Generative engine optimisation for brands that need to be found and cited.

Generative engine optimisation, or GEO, is the work of making a business and its evidence easier for AI answer systems to find, interpret and cite when answering relevant buyer questions. It starts with normal SEO foundations, because a page still needs to be crawlable, understandable and useful. It then focuses on clear answers, named entities, visible source material, consistent facts, internal links and structured data that matches the page. TomHow records a fixed set of questions, the platform, date, returned answer, brand mention and cited URL before recommending changes. A repeat check can show whether those observed answers changed, but it cannot represent every user or prove that one edit caused a citation, enquiry or sale. GEO does not guarantee inclusion in ChatGPT, Gemini, Perplexity or AI Overviews. It creates better evidence and a fairer way to measure selected answer surfaces over time.

Thomas Howcroft's working order is to check the buyer question, improve the page's answer and topic depth, then look for proof that fits the service. That may mean a properly evidenced result, a review on an independent platform, a detailed case study or a relevant industry mention. He checks headings, internal links and page performance as part of the foundation, while keeping those technical checks separate from evidence of AI citation gains. This is ongoing organic work, often over months, rather than a quick placement.

Method source: Thomas Howcroft, founder interview, 11 October 2026. Edited for clarity; no citation gain is claimed. Google guidance for generative AI search

Before GEO, 2019 to 2023: a buyer typed a query into Google and clicked a result. You optimised your page for ranked search and you measured organic sessions. The query terminated at a link to your site.

With GEO, 2026: the buyer asks ChatGPT, Claude or Perplexity. They get a synthesised paragraph. If the answer cites your brand, you may enter the consideration set. If not, you can be missing from that answer path even when you rank on Google.

Generative engine optimisation, usually shortened to GEO, is the work of making a brand easier for AI answer systems to understand, retrieve and cite. When a buyer asks ChatGPT, Claude, Perplexity or Gemini for a shortlist, they get a synthesised paragraph rather than ten links, and there is no second results page to be found on. GEO starts by finding the questions where buyers ask for recommendations, then measuring whether the brand appears, is missing, or is described wrongly. After that the work is practical: fix the technical layer so pages can be retrieved and extracted, keep entity names consistent, and publish source material with a real reason to be used. It is simple to describe and hard to execute. GEO sits on top of SEO rather than replacing it, TomHow runs it as a measured layer with dated evidence, and it never promises a citation.

Most are technical, all are learnable, none are secret. The hard part is shipping them consistently across a real site for months on end.

• Relevant structured data. Describe the actual organisation, service and article where appropriate, with markup that matches visible facts. Google does not require special schema for AI features. Valid markup is not proof that an answer will use a page.
• Optional site map for other systems. An llms.txt file can describe useful public pages for systems that choose to read it. Google says it ignores these files for Search visibility and rankings. It is not access control, a prerequisite or a citation shortcut.
• Entity-led content. Long-form pieces that use consistent names for people, products, services, places, and categories, so the brand is easier to understand.
• Structured Q&A blocks. Content written in question-answer structure with explicit headings, so each answer block can stand on its own.
• Internal link graph. Contextual internal links that help the site read like one joined-up body of work rather than a set of isolated pages.
• Knowledge graph wiring. Wikidata, Wikipedia, and other knowledge graph references where appropriate. Useful when they genuinely clarify the entity.
• Prompt-led publishing. Content plans built from real buyer questions, not vague blog topics. Each piece targets a question where a brand recommendation could matter.

Generative engine optimisation. In plain English, it means making a brand easier for AI answer systems to understand, retrieve, and cite where the evidence supports it.

Close, not identical. AEO (answer engine optimisation) is the broader umbrella covering Google's AI Overviews, Bing's Copilot answers, and any AI answer surface. GEO sits inside AEO and focuses specifically on standalone generative engines (ChatGPT, Claude, Perplexity, Gemini). Most practitioners use the terms interchangeably.

LLMO (large language model optimisation) is often used as a direct synonym for GEO. The practical work is almost the same: make the brand clear, make the useful pages easy to retrieve, and give answer engines enough evidence to cite. TomHow handles both together.

No. SEO helps pages rank in search results. GEO helps a brand get mentioned or cited inside a generated answer. The technical work overlaps, but the measurement is different.

No. Combining them is the right framing. Organic clicks from Google still drive significant business in 2026. AI citations are an additional surface that did not exist three years ago. Brands that ignore GEO lose the new surface. Brands that abandon SEO lose the established one. TomHow does both as one combined retainer.

Agree a question set, platforms, access and checking frequency before work starts. Save dated answers, mentions and cited URLs, and separate unavailable checks from absence. API answers and consumer search experiences are different observations. Google Search Console now documents a Generative AI performance report; availability for your property must be checked before promising it in delivery. Compare repeated observations without claiming that a page edit caused a citation or sale.

Clear entity information, useful pages and corroborating mentions can all help an AI system understand a brand. Link counts alone do not prove that a model will cite or recommend it. That is why the work uses a fixed question set and dated evidence from your own category instead of promising a result from a general correlation.

The llms.txt Builder turns your business details, service pages, market, and avoid list into a copy-ready file for systems that choose to read it. Google ignores llms.txt for Search rankings and AI visibility. Useful content, crawl access and real supporting evidence take priority. Build your llms.txt.

A useful generative engine optimisation brief names the buyer questions, the answer platforms and the pages that need improvement. Start with crawl access and accurate business facts, then improve useful explanations and supporting evidence. Agree how observations will be recorded and repeated. For Google AI features, normal SEO and useful original content apply; no special AI file or schema guarantees inclusion. Treat proposed citation improvements as a test, not a promise.

Do not buy a special-file shortcut. Google says its generative search features use normal SEO foundations and useful content. It does not use llms.txt to improve visibility or rankings. Maintaining that file for other systems is different from buying a Google ranking tactic.

Likewise, do not treat a fixed answer length or extra schema as a guaranteed route into an answer. Ask which reader need a proposed change serves and how the resulting page would be more useful.

Agree a repeatable observation plan. Confirm the question panel, platforms, modes, capture dates and records you will receive. Note unavailable checks and distinguish the business being named from one of its pages being cited.

After changes ship, compare the same panel and conditions. Report what changed, what stayed the same and what remains unknown. A repeat observation can show a difference; it cannot isolate the cause of that difference by itself.

• Read how TomHow runs GEO end to endThe four-step method. What ships every month. How the citation loop closes. Honest hour estimates and honest outputs.
• The AEO pageThe wider answer-surface umbrella that GEO sits inside.
• The AI search pageSEO, GEO and AEO as one plan.
• The paid audit scopeA signed diagnostic with linked evidence rows.
• The recovery planProving whether traffic is gone or tracking has broken.
• Prompt and competitor measurementHow to keep provider evidence separate, dated and reproducible.
• What a citation-gap check can proveMentions, citations and a stable denominator for repeat checks.
• llms.txt builderFree, runs in your browser.
• Schema generatorFree JSON-LD builder, runs in your browser.