Seven in ten AI citations already rank on Google
Ahrefs looked at which pages actually get cited in AI answers, and the result reframes the whole conversation. 38% of the pages cited in AI Overviews (the AI written answers Google now shows at the top of its results) also rank in the top 10 on Google for the same query. Another 31% rank somewhere in positions 11 to 100. Add those together and roughly seven out of ten pages an AI chooses to quote were already ranking in ordinary search.
Read that twice, because it quietly kills a popular myth. There is no secret side door into AI answers that skips search entirely. The pages models reach for are, overwhelmingly, the pages that were already findable.
Why AI reaches for pages that already rank
It makes sense once you picture how these systems work. An AI assistant does not roam the live web the moment you ask it something. It pulls from an index (the library of pages it has already stored), picks a handful it judges relevant and trustworthy, and writes its answer from those few.
The signals it uses to choose them, whether it can crawl you (reach and read your pages), whether your content is clear, whether credible sites point to you, are largely the same signals that decide where you rank. So a page that is invisible to a search crawler is usually invisible to the model too. A page that is the clean, well structured, authoritative answer to a question tends to get surfaced by both. They are drinking from the same well.
The foundation does most of the work
Here is the part that usually gets skipped. The same groundwork that earns you rankings is what makes you quotable in the first place. Most of the result comes from getting the fundamentals right, not from some AI specific trick bolted on at the end.
The fundamentals are unglamorous, and they barely change from one year to the next:
- Crawlability and speed. If a bot cannot reach a page, or gives up waiting for it to load, that page does not exist as far as Google or an AI is concerned. Check that nothing in your robots.txt (a small text file that tells bots which pages they may read) is quietly shutting the door on the crawlers you actually want.
- Clarity. Every page that matters should say, in plain words a stranger could repeat back, what it is and who it is for. Vague homepages confuse people and machines in exactly the same way.
- Earned trust. Coverage, citations and links from sources your buyers already respect. This is the slow part, and the part no shortcut replaces.
Where AEO is genuinely its own thing
None of this means AEO is a myth. It is real. But it is a refinement on top of solid fundamentals, not a substitute for them. A few things are genuinely specific to how a model reads a page:
- Self contained answers. A model lifts a paragraph out of context and drops it straight into a reply. Write so that any single passage still stands on its own, without leaning on the sentence before it.
- Explicit structure. Schema and JSON-LD (invisible code tucked into your page) spell out what a page is in a format machines do not have to guess at. An llms.txt (a plain text file that points AI assistants at your key pages) and an honest sitemap (a plain list of all your real pages) point crawlers straight at what counts.
- Depth across a topic, not a single keyword. One buyer question splinters into many smaller ones behind the scenes. The sites that surface most are the ones that cover the whole subject, not one lonely landing page.
- Being talked about where it matters. Turning up in the sources and conversations a model already treats as reliable, earned honestly rather than manufactured.
What this means for your site
If you have a website you can update, and almost everyone does, this is genuinely good news. You are not starting from zero with AI. The rankings and the structure you already have are most of the battle. The job is to stop leaking on the basics and then add the thin AI layer on top, rather than chasing a separate, mysterious AI strategy that floats free of everything else you have built.
It also means order matters. Fix the foundation first. A polished llms.txt on a site that blocks crawlers, loads slowly or never says clearly what you do is effort spent in the wrong place, decorating something that was never built to be found.
One game, two scoreboards
The expensive mistake we see is treating AEO and SEO as two budgets, two checklists, often two different people, with the new and exciting one getting all the attention. They are not two games. They are one game with a new scoreboard, and the pages that win the old board win most of the new one for free.
So before you buy another tool that promises to ask the AI whether it knows you, ask a duller and more useful question: is my site actually findable? Can it be crawled, is it structured, does it deserve to rank? Fix that, and being cited by AI stops being a mystery and turns into a side effect.