Seven months ago, 76% of the pages Google cited in AI Overviews came from the organic top 10.

That figure is now 38%.

We've been talking about the gap between rankings and AI visibility for a while. But this rate of collapse — in under a year — is something I don't think most SEO teams have fully reckoned with yet. Ranking well and being cited in AI answers are becoming genuinely separate disciplines, requiring different signals, different content structures and different measurement entirely.

It gets more interesting. Only 17% of AIO citations currently come from the organic top 10 at all. Which means the other 83% of what AI Overviews cites is coming from somewhere your rank tracker isn't watching.

Then this week a new arXiv paper added a dimension I haven't seen discussed much in the industry. Researchers demonstrated that deep-research AI agents can be "poisoned" via manipulated user-generated content — Reddit threads, forum posts, reviews — to shift what gets cited and recommended. The authors frame it explicitly as the generative-AI equivalent of black-hat SEO.

So GEO is no longer just an opportunity. It's also a threat surface.

Brands that are invisible in AI answers right now aren't just missing traffic. They're leaving a vacuum that others — including bad actors — are already working out how to fill.

The question I keep coming back to: how many teams are building a GEO strategy at all, let alone a defensive one?