Every few years the search industry produces a new acronym and a wave of people insisting everything before it is obsolete. Most of the time the underlying mechanics have barely moved. This time something did move, but not in the way the louder posts suggest.
The shift is not that search engines got smarter. It is that a growing share of buyers never see a list of results at all. They ask a question, read one generated answer, and act on it. That changes what winning looks like, and it changes it in a specific, tractable way.
The three terms, without the marketing
SEO — search engine optimisation — is the practice of making a site findable and rankable in a list of links. It has not gone anywhere. Crawlability, site speed, internal linking and content relevance are still the substrate every other layer sits on.
AEO — answer engine optimisation — is the practice of structuring content so a system can lift a complete, correct answer straight off the page. Featured snippets, People Also Ask, AI Overviews and voice results are all the same underlying problem: can a machine take a self-contained answer from your page without needing the rest of it?
GEO — generative engine optimisation — targets retrieval-augmented systems specifically. Assistants do not read pages; they retrieve passages. GEO is about passage-level clarity, consistent entity naming, statistics that carry their source, and machine-readable summaries a model can quote accurately rather than paraphrase badly.
Why they are not competing strategies
The framing of AEO or GEO replacing SEO is wrong, and expensively so. They stack.
A page an assistant cannot crawl cannot be retrieved. A page that is retrieved but buries its answer in paragraph nine will lose to one that opens with it. A page that opens with a clean answer but makes an unsourced claim gets deprioritised against one that cites a source the model can verify elsewhere.
Each layer only pays off if the one beneath it holds. That is why programmes that start at the content layer — the most visible, most satisfying place to start — so often produce activity without movement.
What genuinely changed
Four things are meaningfully different from the pre-assistant era.
- Position one is now the only position. A ranked list gives ten businesses a chance. A generated answer usually names three, and most readers act on the first.
- Corroboration outweighs self-description. Models weight agreement between independent sources far more heavily than anything you publish about yourself.
- Passages compete, not pages. Your best page may be retrieved for one paragraph while the rest is never seen.
- Being wrong is worse than being absent. An assistant confidently repeating stale information about you does damage that no amount of new content quickly undoes.
What to do differently on Monday
Start by finding out what assistants currently say about you. Write down twenty prompts a real buyer would type, run them through ChatGPT, Gemini, Perplexity and Copilot, and save the transcripts. That record is the only honest baseline you will get, and it takes an afternoon.
Then audit your top commercial pages for one thing: does a complete answer to the page's core question appear in the first hundred words, in plain prose, without requiring the surrounding context? Most pages fail this, and fixing it is the highest-return content change available.
Finally, check your entity consistency. Search your business name and note every variation of your description, category and address across your site, listings and profiles. Inconsistency here splits one strong entity into several weak ones, and it quietly caps everything else you do.
The honest limitation
None of this is deterministic. Nobody controls model training data or retrieval ranking, and anyone selling a guaranteed placement in a generated answer is selling something they cannot deliver.
What is controllable is whether your information is accurate, structured, corroborated and retrievable. That is a real body of work with a measurable outcome — provided you recorded the baseline before you started.