AI visibility vs traditional SEO: what actually transfers

If you've done SEO for years, the instinct when AI answers started mattering was probably: this is just SEO again, we know how to do this. That instinct is about half right, and the half that's wrong is expensive, because it's where the effort goes.

Here's the split.

What transfers

Technical health. Crawlable pages, clean markup, reasonable load times, no broken canonicals. Retrieval systems still have to fetch and parse your pages. Everything you did to make that easy still helps.

Structured data. This transfers with interest. Schema markup was always a hint to search engines; for language models trying to extract facts about your business, explicit machine-readable facts are worth more than they ever were for blue links.

Genuine topical authority. Being a real, recognised source on a subject (substantive content, earned links, people referring to you) still matters. The mechanism is different but the underlying signal is similar.

Content that answers a question directly. Pages structured as question-and-answer, with the answer near the top rather than buried under 800 words of preamble, extract cleanly. This was good practice for featured snippets and it's better practice now.

What doesn't transfer

Keyword position. There is no position. Nothing is being sorted into an ordered list you can move up. Rank tracking, as a discipline, has no object here. This is the hardest one to internalise because so much of SEO's measurement apparatus is built on it.

Your own pages as the primary lever. This is the big one. In search, you optimise a page you own and that page ranks. In AI answers, comparison questions get answered mostly from pages other people published: roundups, review platforms, community threads. Your own site corroborates once you're named; it rarely establishes that you should be.

Teams that miss this spend six months rewriting their own site and see nothing move, because the surface that decides the answer was never their site.

Link volume as a proxy. Links still matter indirectly, via authority and via being the kind of source that gets retrieved. But a link from a page that never appears in retrieval for your category does nothing here, however good its domain metrics look.

Long-tail keyword expansion. Publishing hundreds of pages targeting keyword variants is a search strategy. Models don't retrieve fifty near-identical pages; they retrieve the one comparison page that answers the question. Volume plays against you if it dilutes what you're clearly about.

What's genuinely new

Non-determinism. The same question produces different answers on different runs. Nothing in traditional SEO prepares you for a metric that legitimately varies run to run. It means single measurements are close to meaningless, and any honest tool reports a range and a trend rather than one number.

Multiple surfaces with different behaviour. ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews retrieve differently, weight sources differently, and disagree constantly. Being first on one and absent from another is the normal case, not an anomaly. There's no single "rank" to hold.

Description accuracy as a ranking-adjacent factor. In search, being described inconsistently across the web was untidy. Here it directly weakens the model's confidence in stating what you are, and models hedge when they're unsure. Consistency across third-party descriptions has become a real lever.

Category clarity over cleverness. Positioning language that wins a brand workshop ("the operating system for modern work") actively hurts, because it tells a model nothing about which questions you answer. Plain category language now has a mechanical payoff.

How the work splits in practice

A reasonable allocation for a team doing both:

• Roughly a quarter on your own property: structured data, a plain-language category page, public pricing, accurate facts, clean technical health. Fast, cheap, fully in your control, and a prerequisite for everything else.
• Roughly three quarters on third-party surface: review-platform listings, roundup inclusion, community presence, digital PR. Slow, external, and where the answer actually gets decided.

That ratio is close to the inverse of how most teams currently spend, which is why the effort-to-result relationship feels broken to people coming from search.

Measuring it

Traditional SEO measurement (position, impressions, clicks) has no direct equivalent. What you can measure:

• Whether you're named at all, per engine, across a fixed set of buying questions.
• How often, as a share of runs rather than a binary, because of the non-determinism.
• Who's named instead, which is usually more actionable than your own number.
• Which sources got cited, which is your target list for the next quarter.

Fix the question set and repeat it on a schedule. A changing question set produces a changing number that tells you nothing about whether you improved.

The honest summary

AI visibility is not a replacement for SEO and it is not the same discipline wearing a hat. It shares the technical foundation and the authority-building, and it diverges completely on where the decisive content lives and how you measure success.

If you keep one thing: the page that decides whether AI recommends you is usually not a page you own. Plan the work accordingly.