Most of your SEO work still counts. A few habits — keyword density, thin location pages, link volume for its own sake — do nothing for an answer engine.
There is a genre of post arguing that SEO is dead and AEO replaces it. It is wrong in a way that costs people money, because the overlap between the two is large and the parts that differ are specific and small.
Here is the honest split.
What carries over unchanged
Crawlability. A page a bot cannot fetch cannot be used by anything. Server-rendered HTML, sane status codes, working internal links, a sitemap that resolves. All the same, all still load-bearing.
Site speed and stability, up to a point. Timeouts matter more than milliseconds here. An AI crawler on a tight budget that hits a twelve-second response will often just give up, and unlike Google it will not come back tomorrow to try again.
Genuinely useful content. The thing that made a page worth linking to still makes it worth citing. If anything the bar went up, because an assistant summarising your page has no reason to name you unless your page had something in it the other three did not.
Structured data. Schema.org markup was always somewhat optional for SEO — a rich-result eligibility thing. For AEO it is closer to essential, because it is the only unambiguous statement of identity on your site.
Being mentioned by other people. Still the strongest thing there is. The mechanism changed slightly (see below), the importance did not.
What changes
Ten slots become three. A first-page ranking is a real result. An answer names three or four companies. This is the single biggest practical difference: the same amount of effort buys you a much narrower target, and "almost made it" pays nothing.
Keywords stop being the unit of work. People type keywords into search boxes and full questions into assistants. The useful unit is a buyer question — "best X for Y under $Z" — and the useful test is whether an assistant names you in the answer, not whether you rank for a phrase.
Entity clarity beats page optimisation. In search, a strong page can rank even if the site is a mess. In an answer, the model has to be confident about who you are before it will name you at all. A brilliant page on a site with no coherent entity is a page that gets summarised without attribution.
Links matter, but so do unlinked mentions. A model reading a sentence about your product does not need an anchor tag to absorb it. This is genuinely new, and it is worth its own post.
Freshness works differently. A live-browsing assistant reads what is on the page today. A model answering from training data is describing the internet as of some months ago. The same question, on the same day, gives different answers depending on which mode is running — which is why a single visibility check is a spot reading, not a metric.
What stops working
These are the habits worth actively dropping.
| Old habit | Why it fails now |
|---|---|
| Keyword density and phrase repetition | A model reads meaning, not term frequency. Repetition reads as low-quality text. |
| Thin location or "X for Y" doorway pages | An assistant looking for a specific answer skips past pages with no substance, and thin pages muddy your entity. |
| Link volume for its own sake | Twenty directory links from sites nothing cites do not put you in an answer. One paragraph on a site that gets retrieved does. |
| Optimising the meta description for click-through | Nobody is clicking. The description's job now is to be an accurate one-sentence definition a model can reuse. |
| Hiding text for machines | Was black-hat, still black-hat, now also trivially detectable and reputationally worse. |
The uncomfortable one: zero-click
If an assistant answers the question using your page, the reader may never visit you. That is real, it is happening, and no amount of optimisation reverses it.
What you can influence is whether you are named in the answer that replaces the click. A cited brand in a recommendation is worth more than an uncited page that supplied the facts — the reader may not click today, but they now know your name, and they will type it directly later. Most of the AEO work in this blog is, at bottom, about being the one that gets named.
A sensible allocation
If you were splitting a week of effort today: keep doing your existing technical SEO and content work, and carve out the first day for the entity layer — crawler access, Organization schema, a clear About page, consistent naming, and the three or four off-site profiles that anchor you. That layer is small, mostly one-time, and almost nobody has done it. It is the highest-leverage unclaimed ground available right now.