Generative Engine Optimisation

Generative engine optimisation (GEO) is the practice of structuring a website so AI answer engines — ChatGPT, Claude, Perplexity, Google’s AI Overviews — cite it as a source. I audit and rebuild sites for AI visibility: static rendering, answer-first content, entity schema, and citable proof structure.

Why does this matter now?

Your customers changed how they search, and most websites haven’t noticed. They ask ChatGPT “best retail ERP for a multi-store business in Dubai” or “who builds ledger systems for NBFCs” — and they act on the answer. The engines composing those answers never execute JavaScript. GPTBot fetches your JS bundle and doesn’t run it; ClaudeBot downloads it and doesn’t execute; PerplexityBot parses static HTML only. If your content mounts client-side, you don’t exist in the conversation where the decision now happens.

No faff about it: that’s the whole game. A decade of content marketing, invisible to the surface your buyers moved to — because of a rendering decision someone made in a framework tutorial.

And the shift is structural, not a fad. Search behaviour that moves to answer engines doesn’t move back; the question is only whether you’re in the answers or absent from them. Every month a competitor is citable and you aren’t is a month of buyers who never knew you were an option.

What does the service include?

• Rendering audit. What AI crawlers actually receive from your site, page by page — fetched the way GPTBot fetches, no JavaScript execution, no charity. The answer is usually “less than you think”, and the audit shows you exactly which pages are ghosts.
• Answer-first restructuring. Every page opens with the sixty-word definitive answer an engine can lift verbatim: what this is, who it’s for, why it’s credible. Engines extract passages; self-contained passages win. The elaboration, the personality, the proof — all of it comes after the answer, never instead of it.
• Entity schema. Organization, Person and Service in one connected graph with stable IDs, consistent across every page — so engines composing an answer can resolve who claims this without guessing. Ambiguity costs citations; entities that contradict themselves across pages read as risk and get skipped.
• Citability engineering. Facts with named sources, FAQs emitted as FAQPage schema, mechanisms explained rather than adjectives stacked. Engines lift reasoning and specifics far more readily than marketing copy — “we deliver scalable solutions” is uncitable by construction, because there’s nothing inside it to quote.
• llms.txt and crawler policy. Explicit welcome mats for GPTBot, ClaudeBot and PerplexityBot in robots.txt, plus an llms.txt that maps the site for agents — done right and priced at what it’s actually worth, which is a courtesy, not a strategy.
• The measurement panel. A fixed set of your buyers’ real questions, run against the major engines on a schedule, citations logged. You watch yourself start appearing — or you watch nothing move and hold me to account. Either way, the panel replaces faith with data.

How does the work actually run?

The audit comes first and stands alone. Two weeks: every page fetched as the crawlers fetch it, the rendering verdict per page, the entity graph mapped, the citability of your key pages scored against what currently gets cited for your queries. Fixed fee, written findings, and the top three problems flagged by impact — usable even if you never hire me again. Most audits find the same brutal headline: the site’s best content is invisible, and its visible content says nothing quotable.

Then the rebuild, in priority order. Rendering first — static generation or prerendering for the pages that matter, because nothing else counts until the words are in the HTML. Structure second — answer-first openers, question-phrased headings, FAQ schema. Entity graph third. The measurement panel runs from day one, so the before-and-after is recorded rather than remembered.

What I will not do: keyword-stuff for robots, generate AI-slop pages at scale, or fake authority with schema that overclaims. Engines are getting better at detecting all three, and the penalty for being caught is the invisibility you hired me to fix. The durable strategy is being genuinely citable — which is harder, slower, and the only version that compounds.

What does answer-first actually look like?

The single highest-leverage change is also the most concrete, so here it is as a before-and-after. A typical services page opens like this:

“In today’s fast-paced digital landscape, businesses need partners who understand their unique challenges. With over a decade of combined experience, our passionate team delivers end-to-end solutions tailored to your needs.”

Fifty words, zero extractable claims. An engine reading this learns nothing it can repeat: no what, no who-for, no evidence. It is skipped — not penalised, just ignored, which is worse.

The same page, answer-first:

“We build payment reconciliation systems for Indian NBFCs: automated daily matching between ledger, gateway and bank, with exceptions queued and reasoned. Typical deployment is four weeks; books that previously reconciled monthly reconcile every morning.”

Forty words, four extractable claims: what, for whom, how it works, what changes. An engine composing an answer to “payment reconciliation for NBFCs” can lift that passage whole — and the passage carries your name with it. Every page on a site gets this treatment: the definitive answer first, in clean declarative English, with the personality and the persuasion following it rather than replacing it.

The same rewrite improves human conversion, which is the quiet second dividend of this work — buyers skim exactly the way engines extract, and a page that answers immediately wins both readers. The disciplines converge: what quotes well, sells well.

What does the timeline look like?

Weeks one and two — the audit. Every page fetched as GPTBot fetches it, the rendering verdict per page, the entity graph mapped, your key pages scored for citability against what currently gets cited. Written findings, top three problems flagged by impact, fixed fee.

Weeks three to six — the rebuild, in priority order. Rendering fixes first, because nothing counts until the words are in the HTML. Answer-first restructuring of the pages that sell. Entity schema across the site. FAQ markup where real questions exist. Your developers can implement from my specs, or I implement directly — the audit tells us which is faster for your stack.

From week one, continuously — the measurement panel. Your buyers’ questions, asked monthly across ChatGPT, Claude, Perplexity and Google’s AI Overviews, citations logged. The first months usually move slowly — then the re-crawls compound. The panel is also your protection: if nothing moves by month three, you have the data to fire me, which is precisely why most vendors in this field don’t offer one.

After that — quarterly checks, not a retainer heartbeat. Citability decays through releases and redesigns, not through time; the check re-runs the audit on whatever changed. You pay for inspection when there’s something to inspect.

The honest part — what GEO can and cannot do

The mechanics are real: static rendering is non-negotiable, answer-first structure measurably increases extraction, entity schema resolves ambiguity, and proof-dense pages get lifted while adjective-dense pages get ignored.

But nobody — nobody — can promise you’ll be the answer. Engines blend sources, weight authority you build over years, and change behaviour without notice or changelog. Anyone selling guaranteed AI rankings is selling jugaad with a London accent. What I sell is narrower and real: your site stops being invisible, starts being citable, and you can watch the citations arrive in the panel. The gap between invisible and citable is enormous; the gap between citable and guaranteed is where the charlatans live.

One more honest thing: content that would embarrass you when quoted verbatim will, eventually, be quoted verbatim. Engines strip context and keep claims. Every page I write for this work is written to survive that — which is a quality bar most marketing copy has never had to meet, and the real reason this discipline improves sites beyond their AI visibility.

What’s the proof?

This site is the demo. Every page you’re reading is prerendered, schema’d and answer-first — view source, and everything is in the HTML, exactly as the crawlers receive it. The robots.txt explicitly welcomes the AI crawlers; the llms.txt maps the site; the entity graph resolves one person across ninety-nine pages. Does what it says on the tin, because the tin is the product.

The method is also given away in full — the GEO method article explains the four disciplines, and Ranking in AI answers is the practitioner’s sequence. Selling the work and publishing the method is deliberate: it’s the confidence trick where the trick is competence.

Related reading

Generative engine optimisation — the method · Ranking in AI answers · Technical SEO · Google updates — the foundation this builds on: /services/technical-seo

The thirty-second test

Ask ChatGPT who does what you do, in your market, right now. If you’re not in the answer, you now know the size of the problem — and the size of the head start still available, because most of your competitors haven’t run the test either. Get the audit — sorted.