How to Do Generative Engine Optimisation (GEO)

Generative engine optimisation (GEO) is the practice of improving how clearly, accurately and prominently a brand or source appears in answers generated by AI systems.

You may also hear it called answer engine optimisation (AEO), AI search optimisation, LLM optimisation, AI SEO or generative search optimisation. The labels overlap. The practical objective is the same: make reliable information easy for machines to find, interpret, retrieve and use when answering a relevant question.

How does GEO work?

AI answer engines do not all work in exactly the same way. Some use web search or retrieval when a question is asked. Some draw on indexed pages, data partners or model knowledge. Some show citations; others mention brands without a visible link. Results can change with the model, prompt wording, location, personalisation and date.

That means GEO is not a hidden switch or a list of magic keywords. It is a coordinated way to improve five things:

• Relevance: your content directly answers questions that matter to your audience.
• Clarity: the page states who you are, what you know and what each passage means.
• Evidence: claims are supported by named expertise, examples, sources, dates or original research.
• Authority: credible websites and people independently mention, review or cite you.
• Discoverability: crawlers can access, index and understand the content and its relationships.

GEO vs SEO vs AEO

GEO does not replace SEO. Strong technical SEO, helpful pages, internal links and recognised authority remain core inputs. GEO adds a new research unit — the prompt or question — and new outcomes such as citation share and answer accuracy.

How to do GEO in nine steps

1. Define the questions that matter. Build a prompt set from customer interviews, sales calls, support tickets, on-site search, keyword research and the questions prospects ask before buying. Include definitions, comparisons, recommendations, objections, local queries and high-intent tasks.
2. Record an AI visibility baseline. Test the same prompts across the answer engines relevant to your audience. Record whether your brand appears, which competitors appear, which pages are cited, whether the description is accurate and where the answer lacks confidence.
3. Make your entity unambiguous. Use one consistent organisation name and plain-English description. Keep founder names, locations, services, credentials, contact details and social profiles aligned. Create strong About and author pages so people and machines can connect the facts.
4. Create answer-first pages. Put a concise, self-contained answer near the top. Follow it with definitions, steps, tables, examples, limitations and FAQs. Give each page one clear purpose and use headings that resemble genuine questions rather than vague slogans.
5. Add evidence and original value. Include first-hand experience, named contributors, methodology, dates, case studies, useful statistics and links to primary sources. Update material when facts change. A page that merely rewrites everybody else gives an answer engine little reason to prefer it.
6. Earn relevant third-party authority. Build relationships with trade publications, professional bodies, local media, reviewers, partners, podcasts and respected specialists. Useful expert commentary and independent coverage reinforce your entity and claims beyond your own website.
7. Strengthen technical discoverability. Keep important information in crawlable HTML, use descriptive titles and internal links, submit an accurate sitemap, avoid accidental blocking, provide canonical URLs and add structured data that matches the visible page.
8. Publish in formats people can use. Turn expertise into definitive guides, glossaries, comparison pages, original research, case studies, videos with transcripts, tools and FAQs. Cover the whole decision journey without producing dozens of thin, repetitive pages.
9. Measure, learn and improve. Re-run the same prompt set on a regular schedule. Compare the results with your baseline, investigate which sources are gaining citations and improve the pages or authority gaps connected to valuable questions.

What should a GEO-optimised page contain?

A useful GEO page is built for a person first, but it is unusually easy to extract and verify. Use this checklist:

• A descriptive title and one clear H1 using the language customers use.
• A direct answer or definition in the opening section.
• Short, meaningful sections that answer related sub-questions.
• Specific facts, examples, dates and limitations instead of unsupported claims.
• A named author or reviewer with relevant experience.
• Links to primary evidence and useful supporting pages.
• Accurate Article, Organisation, Person, Product, Service, Event or FAQ schema where appropriate.
• Original images, tables, transcripts or data that add information rather than decoration.
• A visible published or updated date for time-sensitive material.
• Fast, mobile-friendly HTML that crawlers can access without logging in or running a complex interaction.

The best GEO content is not written for a robot. It is written so clearly that a person and a machine reach the same understanding.

How to measure GEO performance

No single metric tells the whole story. Build a small scorecard around a fixed, commercially relevant prompt set.

Keep the method stable. Save prompt wording, system, date, location and whether browsing or personalisation was active. AI outputs vary, so trends across repeated tests are more useful than one screenshot.

Common GEO mistakes

• Keyword stuffing: repeating “GEO” or every product name does not create evidence or authority.
• Publishing thin AI copy at scale: volume without first-hand value can make a site less distinctive and less trustworthy.
• Treating schema as a ranking hack: structured data describes content; it does not turn weak content into a trusted source.
• Ignoring earned media: your own site explains your claim, but independent sources help validate it.
• Testing random prompts: changing the questions every month makes progress impossible to judge.
• Promising guaranteed placement: no business controls the answer produced by a third-party AI system.
• Forgetting conversion: a citation has limited value if the linked page is confusing, slow or offers no sensible next step.

A practical 90-day GEO plan

Days 1–30: research and baseline

Choose one audience and one commercial topic. Build 30–50 representative prompts, test current visibility, audit the sources being cited and identify the largest content, entity and authority gaps.

Days 31–60: improve the evidence

Update or create a small number of definitive pages. Strengthen author and organisation information, technical access, internal links and relevant schema. Add first-hand examples and begin targeted expert outreach around genuine knowledge.

Days 61–90: distribute and measure

Repurpose the strongest material into useful supporting formats, earn relevant third-party discussion, repeat the original prompt tests and connect any movement to referral traffic and enquiries. Keep what works; replace activity that only creates noise.

Harry Lang is the founder of The Oxford AI School and a former agency and client-side marketing leader. He trains marketing, PR, content and leadership teams to understand AI search and turn GEO into practical, measurable work.

Marketing teams may also find AI for Marketing Teams, our analysis of Reddit and AI search and our website optimisation workshop useful.

Further reading: the original academic paper that popularised the term Generative Engine Optimization, plus Google Search Central guidance on AI features and your website and structured data.