Wikipedia Is the Most Cited Domain in AI Search. It's Also Losing Readers.

Wikipedia is the single most-cited domain in Google's AI Overviews. It's the source the system trusts most, the one it reaches for when it needs to sound authoritative. By every measure that matters in the new game of Generative Engine Optimization — the practice of structuring content so AI platforms quote it — Wikipedia should be winning.

Its pageviews are down 8 percent.

That number doesn't make sense until you realize the new game doesn't reward being read. It rewards being referenced. And there's a difference.

The industry calls the shift GEO, though some people still use AEO (Answer Engine Optimization) or LLMO (Large Language Model Optimization). The name doesn't matter. What matters is the goal: get your content cited inside an AI-generated answer. The logic is obvious. Google's AI Overviews now appear on roughly 48 percent of all tracked queries, up from about 6 percent in early 2025. Organic click-through rates on queries with an AI Overview dropped 61 percent over the past year, from 1.76 percent to 0.61 percent, according to Seer Interactive's study of over 3 billion impressions. Global publishers lost roughly a third of their Google search referral traffic in 2025. Forbes, HuffPost, and Business Insider each lost half or more.

So the move to GEO is a rational response to a real problem. But rational responses to structural changes often look like adaptation and turn out to be just a different version of the same mistake.

The mistake is optimizing for a metric the platform controls.

With SEO, you optimized for ranking. Google controlled the algorithm. You built content, earned backlinks, and hoped the ranking factors hadn't changed overnight. The ranking was a proxy for traffic — imperfectly, but well enough that the game was playable. With GEO, you optimize for citation. You structure paragraphs so they can be extracted, front-load your answer in the first forty to sixty words, pack facts and statistics every 150 to 200 words, and sprinkle authoritative citations. A Princeton and IIT Delhi study found that research-backed strategies can lift citation frequency by up to 40 percent. Content with structured data markup is 2.5 times more likely to be featured.

But citation frequency is a much worse proxy for the thing you actually want than ranking was.

Wikipedia proves it. Wikipedia gets cited everywhere and reads nowhere. When Google's AI system summarizes an article and names Wikipedia as the source, the user doesn't click through. Only about 19 percent of users click the sources cited within AI Overviews, according to Exploding Topics. The rest get the answer and leave. Citation gives Wikipedia visibility without traffic — which sounds great until you realize that visibility without traffic is just the name of the problem GEO is trying to solve, not the solution.

The industry knows this but hasn't sat with it long enough. The marketing articles tell a different story. They report that AI search visitors convert at 14.2 percent versus 2.8 percent for traditional organic search, a five times difference. They report that traffic from ChatGPT and Perplexity grew 44 and 71 percent respectively over six months. They're not wrong about those numbers. But five times the conversion rate on a vanishingly small share of visits is still a rounding error if you're the kind of business that depends on search for discovery.

And there's another problem. GEO is harder to measure than SEO was, even at SEO's worst. In the old world you had rankings, impressions, click-through rates, and tools like Semrush, Moz, and Ahrefs that gave you a dashboard. In the new world, there is no dashboard. The industry is in a pre-Semrush era for language models. You track AI bot user agents in Google Analytics — GPTBot, PerplexityBot, Claude-Web — and manually audit how often your domain appears in responses to your core queries. There's no position, no impression count, no CTR. You have a rough frequency estimate and a lot of hand-waving.

If you can't measure it, you can't optimize it. You can only hope.

The citation decay makes the game even harder. Half of all content cited in AI answers is less than thirteen weeks old. Stale statistics, new competitor content, or structural changes in how the model extracts information can displace your citations overnight. GEO is not a moat. It's a treadmill.

Now here's the part that most GEO advocates don't address. Google isn't building this system to help brands get discovered. Google is building it to keep users inside Google. In Q2 2026, Alphabet reported $119.8 billion in revenue, up 24 percent year over year. Search advertising grew 17 percent to $63.3 billion. That's the first sequential deceleration in six quarters, down from 19 percent in Q1. Growth hasn't collapsed, but the acceleration phase has paused. Meanwhile Google spent $44.9 billion on AI infrastructure in the quarter alone, producing its first-ever negative free cash flow. That spending was funded partly by a $49.6 billion equity raise.

Google is spending roughly $200 billion this year to make sure people never need to leave Google. The business incentive is to give people a good enough answer on the results page that they don't click through, then monetize the engagement with ads and the AI interaction itself. Brands optimizing for citation are playing on a field where the house is actively trying to make the game produce less traffic for everyone else.

So what's the right move?

The honest answer is that GEO is worth doing, but only as a supplement to whatever else works. Structuring your content so AI systems can extract and cite it — clear headings, direct answers, fact density, schema markup — tends to improve traditional SEO anyway, because it aligns with Google's own helpful-content guidelines. But treating GEO as a replacement for SEO would be like switching from driving to flying because traffic was bad, without checking whether you had a runway.

The real question for brands isn't how to optimize for citation. It's what to do when citation doesn't produce clicks. If you're a B2B company selling to a small niche where informational queries matter and each qualified lead is worth thousands, GEO might move the needle. If you're a consumer brand that depends on discovery volume, the math doesn't work the same way. Being cited fifty times by AI systems that send you two actual visitors per thousand citations is not the same thing as ranking.

The thing about optimization games is that they always look like strategy until you notice what you're optimizing for isn't what you want. SEO was a proxy for traffic. GEO is a proxy for citation. The proxy is getting worse, not better.

I suspect the brands that survive this shift won't be the ones who optimized best for AI answers. They'll be the ones who built enough name recognition that people ask for them by name — the queries where the AI can't dodge, because the user's intent is specific and the answer is a brand, not a summary. Search for "best CRM" gets you a synthesized answer that cites four competitors and sends you to none of them. Search for "Salesforce pricing" sends you to Salesforce. The more your business becomes the answer rather than a source of information about the answer, the less you need to play the citation game at all.

If you want to test whether GEO is worth your time, try this. Take your top ten core queries — the ones that bring most of your search traffic. Ask ChatGPT, Perplexity, and Claude the same questions. Note whether your brand is cited. Then note whether any of those citations produce a click to your site. Do the math on actual click-through from citations versus the hours you're spendingon GEO tactics. If the ratio doesn't justify the investment, the problem isn't your optimization. It's the game.

The smarter move might be to build a product or a reputation so specific that the citation becomes unnecessary. Because at the end of the day, being mentioned in passing by an AI system is not the same thing as being the place people want to go.