There was a comfortable assumption in SEO for a long time: rank well, and everything downstream takes care of itself. That assumption is getting harder to defend. AI systems now decide who gets cited using a partly overlapping, partly different set of signals than classic Google rankings — the discipline for working with those signals deliberately is what we call GEO — and the gap between the two is widening.

Ranking well isn't enough on its own anymore

An Ahrefs study analyzing 863,000 keywords and 4 million AI Overview URLs in March 2026 found that only 38% of AI Overview citations now come from pages ranking in Google's top 10 — down sharply from 76% in mid-2025. Pages ranking 11 to 100 account for 31.2% of citations, and pages ranking beyond position 100 account for another 31.0%. In other words, nearly two-thirds of AI citations now go to pages that wouldn't show up on the first page of a normal Google search.

That's a real shift in how visibility works. It happens because AI systems don't just answer the literal query — they break a question into several related sub-questions and pull sources for each one, reaching deeper into the results than a person scanning ten blue links ever would.

What actually predicts a citation

The same research, along with a separate Digital Applied study that manually analyzed 1,000 AI Overviews, points to a specific, fairly consistent set of signals:

  • Domain authority is the strongest single correlate with citation likelihood (a +0.61 correlation in the Digital Applied study) — established sites still have an edge, even if it's no longer the whole story.
  • Content length matters more than expected. Pages over 2,500 words are cited 1.6 times more often than pages under 800 words, likely because longer pages simply cover more of the sub-questions an AI system is trying to answer.
  • Named sources inside the content create a 2.1x lift. Pages that cite their own sources — studies, data, named experts — get cited more often themselves. AI systems appear to treat sourcing as a trust signal, the same way a careful human reader would.
  • Schema markup lifts citation likelihood by roughly 2.3x, according to Ahrefs' Brand Radar research — structured data makes it easier for AI systems to parse exactly what a page is saying and about what.
  • URL accessibility is the floor, not a bonus. A page blocked by robots.txt, sitting behind a paywall, or using a nosnippet directive is often invisible to AI citation entirely, regardless of how good the content is.

Which sources actually get cited

Domain-level patterns matter too. Everything-PR's synthesis of six independent studies covering more than 680 million citations across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews found that the top 15 cited domains capture roughly 68% of all citation share — and that Reddit alone accounts for close to 40% of that. YouTube has become the fastest-growing single source specifically within Google AI Overviews, growing 34% over six months to represent 5.6% of all AI Overview URLs, largely because AI systems increasingly favor video answers for how-to and demonstration-style queries.

The takeaway isn't that brands should abandon their own sites for Reddit and YouTube. It's that citation is now genuinely multi-channel — a strong brand presence in the communities and formats AI systems already trust, including Reddit specifically, is doing real work that a company blog post alone can't.

Why this is worth the effort

Citation isn't just a vanity metric. Seer Interactive's research, published in November 2025, found that brands cited in AI Overviews see 35% higher organic click-through rate and 91% higher paid click-through rate compared to uncited brands on the exact same queries — being the answer an AI system trusts appears to build trust with the human reading it, too.

That matters more as fewer of those readers click through to a traditional search result at all. 5W's synthesis of citation research found zero-click searches rose from 56% of queries in 2024 to 69% by May 2025, and a separate SparkToro study from June 2026 put the figure at 68% of US Google searches ending without a click. Citation, not click-through, is fast becoming the thing worth optimizing for.

Where to start

If nothing else, three moves are worth doing regardless of industry: add schema markup to priority pages, restructure your highest-intent content to run past 2,500 words where the topic genuinely supports it, and start citing your own sources inside your content rather than just asserting claims. None of that guarantees a citation — nothing does — but it moves the odds in the right direction, and the data behind each one is specific enough to act on today.

Common mistakes brands make chasing citation

A few patterns show up repeatedly in teams that start optimizing for AI citation and don't see movement. The first is treating it as a one-time project rather than an ongoing signal to track — citation patterns shift as models update, and a page that was getting cited in March can quietly stop by June without anyone noticing until traffic drops. The second is adding schema markup without actually restructuring the content underneath it, which is a bit like adding a table of contents to a book that still doesn't answer the question on the cover. The third, and probably the most common, is chasing every AI platform equally instead of figuring out which one actually sends the client's audience — a B2B software brand's buyers and a consumer wellness brand's buyers aren't necessarily using the same AI tools in the same way, and treating them identically wastes effort on platforms that were never going to move the needle for that particular business.

A starting checklist

None of the individual fixes here are complicated on their own — the difficulty is usually prioritization. A reasonable order to work through:

  • Confirm nothing important is blocked by robots.txt, paywalled, or marked nosnippet — worth checking before anything else, since it can make otherwise excellent content invisible to AI citation entirely
  • Add schema markup to the pages most likely to answer a specific, citable question — the same technical SEO foundation that helps traditional rankings
  • Pick five to ten priority pages and restructure them specifically for extraction: a direct answer near the top, named sources cited in the body, enough depth to earn a 2,500+ word count where the topic genuinely supports it
  • Set up a simple way to track citation — manually checking target queries across ChatGPT, Perplexity and Google's AI Overview is tedious but doable at a small scale, and several paid tools now track it automatically at larger scale
  • Revisit the list quarterly, since citation patterns are volatile enough that a page earning citations today isn't guaranteed to keep them