Three rules the method follows.
Prompts before keywords
AI engines answer questions, so the unit of measurement is the buyer prompt, not the keyword.
Every finding is reproducible
Each result is stored with the prompt, the engine, and the date, so anyone on your team can re-run it.
Engines are tested separately
Each engine draws on different sources, so a change that moves one engine may not move another.
Six steps, in this order.
Build the prompt set
10–20 prompts per client, written the way buyers ask, and agreed with your team before testing.
- Service prompts: “emergency plumber near me”
- Comparison prompts: “best hot water installer in Parramatta”
- Problem prompts: “why is my drain gurgling”
Run and record
Each prompt is run on ChatGPT, Gemini, Perplexity, and Google AI Overviews. The result is recorded as named, cited, or absent, along with the sources the engine cited and the date.
Diagnose the gaps
For every prompt where the client is absent, the cause is traced to one of five areas: page answers, structured data, entity consistency, third-party citation sources, or reviews and profile signals.
Rank the actions
Each fix is scored on expected impact and effort, then split into a 30 day list and a 90 day list.
Brief the work
The highest-impact actions get page-level briefs your writers and developers can execute without a strategy call.
Re-test monthly
The same prompt set is re-run each month. AI answers vary between runs, so the trend across the full set is reported, not a single answer.
Sources for every report.
| Source | What it shows | How often |
|---|---|---|
| Prompt set results | Named, cited, or absent, by engine | Monthly |
| Cited sources | Which sites the engines rely on for the category | Monthly |
| Search Console AI reports | Impressions in AI Overviews and AI Mode | Monthly |
| GA4 AI assistant traffic | Visits referred by AI assistants | Monthly |
Lines that are not crossed.
No invented proof
No fabricated reviews, statistics, or testimonials on client sites. If a fact cannot be verified, it is left out.
No penalty-risk tactics
No private blog networks, link schemes, or doorway pages, even where competitors use them.
No placement guarantees
AI answers change between runs, so no one can guarantee a placement. Progress is reported as a trend across the prompt set.
See it on one of your accounts.
The fastest way to judge the method is to see it run on a client you know well. The audit applies all six steps to one site.
See the white-label AEO audit →