Why GEO is not SEO with extra steps — a methodology.

The TL;DR for anyone who's read enough of these.

Generative-engine optimisation (GEO) and search-engine optimisation (SEO) share a vocabulary — keywords, intent, freshness, structured data — but the engines weight those vocabulary words differently enough that a page optimised for one will reliably under-perform on the other. The five differences below explain most of the gap.

1. The unit of output.

SEO optimises for the click-through; GEO optimises for the paragraph the model returns before the click even happens. A page that ranks first on Google but doesn't get quoted by Claude when asked the same question is failing the GEO test. The two metrics will agree for the long-tail informational query and disagree on almost everything commercial.

The practical consequence is that the page has to be readable as a self-contained answer to a single question. Pages with thirty links and a sidebar full of cross-references rank fine; they get poorly quoted, because the model has nothing crisp to quote.

2. The role of dates.

Search engines have decades of priors about freshness; the LLMs have weeks. Perplexity in particular gives recency more weight than any traditional search engine — a page published this month with the same content as a page published two years ago will out-cite its older twin most of the time. ChatGPT is the opposite extreme: its citation behaviour is heavily anchored to the pre-training cutoff and only nudged at the margin by browse-search at inference time.

This means a published-at date in machine-readable form (HTTP header, schema.org datePublished, or both) is much more important to GEO than to SEO. Pages without dates are systematically penalised by Perplexity and ignored by the recency tail of the others.

3. The shape of structured data.

Schema.org markup has been a soft SEO signal for fifteen years; for GEO it is much closer to a hard requirement on certain page types. The FAQPage schema, in particular, lifts an answer paragraph verbatim into Claude and ChatGPT's responses far more often than a loose HTML question/answer block does. We see the largest citation lifts on customer FAQ pages where the markup is present and on customer FAQ pages where it isn't, the lift is roughly half.

4. The geography problem.

Both surfaces handle local intent — "best dentist in Manchester" — but they handle it differently. Google's local pack relies on Google Business Profiles and the proximity signal from the user's device. The LLMs have no proximity signal at inference time; they fall back to whichever page named the city most explicitly in its own content. This is why a page titled "Family law solicitors in Greater Manchester" out-cites a page titled "Family law solicitors" that happens to be hosted in Manchester. The geography lives in the prose, not in the host's location.

5. The role of authority.

Backlinks are the foundational SEO signal; for GEO they matter only indirectly, and only via two mechanisms. First, backlinks predict crawl frequency — the LLMs trust pages they've fetched many times. Second, well-linked pages tend to be cited by other pages the LLMs have already learned to trust. So the moat compounds, but it compounds on quoting, not on linking. A page that nobody backlinks but that gets quoted by ten other pages will out-cite a page with ten backlinks that nobody else quotes.

What this means for the brief.

For most small service businesses the practical translation is short: write fewer, longer, dated pieces that answer one specific question with prose a person would forward to a colleague. Add the schema. Mention the geography in the title. Don't try to win every prompt at once; pick nine where you have something to say, and say it.

This is genuinely different from "do SEO harder". The behaviour the engines reward is closer to magazine writing than to search writing. Most agencies who have learned to write the second won't naturally produce the first.