The claim that AI killed SEO has been made confidently for two years now, usually by people selling the replacement. What actually happened is narrower and more useful to understand.
Answer engines still need sources. They still choose between pages. What changed is the basis on which they choose, and that change punished a specific kind of content that had been working for a decade.
What actually changed?
Search used to reward coverage. Publish enough pages on enough related terms, with enough words on each, and the topical footprint alone would move rankings. Depth was expensive, so volume was a real moat.
Then producing volume became almost free, and everybody did it at once. A strategy that depended on effort being scarce stopped working when the effort requirement went to zero.
Answer engines accelerated it by removing the click. When the engine answers directly, the value of ranking for a query you cannot uniquely serve drops toward nothing. You get the zero-click outcome, where your content informed the answer and nobody visited.
Why does generic content fail specifically?
Because an answer engine has no reason to cite it. If four sources say the same thing, the engine needs one of them and the choice is close to arbitrary. Being the fifth page explaining what a marketing funnel is does not create a reason to be chosen.
We watched this in our own data. Our largest pages target the broadest national terms and sit in the twenties and thirties, generating enormous impression counts and almost no clicks. Meanwhile a narrow glossary article on direct response advertising sits at position 5.4 with a click-through rate of 1.49 percent, roughly seventy times better, from a fraction of the impressions.
Broad and generic accumulated visibility. Narrow and specific accumulated visitors.
What earns a citation now?
Content with something the engine cannot get from its other sources. In practice that means:
Original data. Numbers that exist because you collected them. First-party results, aggregated performance figures, survey data. No other source can restate what only you measured.
Specificity with sources. A claim attached to a figure, a year and an origin is quotable in a way that a general statement is not.
Structure that can be extracted. Direct answers near the top, definitions phrased as definitions, comparison tables, headings written as the questions people actually ask. Answer engines quote passages, so passages have to be self-contained.
Freshness. Content updated recently is favoured, which makes a maintenance habit part of the strategy rather than an afterthought.
We are deliberately not quoting the citation-lift percentages circulating in AEO decks, including our own internal ones, because we have not been able to trace them to a primary study we would stand behind. The argument does not need them. Our own search data makes the same point with numbers we collected ourselves.
Is AI-written content the problem?
The tool is not the problem. Using it to produce more of what already exists is.
AI drafting a piece built on data you gathered, structured around a position you hold, is fine and we use it that way. AI producing the fortieth explanation of a concept with no first-party input is content that had no reason to exist before the technology arrived, produced faster.
The uncomfortable version for agencies: a great deal of content marketing was always generic, and it worked because generic used to be expensive enough to be scarce. That protection is gone.
What we do instead
Fewer pages, each with a reason to exist. Before anything goes into our pipeline it has to pass one test, which is whether a competitor could publish the identical article with their logo on it. If they could, it does not get written.
That pushes the work toward things only we can say. What our own accounts show. What went wrong on a campaign and what we changed. The unusual clients that produce problems nobody else has had to solve.
The technical layer still matters. Schema markup, llms.txt and clean answer engine optimization fundamentals make content easier to select. They cannot make an unremarkable page worth selecting.
SEO did not die. The part of it that was a volume game did, and it was replaced by something closer to publishing, where being worth citing is the whole job.
Frequently Asked Questions
Has AI search made SEO obsolete?
No. It changed what gets surfaced. Answer engines still need sources, and they select them on structure, specificity and extractability rather than on volume. Pages that were only ever competing on word count lost their advantage.
Why is AI-generated content underperforming?
Because most of it restates what already exists. An answer engine has no reason to cite a page that adds nothing its other sources already contain. Volume was a viable strategy when volume was expensive, and it stopped being one when it became free.
What kind of content gets cited by AI engines?
Content with something only that source can provide. Original data, first-party results, clear definitions, comparison tables and specific sourced figures. Structure matters as much as substance, because extractable passages are what get quoted.
Should we stop publishing blog content?
Publish less and make each piece defensible. A page that restates the category consensus has no reason to be chosen. A page reporting something only you know does, and it keeps working across both search and answer engines.