ai search optimisation — geo, aeo, and what's underneath
AI search optimisation as engineering.
Call it GEO, AEO or AI SEO — the job underneath is the same. Answer engines quote the sites they can fetch, parse and trust. I make a site legible to them: open to the right crawlers, extractable without a headless browser, described by a structured graph that agrees with the prose, and measured by instruments rather than vibes. This site is the reference build: it grades itself at 100/A on the model it uses for clients.
in practice since 2007 · uk & us · one operator · no forms, no funnels
who this is for — three situations
01
Competitors are cited in ChatGPT and you aren't.
Citation is mostly plumbing. If OAI-SearchBot can't fetch the page, or the answer sits inside a client-rendered component, the model never sees it. Access and extractability come first; the rest follows.
02
You bought an AI-visibility dashboard and can't act on it.
Prompt tracking measures a string nobody types twice. I instrument what you control — crawler hits by product, citation referrers, structured-data validity, the grader's gates — and put the numbers where decisions get made.
03
Your schema plugin says everything is fine.
Plugins emit islands. Machines want a graph: Organization, Person, Product and Article linked by @id and agreeing with the page. Three small businesses on three platforms went to zero validator errors on exactly that work.
how the work runs — four stations
01
Gates
Crawler access for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended, each decided on purpose. Extractability tested against the raw HTML. Snippet eligibility checked.
02
Graph
Structured data rebuilt as one connected entity graph, validated, and reconciled with the visible copy so the site cannot contradict itself across pages.
03
Surfaces
llms.txt, markdown mirrors, clean feeds and — where it earns its place — an agent endpoint. The machine readers get a front door instead of a side window.
04
Measure
Bot hits by product, citation referrers, grader scores over time. The machine guestbook on this site's lab page shows the method running live.
what lands — the receipt
- 01Agent-Ready Grader report with the three gates explained
- 02Crawler access policy, decided bot by bot and implemented
- 03Entity graph rebuilt and validated — zero errors at sign-off
- 04llms.txt and machine-readable mirrors published
- 05Measurement plan: bot logs, citation referrers, grade over time
- 06A plain-words brief your marketing team can act on
signed — one operator, no handoffs
06 items
proof — on the record
sealed file · Leeds dental practice
74 → 99
grade C → A
source — Agent-Ready Grader, this site
sealed file · Three small businesses, three platforms
17 / 17
pages validated on one site — no parse errors, no dangling references
source — Rich Results Test + schema.org validator
sealed file · UK data-cabling contractor
10
industry leaves under the rebuilt wireless hub
source — site structure
built · Agent-Ready Grader
7 + 3
weighted categories plus gates
source — scoring model
asked straight — answered straight
01What is generative engine optimisation?
Making your site the source an answer engine chooses to cite. In practice it is technical: can the engine's crawler reach the page, can it extract the answer from the HTML, does the structured data describe the entity clearly, and does the site agree with itself across pages. The copy matters — after those.
02Is GEO different from SEO?
The readers multiplied; the discipline didn't split. The engines fetch and parse the same HTML Google does. What changes is the weighting: extractability, entity clarity and freshness count for more, and the measurement layer is weaker, so you need your own instruments.
03Can you guarantee citations in ChatGPT or AI Overviews?
No, and be wary of anyone who does. What I can guarantee is the inputs: access, extractability, a valid graph, all measured over time. Citation follows from being the best available source, which is the part we build together.
04Should I block AI crawlers from training on my content?
That's a business decision, and the grader never penalises a training-bot block. The mistake is blocking the search crawlers — OAI-SearchBot, PerplexityBot — by accident while trying to block the training ones. Decide bot by bot, on purpose.
05How do I know if my site is readable by AI search engines today?
Run it through the Agent-Ready Grader on this site. It's free and ungated and gives the full report: gates, structured data, metadata, Core Web Vitals and the frontier bonus for llms.txt.
start — the brief
A few lines about the site, the problem and the evidence you have.
The reply is a straight read on fit and shape, from the person who does the work. If what you need is small, it says so. If it is not a fit, it points somewhere better.