Skill
AI search, AEO and GEO
More people now get an answer without visiting a site. That does not make the site pointless; it changes what the site is for. It is the thing the answer is built from. Google says there is nothing extra to do to appear in AI Overviews and AI Mode beyond being indexed and eligible for a snippet, and since May 2026 its spam policies cover trying to game the answers. So the work is not a new discipline with a new acronym. It is measuring something rankings never measured, and writing pages so the first 150 words can stand alone as the answer.
What this covers
| Area | What goes wrong | What I do about it |
|---|---|---|
| Brand visibility in assistants | Nobody knows whether ChatGPT, Gemini, Perplexity or AI Mode name you, or what they say | A fixed prompt set from real buying questions, run on a schedule, recording who is named and which URL is cited |
| Answer-first pages | The page ranks and never gets quoted, because the answer arrives in paragraph four | Rewrite the top of the page so the claim, the number and the definition each stand alone, then the context |
| Generative AI report | AI impressions are invisible inside the Web total | Separate them, find pages with AI impressions and falling clicks, and start there |
| Structured data and entity clarity | Machines cannot reliably parse what the page asserts or who is behind it | Organization, Product and Article schema that matches the visible page, plus FAQ only where the questions are real |
| Crawler and snippet policy | AI crawlers and snippet directives were set by accident | A written decision per crawler and per directive, matched to whether you want the exposure |
| Comparison content | Absent from best of and versus answers entirely, because the assistants draw those from third-party pages | Build the comparison pages with defensible criteria, and earn mentions on the third-party ones |
| Measurement | AI referrals land in Direct and the prompt results live in someone’s memory | A GA4 channel for AI referrers and a prompt log that makes the next run comparable to the last |
Signs you need this
- You have never read what an assistant says when asked about your category.
- Competitors get named in AI answers and you do not, even where you outrank them.
- Impressions are up in Search Console’s generative AI report and clicks on the same pages are down.
- Your content answers the question, but only after several paragraphs of setup.
- Structured data was added once, by a plugin, and never checked against the page.
- Nobody has decided which AI crawlers can access the site.
A visibility baseline: a fixed prompt set run across the major assistants and AI Mode, plus the generative AI report and AI referrals pulled into one sheet. You see how the brand is described today, which pages are being quoted, and where competitors are named instead. Repeatable, so the next run is comparable.
How I work on it
- Baseline first. Run the prompt set, record what gets said, who gets named and which URL is cited.
- Fix what is parseable. Answer-first page tops, schema that matches the page, claims that stand alone.
- Open the right doors. A crawler and snippet policy set on purpose rather than inherited.
- Re-run and compare. Same prompts, same method, plus the generative AI report, so the change is a number and not a feeling.
Related projects
WorkWise Compliance
AI searchCompliance catalogue rebuilt so the pages answering the question were the ones being cited.
Read the project →Marvel Refrigeration
IndexationPre-migration audit that sorted every URL by evidence before a redirect map was written.
Read the project →Common questions
Is GEO different from SEO?
Google’s answer is no: no additional requirements, no special optimisations, the same crawl and index rules, and the same spam policies, which since 15 May 2026 explicitly cover attempts to manipulate generative AI responses. The honest practitioner answer is “mostly no, with different measurement”. You cannot rank-track an answer, so you build a prompt set. You cannot see AI clicks yet, so you watch the generative AI report’s impressions and the AI referrer channel. The page work is the same discipline written more directly.
Can it actually be measured?
Partly, and better every quarter. Search Console’s generative AI performance report gives impressions for AI Overviews and AI Mode, no clicks yet. A fixed prompt set gives share of voice across ChatGPT, Gemini and Perplexity. GA4 can count AI referrals once you segment them. None of it is as clean as a rank tracker and all of it is more than most sites have.
Should we block AI crawlers?
It is a business decision, not an SEO one, and I will help you make it deliberately. Blocking Google-Extended does not remove you from AI Overviews or AI Mode; those use the normal index. Blocking GPTBot removes you from ChatGPT’s browsing and training. If you sell to people who ask assistants what to buy, blocking is usually the wrong call. Either way, write it down.
Want to talk through a site?
Twenty minutes, screen shared, your Search Console open. You will leave with at least one thing worth fixing whether or not we end up working together.
Typical reply time: within one working day