For most of the last two decades, "being findable" had one meaning: rank on Google. That assumption is now unreliable, because the systems answering questions no longer draw on the same sources in the same proportions.
The overlap is smaller than most people assume
Studies comparing what Google ranks against what AI assistants cite have found the correspondence is weaker than expected, and weakening. Google's own AI surfaces remain closely tied to organic results — around half of AI Overview citations come from the top 20, and nearly all Overviews cite at least one top-20 page.
ChatGPT is a different case. Even where it queries a search index, it applies its own selection on top, and the pages it ends up citing overlap much less with what Google ranks. Its most-cited single source is Wikipedia, at a share far above any commercial site.
The practical reading: Google's AI features are downstream of search. ChatGPT largely is not. Two surfaces, two sourcing behaviours, one strategy will not serve both.
Why the systems differ
Google is optimising an existing index it already trusts and has spent decades ranking. Its AI features inherit that machinery, so the sources look familiar.
Assistants without that inheritance lean on what is easy to verify and widely corroborated — reference works, well-established publications, sources that agree with each other. That favours breadth of independent mention over on-page optimisation, which is a different game from ranking.
What this changes about planning
It argues against buying "AI visibility" as a single undifferentiated service, and for being specific about which surface matters for a given business.
If your buyers use Google, conventional SEO still carries much of the load and the AI-specific work is incremental. If they ask assistants directly — increasingly common in B2B research, where a shortlist forms before any vendor is contacted — then ranking is a weak proxy and independent corroboration matters more.
Most businesses need both. The mistake is assuming that work on one automatically delivers the other.
Citation is also less stable than ranking
Rankings move gradually and can be tracked. AI citations are noticeably more volatile: the same question asked twice can return different sources, and cited sets shift between model updates.
That has a measurement consequence. A single check tells you very little. What is worth tracking is whether a brand appears across repeated samples of the questions its buyers actually ask, over time — not whether it appeared once on a Tuesday.
What holds across both
Some things help everywhere. Consistent, verifiable facts about the company. Clear entity information so it is not confused with another. Content that answers real questions in passages that survive being extracted. And independent sources that say something substantive about you.
That last one is the scarcest and hardest to fake, which is exactly why it carries weight in systems built to weigh corroboration.
Where to start
- Establish which surface your buyers actually use before deciding what to optimise.
- Sample AI answers repeatedly rather than checking once.
- Keep the SEO fundamentals — they still drive Google's AI features.
- Invest in independent coverage, which is what the assistants weigh and the only part competitors cannot trivially copy.
See our AI visibility service, or how to appear in Google AI Overviews for the Google-specific half.