GEO vs SEO: what actually changes when LLMs become the front door

For two decades, the front door of the internet was a results page. Ten blue links, one cursor, one click. Every SEO playbook ever written was implicitly written for that surface.

That surface is dissolving. ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews now answer the question before the user reaches a page. Sometimes they cite their sources. Often they don't. Either way, the bar for visibility just moved.

If you're still running pure 2018-era SEO, you are competing for the wrong real estate.

The mental model shift

SEO optimised for ranking: be on the first page, ideally above the fold.

GEO — Generative Engine Optimization — optimises for being the answer, or at minimum for being a source the answer is built from.

It sounds like a small distinction. It isn't. The implications cascade through every decision a content team makes:

• Goal: not "rank #1 for X", but "be the entity an LLM trusts to define X".
• Unit of work: not "page", but "claim". An article is a bundle of factual statements; LLMs pluck individual claims, not whole pages.
• Distribution: not "build links", but "be cited by sources LLMs already trust" — Wikipedia, Wikidata, Stack Overflow, Reddit threads, GitHub READMEs, top-tier news.
• Measurement: not "keyword rankings", but "citation share, brand-prompt accuracy, AI Overview presence".

The four signal categories that matter now

After eighteen months of GEO engagements I keep coming back to the same four buckets. If you're rebuilding a strategy, audit yourself against these — not the old technical-SEO checklist.

1. Entity clarity

LLMs reason in entities, not URLs. Does your brand resolve to one stable, well-described entity across Wikidata, Knowledge Graph, your own schema and the corpora the model was trained on? If two entities exist, or if the description disagrees with itself between languages, your retrievability collapses.

2. Source quality

Is your content the kind a careful researcher would cite? Original numbers, named authors, dates, methodology, primary references. LLMs are trained — and increasingly retrieved — through datasets that disproportionately reward this register. Generic "ultimate guide" listicles get filtered out.

3. Retrievability

When a model retrieves at inference time (RAG, browsing, plugin), can it actually parse your page? Predictable structure, clean HTML, summary blocks early, schema markup, no critical content trapped behind JavaScript or interactivity. Retrievability is the new Core Web Vitals.

4. Citation worthiness

The signal that finally matters: is your URL the one the model points at when it shows its work? Citation worthiness is built deliberately — strong claims, defensible numbers, named authority, freshness, and a footprint across the citation graph that LLMs lean on.

What stays the same

A lot, actually. Topical authority still matters. Internal linking still matters. Page experience still matters. Disambiguated entities still matter. GEO does not throw away SEO — it composes on top of it.

The teams that win the next five years won't be the ones who pivot wildly to "AI SEO". They'll be the ones who keep doing the durable parts of SEO well, then add a deliberate, measured GEO layer that makes them legible to the new front door.

Where to start this week

If you're a team of one trying to act on this, three concrete moves:

1. Pick five queries your buyers ask AI assistants today. Run them in ChatGPT, Perplexity and Gemini. Record who is cited. That's your real baseline.
2. Audit one entity surface — your Wikidata page (if any), your Person / Organization schema, the first paragraph of your "About" page. Make them agree.
3. Rewrite one cornerstone article as a citation-first piece: tight thesis, named author, dated, primary numbers, a clean summary block at the top. Watch what happens over six weeks.

The era of clicks isn't ending. It's just sharing the front door with the era of cited answers. Plan accordingly.