AI did not replace search engine optimization. It inserted an answer layer in front of it, which split one job into two. Ranking still gets you into the results. Being quotable is what gets your name into the answer someone reads instead of the results. Most sites are still doing only the first.
The 2024 version of this question was whether SEO survives. It did. The more useful question two years on is narrower: which parts of the work still behave the way they used to, and which quietly stopped.
What actually changed
A search used to end in a list of links. Now it often ends in a written answer, assembled from several sources, with the sources named underneath or not at all.
That changes where the decision happens. A buyer who reads an assembled answer has already formed a view before any click. If your page fed that answer, you arrive pre-endorsed. If it did not, the click you were optimizing for may never happen, because the question got resolved upstream of your result.
The important consequence is not fewer clicks. It is that the page which gets quoted and the page which ranks are selected on different criteria. Ranking rewards relevance, authority and links. Quoting rewards a passage that answers a question cleanly enough to lift without editing.
Those often overlap. They are not the same test.
What did not change
Quite a lot, and this is the part the panic pieces skipped.
Crawling and indexing work as they did. A page that cannot be found or read still cannot be used, by an answer engine or anything else. Site structure, internal links, and clean technical foundations all still matter, and arguably matter more now that a second class of crawler is reading you.
Links still carry weight. So does topical depth. So does the plain, unglamorous fact that a site publishing consistently on one subject accumulates something a site publishing sporadically on ten does not.
And most published pages still get no traffic at all. Ahrefs' study of roughly 14 billion pages found 96.55% of them get zero traffic from Google, with the caveats they state themselves: the sample skews toward higher-quality content and their traffic estimates miss long-tail queries, so the real figure is lower. It was true before AI answers and it is true now. Publishing more was never the lever.
The second job
If ranking is the first job, being citable is the second.
A page that gets quoted tends to share a few properties, and none of them are exotic. It answers the question in its heading, in the first sentence under that heading, before any context or throat-clearing. It uses specifics that survive extraction: numbers, dates, named tools, concrete examples. It states tradeoffs honestly, including where the approach fails, because an answer engine assembling a balanced response needs something balanced to draw from. And it is organized so a machine can find the boundary of an idea without guessing.
None of that requires writing more. It requires writing so the useful part is not buried three paragraphs into a section.
The full mechanics of that are a separate piece: AEO vs SEO when nobody has heard of you covers the audit, the structure and the plan. This page is about what changed. That one is about what to do.
The technical trap almost nobody checks
Here is the part that catches modern sites, and it is invisible from the browser.
Googlebot executes JavaScript. Most AI crawlers do not. If your site renders its content client-side, Google eventually sees a full page and GPTBot sees an empty shell. You can be perfectly indexed and completely unquotable at the same time, and nothing in your analytics will tell you.
This is not theoretical for anyone running a React or Vue front end. It is the single highest-return technical check available right now, and it takes a minute: fetch your own page without JavaScript and look at what comes back. If the article text is not there, no amount of writing quality matters to the engines that do not render.
We wrote up the mechanics separately in Googlebot renders React, AI crawlers do not.
What this means for work you are already doing
Most of it survives, with an adjustment.
Keyword research still tells you what people type, and it still misses what they ask. Treat the search volume tool as one input and the actual prompts people put to assistants as another. The second is free to check and most teams never do.
Content briefs need one addition: for each section, what is the single sentence you would want quoted. If you cannot name it, the section is probably not saying anything specific enough to quote.
Technical work gains a new item, the one above. Everything else on the list stays.
Measurement is where honesty is hardest, because attribution got worse. An assistant that names you produces no referrer. You will see the effect as direct traffic, branded search, and people arriving already knowing what you do. That is real and largely unmeasurable, which is uncomfortable and does not make it less real.
What to stop doing
The 2024 advice that aged badly, in order of how much it now costs you.
Publishing volume to establish topical authority. That was a weak strategy then and it is a worse one now, because the marginal page adds nothing an engine wants to quote.
Writing that defers the answer. The introduction that sets context for four paragraphs before saying anything was always bad for readers. It is now also the reason a competitor's paragraph gets lifted instead of yours.
Treating AEO as a separate budget line from SEO. It is the same work, structured better. Teams that split them end up producing two versions of everything.
Chasing featured snippets specifically. The format keeps changing. The underlying behavior that earns them, answering directly and structuring clearly, is what to optimize for, and it survives every format change.
How to tell whether any of it is working
Run the same three prompts through ChatGPT, Perplexity and a Google search that triggers an AI Overview, once a month. Ask what your category's best tools are, what the difference is between your category and the obvious alternative, and how to evaluate a vendor in your space.
Note who gets named. Note which sources keep appearing. If the same competitors are cited every month and you never are, that is your gap, described precisely, for free.
It is a cruder measurement than a rank tracker and it tells you something a rank tracker cannot.