SEO for AI search: what actually changed

AI search changed one thing about SEO, not the whole discipline: getting named now matters as much as getting clicked. Ask ChatGPT, Claude, Gemini or Perplexity a question that a search engine used to answer with ten blue links, and the model just answers it, in its own words, citing whichever site it trusts most. A lot of the current advice treats that as a reason to rebuild your SEO strategy from the ground up. It isn't. Here's what actually moved, and what's exactly where you left it.

The one real change: a mention now counts as much as a click

A traditional search result is a list of links, and you click through to find the answer yourself. An AI answer is the answer, already assembled: the model reads what's out there and hands you the finished version, so visiting the source becomes optional rather than the next step. That moves the goalpost. Ranking well used to mean earning the click. Now it also means earning the citation: getting named as the source the model actually used, whether or not that visit ever happens.

That single change ripples into three practical ones.

What that shift actually changes

Three things now carry more weight than they used to: how citable your copy is, whether your structured data is in order, and how consistent your story is everywhere it's told.

• Being citable matters as much as being clickable. A model assembling an answer favours content that's unambiguous and easy to lift a clean fact from. Copy that makes a reader infer what you actually do, rather than stating it, is exactly what gets skipped for a competitor who just says it plainly.
• Structured data is cheap hygiene, not the main lever. Schema markup gives a model an explicit description of what a page is about instead of making it infer one from prose, and it genuinely helps on some platforms: Bing and Copilot are documented to use it directly, and Gemini reliably reads it when asked. But Google's own guidance says structured data isn't required for AI Overviews or AI Mode, and both Claude and Perplexity lean far more on clean, well-organised content than on the markup wrapped around it. Add it. Don't mistake it for the fix.
• Consistency across the web matters more. A model often draws on more than your own site: reviews, directories, third-party mentions. If those tell a different story to your own pages, that inconsistency can show up in the answer, sometimes as a model citing someone else's version of your own facts.

None of that replaces the basics. It sits on top of them.

What hasn't changed at all

Everything that made a site work before AI search still works exactly the same way: crawlable pages, solid technical health, and writing for an actual reader.

• Crawlability is still the prerequisite. A model testing what to say about you, like a search engine indexing your pages, still has to find and parse your content first. A broken sitemap, a redirect chain, or an orphaned page does exactly the damage it always did, arguably more: a page a model can't parse cleanly just doesn't get cited, and there's no ranking-adjacent consolation prize for landing at position eleven.
• The technical fundamentals are still fundamental. Page speed, mobile usability, duplicate meta descriptions, broken internal links: none of that stopped mattering. It's still the floor everything else stands on.
• Writing for the reader is still the right instinct. Answer-first structure, plain language, and content that's genuinely useful were already the right call for a human. It turns out they're also exactly what makes a page easy for a model to extract and cite. The two goals didn't diverge. They converged.

So the actual to-do list is shorter than the panic suggests.

So what should you actually do about it?

Don't treat AI visibility as a separate discipline bolted onto SEO. Treat it as the same foundation, crawlable, well-structured, technically sound, honestly written, with one more question added on top: would a model confidently cite this page, and is what it would cite actually true and current?

If the answer's no on either count, that's usually a small, specific fix, not a strategy overhaul. Find the gap, close it, and check again next week. Nothing on the list changed. What happens after it did: sometimes a click, increasingly, just a mention with your name on it.

A few things worth knowing

Do I need schema markup to get cited by AI search? It helps on some platforms and does nothing on others, so treat it as easy hygiene rather than the fix. Bing and Copilot are documented to use it directly, and Gemini reliably reads it when asked; Google's own guidance says it isn't required for AI Overviews or AI Mode, and Claude and Perplexity lean on clean, well-structured content over the markup itself. Add it, then put the real effort into the writing.

Does this mean traditional SEO doesn't matter anymore? No. Crawlable pages, fast load times, working internal links and a real technical health check underneath it all matter exactly as much as they did before any of this. AI visibility is one more layer on top of that foundation, not a replacement for it.

How do you actually know if a model would cite your page? Ask it. Put the real questions a buyer would type to ChatGPT, Claude, Gemini and Perplexity, and read back what each one actually says, in its own words: whether you're named, whether your site's cited, and who gets mentioned instead when you're not. Guessing from rankings alone won't tell you that; the answers will.