes, reviews and tickets is how buyers actually think about you.
Then write each theme as a claim, grouped by the kind of question it answers. The example below is for a fictional scheduling tool for small clinics.
| Kind of claim | What it answers | Example |
|---|---|---|
| Category | What are you? | Appointment scheduling software |
| Audience | Who is it for? | Independent clinics with one to ten practitioners |
| Situation | When is it the right choice? | Replacing phone booking and a paper diary |
| Problem | What does it fix? | Patients who miss appointments |
| Integration | What does it work with? | The calendar and payment tools a clinic already uses |
| Constraint | What does it rule out? | Not built for hospital groups with central scheduling |
Choose where to start
A full list is too much to work on at once. Start where three things overlap:
- The claim decides purchases: buyers use it to choose, not just to describe.
- The gap is large: what engines say today is furthest from what you wrote.
- You can back it: there is evidence, on your site or elsewhere, that makes it true.
What the list is for
Everything after this chapter traces back to a claim. Each tracked question tests one (chapter 3). Each finding in a baseline is about one (chapter 4). Each page brief and each third-party request exists to strengthen one (chapter 5). If a piece of work cannot name the claim it serves, it is not AEO work.
Key takeaways
- Write the plain description you want engines to give, including who you are not for.
- Build claims from what customers already say, and trust themes that repeat across sources.
- Group claims by the kind of question they answer.
- Start with claims that decide purchases, have the biggest gap, and can be backed with evidence.