GEO: getting cited by ChatGPT, Perplexity and AI Overviews

A growing share of B2B research now ends without a click. Someone asks ChatGPT what an MVP should cost, or Perplexity which architecture suits their case, and gets a synthesised answer with three or four citations. If you are not one of those citations, you were not in the conversation — regardless of where you rank.

Generative engine optimisation is the work of being quotable. It overlaps heavily with good SEO, but the unit of success is different: not a ranking position, a sentence someone else's model repeats.

How it differs from SEO

SEO optimises a page to be chosen from a list. GEO optimises passages to be extracted and reused. That changes what "good" looks like.

The most useful consequence: a paragraph that only makes sense after reading the three above it is nearly useless for GEO. A model lifting that passage into an answer would produce something incoherent, so it will not lift it.

Write in liftable units

The single highest-leverage change is structural. Make each section answer one question completely, in a way that survives being read alone.

• Lead with the answer, then justify it. Models — like readers — take the first sentence most seriously.
• Repeat the subject instead of leaning on pronouns. "An MVP typically costs…" travels; "It typically costs…" does not.
• Use headings that are the actual question someone types, not clever labels.
• Put concrete numbers, ranges and named technologies in the text. Specificity is what gets quoted; "it depends" never is.

The technical layer

Some of GEO is plumbing, and it is the part most sites get wrong without noticing.

1. Server-render your content. Many AI crawlers execute little or no JavaScript. If your text only exists after hydration, it may as well not exist. Fetch your own page with curl and read what comes back.
2. Allow the AI crawlers in robots.txt — GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended. Blocking them is a choice to be absent.
3. Publish an llms.txt: a plain-text map of what your site covers and where. It is cheap and increasingly read.
4. Use FAQPage and Article schema. Structured question–answer pairs are the most directly reusable format there is.
5. Keep pages fast and stable. Crawlers have budgets and low patience.

Being an entity, not just a page

Models answer questions about things they can identify. If your company is a consistent entity across the web — same name, same description, same address, linked to the same legal entity in your schema — you are far more likely to be recalled as an option than a business whose details vary by directory.

Practical version: pick one canonical description and use it verbatim everywhere. Mark up Organization schema with your registered details. Make sure your site states plainly what you do, who for, and where — in text, not only in a hero image.

Third-party mentions matter more than links

For classic SEO, a link passes authority. For GEO, an unlinked mention in a comparison article, a listicle or a forum answer can be enough — models ingest the text either way.

This shifts where effort pays. Being included in "best X for Y" roundups, answering substantively in communities where your buyers ask questions, and publishing data nobody else has are all more valuable per hour than chasing directory links.

How to tell whether it is working

Attribution is genuinely harder here, because a citation often produces no referrer. What you can do:

• Ask the engines directly, on a schedule. Put your top ten buyer questions into ChatGPT, Perplexity and Google AI Overviews monthly and record whether you appear.
• Watch for referrals from chat.openai.com, perplexity.ai and similar in analytics — small numbers, but they are real and they convert well.
• Track branded search volume. People who meet you in an AI answer often search your name afterwards, and that shows up in Search Console.

The uncomfortable summary

Most GEO advice reduces to writing more clearly and being more specific than your competitors, then making sure a crawler can read it without executing JavaScript. There is no trick layer underneath. The sites getting cited are the ones making concrete, checkable claims in self-contained paragraphs — which was always good writing, and is now also good distribution.

Frequently asked questions