AI discovery vs SEO

Two different questions

The temptation is to treat AI search as the next version of Google, and therefore AI visibility as the next version of rankings. It is not a version of anything. It is a second surface, and it answers a different question.

That bottom row is the real difference. On a results page, position eight is bad but recoverable — you are on the page, someone might scroll. In an AI answer there is no page to be eighth on. There are three names, and either yours is one of them or the conversation happened without you.

What carries over from SEO

More than the “SEO is dead” posts suggest. Search-backed engines find pages by searching, so the work that makes a page findable still does its job.

• Pages that answer a specific question. A page built to answer one buyer question well is exactly what a search-backed assistant wants to cite. This was good SEO before and it is good AI discovery now.
• Being crawlable and fast. Nothing changed. A page an engine cannot fetch cannot be cited.
• Structured data. Machine-readable facts about what you are, who runs you and what you charge. Arguably more valuable now, because the consumer is literally a machine assembling an answer.
• Third-party presence. Directories, comparison articles and review sites were always part of SEO. They matter more here, because an assistant will happily cite someone else’s page about you.

What doesn’t carry over

• Keyword volume. Nobody types keywords at an assistant. They type sentences, with constraints — budget, team size, country. The unit of demand is a question, not a phrase, and there is no search-volume tool for questions asked inside a chat window.
• Position tracking. There is no position. Ordering within an answer exists but is weak evidence; the same question asked twice can reorder its list without anything having changed.
• Click-through rate. The answer often ends the task. You can be the recommendation and receive no visit, which makes traffic a poor proxy for whether this is working.
• One number to watch. Rank tracking gave you a tidy line on a chart. The equivalent here is a set of counts across a set of questions, which is less tidy and considerably more honest.

Why AI discovery is harder to measure

Three reasons, and it is worth being clear-eyed about all of them before buying anything — including ours.

Answers are generated, not retrieved

Two identical questions produce two differently-worded answers. The interesting question is whether the substance changes, and that is testable rather than assumable.

Some surfaces are closed

Google AI Overviews has no public API. Anyone tracking it is buying scraped search results. That is a legitimate choice with real caveats, and it should be disclosed rather than counted silently alongside API-measured engines.

There is no equivalent of a crawl

SEO tools can check every page you have. Here, you can only ask questions — and you pay per question, per engine, per run. Coverage is a budget decision, which means the choice of which questions to ask is the single most consequential part of the exercise.

Citations are where the two meet

The practical connection between your SEO work and your AI visibility is the citation list. Search-backed engines tell you which pages they used. Those pages fall into three buckets:

• Yours. Good. Whatever that page does, do more of it.
• A competitor’s. They have published something you have not, and an assistant is now repeating it as fact about your category.
• A third party’s. A directory, a review site, a “best X tools” listicle. Frequently the largest bucket, and the one most people never look at.

That third bucket is where SEO thinking pays off immediately. If the same comparison article is cited in five answers in your category and you are not in it, that is a concrete, addressable gap — and it is an outreach task, not a content task.

So should you stop doing SEO?

No, and anyone telling you otherwise is selling something. The honest position is narrower and more useful:

• Search still sends most of the traffic for most businesses today. That is an observation about now, not a prediction.
• The two surfaces share an input — pages that answer real questions clearly — so most good work counts twice.
• They need separate measurement. Your rank tracker cannot see whether Perplexity names you, and no amount of position data will tell you.
• The new work is mostly the third-party bucket: being present in the sources assistants actually cite, which is closer to PR than to on-page SEO.