SEO in the Age of AI Search: What Changed

Does SEO still matter now that AI answers questions directly?

SEO still matters, and the reason is mechanical rather than sentimental. Google's own documentation describes its AI-generated search answers as grounded: a conventional search runs first to retrieve relevant pages, and the response is then drafted from what those retrieved pages say. A page a search engine cannot reach, index or understand is therefore not available to be drafted from either.

What changed is narrower than the headlines suggest, and worth stating exactly. A ranking used to be a reasonable proxy for being found. A ranking is now a proxy for being retrievable, which is a precondition for being named and not the same thing as being named. Two channels sit on top of the same site, and reporting that tracks one of them describes half the outcome.

What does SEO actually involve now?

The same three failures, in the same order they have always occurred. Whether a crawler can reach, render and index the pages that matter. Whether the architecture puts the commercial pages where they can be found, with internal links that state which pages carry weight. And whether anything the company owns actually answers the query, in the words a buyer uses rather than the words the industry uses.

One addition belongs on that list, and it is a change of emphasis rather than a new discipline. A page written for a ranking can rely on its own context: a heading above, a paragraph before, a diagram beside it. A passage pulled into a composed answer arrives on its own, without any of that. Writing each claim so it survives being quoted alone costs nothing in ordinary search. A company can check the result for itself: read one paragraph out of its page and see whether it still states something. Whether a system then quotes it is a separate question.

How is it different from optimizing for an answer?

SEO competes for a position among visible options, and the reader makes the final choice. Answer work competes to be one of the options mentioned at all, inside a paragraph where much of the choosing already happened before the reader saw anything. One is a contest for attention; the other is a contest for inclusion.

Measurement is where the two separate hardest. A ranking is a stable enough number that two people checking on the same day will broadly agree about it. An answer is composed fresh, varies between runs, and changes as providers update their systems, so a single check records a moment rather than a position. Falkview therefore records an answer rather than a position: the full text, the companies named in it, and the sources it cited, kept run by run so a change is visible as a change.

Overlap between the two is still substantial, and pretending otherwise would sell work nobody needs. Crawlability, indexation, architecture and plainly written pages serve both channels. The working assumption here is that entity clarity and independent sources matter more to being named than to being ranked. That assumption is stated as a hypothesis the measurement is designed to test, not as a published fact. Providers do not publish the weighting, and it cannot be read off an answer from outside.

How is it measured?

SEO reporting starts from a query set agreed before the work begins and then frozen, so later runs compare rather than reconcile. Four readings follow from it. Observed position, under recorded conditions. The share of the query set where the site appears at all. Indexation and crawl health over time. And which pages earn the positions, rather than which pages were meant to.

Recorded conditions matter as much as the number. A position varies with location, device, personalization and the moment it was taken, so a position quoted without the conditions behind it is a number without a unit. Two runs are comparable only when both state how they were taken.

AI answers get their own record alongside that, on the same schedule and the same questions where the two overlap. Keeping both in one report is the only way to see the case that matters most: a query the site ranks well for, where the composed answer on the same subject names somebody else.

What usually goes wrong?

Declaring SEO finished is the expensive version. Technical debt accumulates quietly. A redirect chain, a noindex left on after a migration, a template that stopped rendering internal links: none of it announces itself in a ranking report until months of movement have already been lost.

Abandoning SEO for answer work is the other version, and it costs more, because it removes the retrieval step those answers depend on. A site that falls out of an index loses both channels at once, and recovering an index position is slower than losing one.

Two smaller failures are worth naming. Chasing search volume rather than buying intent fills a site with traffic that never buys, while the pages that would win commercial queries go unwritten. And reporting on positions alone leaves the obvious question unanswered: whether any of those positions produced an inquiry.

How much traffic composed answers divert from ordinary results is not something this firm can measure from outside, and published estimates of it vary widely enough to be worth treating with care. The measurable version of that question is your own: what your analytics record over a period, on the same pages, read alongside the answer tracking for the same weeks.

What else is worth reading on this?

Falkview's SEO service covers the full scope, from technical foundations through the pages built for buying-intent queries, and the machine-readable foundation underneath it is technical SEO. Both channels share the same published measurement procedure.

What a Search Engine and an Assistant Disagree About, and Why explains why the two channels frequently name different companies for the same question. What Is AI Search Optimization? covers the discipline that competes for the answer rather than the ranking.