Generative Engine Optimization (GEO)

Generative Engine Optimization, or GEO, is the work of appearing in the answers AI systems give. Not at position one of a results list, but inside the text ChatGPT, Perplexity or Google AI Overviews hands somebody as an answer. What gets measured is therefore mention, not position.

How does GEO differ from SEO?

Both pursue the same goal of being found, but they optimise for different output shapes.

SEO optimises for a list. Success is a position, and the click is the prize.

GEO optimises for a paragraph. Success is your statement appearing in that paragraph with your brand named beside it. Whether a click follows is open, and often secondary, because the mention already does work.

One practical difference follows from that. For a list you write pages. For a paragraph you write sections that still make sense away from their page, because that is exactly how they get quoted.

What actually works

Four things, in this order.

Sections that stand alone. A clear question as the subheading, then two to four sentences that make sense without the rest of the page. Anything needing a back-reference like "as described above" cannot be quoted.

Verifiable specifics. Numbers with a source, a date and a unit. Models pick up statements that look checkable, and so do editors and readers.

Machine-readable structure. Structured data helps a machine attach statements to an entity.

An unambiguous entity. Company name, offering and remit should be described identically everywhere. Describe yourself three different ways across three pages and a model sees three half-companies.

Why this is surfacing now

Because clicks are falling even where rankings hold. Ahrefs found that AI Overviews cut the click-through rate on position one by roughly 58%. For many companies that means visibility looks unchanged in the tools while traffic declines.

The second reason is the shape of the questions. In an answer engine people ask full sentences rather than keywords, and they ask earlier in the buying process. Not appearing there means missing the shortlist before any comparison begins.

How to measure your own position in this is covered on the LLM optimisation page.

How to start

Three steps, in this order, with no tools to buy.

First, take the baseline. Write down ten to twenty questions your customers typically use to enter a topic. Put them to two different answer engines and record who gets named and who does not. Without that number, no later effect can be demonstrated.

Second, rework existing pages rather than writing new ones. On most websites the answer is already there, just embedded in prose that makes it unquotable. Turn subheadings into questions, put the answer directly beneath, remove back-references.

Third, make the entity unambiguous. Name the company, the offering and the remit identically everywhere, machine-readably via structured data.

Something stands out about that order: two of the three steps are cleanup on content you already have. The common reflex of producing new content for GEO is usually the most expensive route to the smallest effect.