For twenty years, we played the "Ten Blue Links" game. We obsessed over keyword density, metadata, and backlink chasing to climb the Google rankings. But the game has changed. When someone asks ChatGPT, Claude, or Perplexity a question, they aren't looking to browse—they are looking to know.

This shift has created a new discipline: Generative Engine Optimisation (GEO).

This guide covers what GEO actually is, how AI engines decide what to cite, and the specific changes you can make to your site this week to get cited more often.

What is GEO?

GEO is the practice of structuring content so AI models can understand, synthesise, and cite it.

In the SEO era, you were a destination. In the GEO era, your content is an ingredient. The goal isn't to be at the top of a page; it's to be the source the model reaches for when it builds an answer.

The practical difference: SEO success was a click. GEO success is being named in a response you never get a visit from — and then getting the visit later, when the user searches your product by name because an AI mentioned it.

The Three Pillars: How GEO Flips the Script

Why Intent Trumps Phrasing

AI models don't match strings of text; they interpret underlying meaning.

SEO asks: "Does this page contain the phrase 'Golden Retriever'?"

GEO asks: "Does this page credibly solve the problem of a 3-month-old puppy that bites?"

This is why keyword stuffing actively hurts you now. A model reading a page padded with repetitions of "best project management software" extracts less usable meaning than one reading a page that plainly explains which tool suits a five-person team and why.

To win at GEO, your content must be structured, specific, and verifiable. Models favour precise information that is easy to lift into a direct answer.

How AI Engines Actually Find You

Modern AI systems read your work in three ways, and each rewards something different:

1. Pretrained knowledge. Your content was in the training data. Static, slow-moving, and largely outside your control.

2. Retrieval (RAG). Engines like Perplexity and ChatGPT's search mode fetch live pages at query time. This is the one you can influence most directly — it rewards clean HTML, fast loads, and crawlable pages.

3. Hybrid. Google's AI Overviews blend the web index with internal knowledge graphs.

The practical takeaway: optimise for retrieval. It is the fastest-moving channel and the only one where a change you make today can show up in an answer next week.

Nine Things You Can Change This Week

GEO advice tends to stay abstract. Here is the concrete version.

1. Answer the question in the first 100 words

Retrieval systems chunk your page and score the chunks. If your answer arrives after six paragraphs of throat-clearing, the chunk that gets retrieved is the throat-clearing. State the answer up front, then explain it.

2. Use descriptive headings, not clever ones

"How Much Does It Cost?" retrieves. "The Price of Admission" does not. Headings are strong chunk boundaries — write them as the literal question a user would ask.

3. Put facts in tables and lists

Structured data survives summarisation. A pricing table gets quoted accurately; the same numbers buried in prose get paraphrased, and sometimes paraphrased wrongly.

4. Add real numbers, dates, and named specifics

Models preferentially cite content that contains verifiable detail. "Submissions are reviewed in about four to six weeks" is citable. "Reviews take a while" is not.

5. Keep a visible last-updated date

Retrieval engines weight freshness heavily, especially for anything with a year in the query. An undated page competing against a dated one loses.

6. Implement Schema.org markup

Use Article, FAQPage, Product, and Organization markup. This is the most mechanical, highest-certainty item on this list, and most indie sites still skip it.

7. Make your pages work without JavaScript

Many crawlers execute JS poorly or not at all. If your content only exists after hydration, a meaningful share of AI crawlers see an empty page. Server-render anything you want cited.

8. Be consistent about your own name and description

If your product is described five different ways across your site, your directory listings, and your social profiles, models produce a blurry, hedged summary — or invent one. Pick one sentence and repeat it everywhere.

9. Get listed where models already look

Retrieval leans on sources it already trusts: comparison directories, review sites, community threads. A listing on a well-indexed directory is often what puts you in an AI answer, because the model finds you there rather than on your own site.

The "Machine Trust" Economy

There is a structural change underway. Referral traffic is falling for publishers because AI absorbs the query and answers it in place.

But being cited by an AI is becoming the new verified badge. When a model names your product in an answer, that is machine trust, and it converts into human trust. If you are not surfaced in generative answers, then to a growing share of users, you effectively don't exist.

This matters disproportionately for small products. You are not going to outrank an established competitor for a head keyword any time soon. But you can absolutely be the more useful, more specific, better-structured answer to a narrow question — and narrow questions are exactly what people ask AI assistants.

How to Tell If It's Working

GEO has a measurement problem: citations often produce no referrer, so your analytics show nothing.

Three things worth tracking:

• Ask the engines directly. Once a month, put your core questions to ChatGPT, Claude, Perplexity, and Google AI Overviews, and record whether you appear. Crude, but it is a real baseline.

• Watch branded search. A rise in people searching your product name without a matching rise in generic traffic usually means something is recommending you.

• Check your server logs for AI crawlers. GPTBot, ClaudeBot, PerplexityBot, and Google-Extended all identify themselves. If they are not fetching your pages at all, nothing else on this list matters yet.

What's Coming Next

We are in the early-adopter phase of GEO — roughly where SEO was in 2003. Expect:

• AI-specific metadata: schema designed for LLM synthesis rather than rich snippets.

• Protocol-level access: standards like MCP that let creators expose structured data to models directly, instead of hoping a crawler infers it from HTML.

• Citation analytics: tooling that reports how often a model used your content, closing the measurement gap above.

The Verdict: Write for Machines to Reach Humans

The good news is that GEO and genuinely good writing point in the same direction. Clear structure, specific claims, honest detail, and a direct answer near the top all serve both a model and a human in a hurry.

Don't just build a website; build a knowledge source. In a world where AI synthesises the answer, the most valuable thing you can be is the most reliable ingredient in it.