Ask a room of marketers where their organic traffic went and most will say ChatGPT. The data says otherwise — and the difference matters, because it changes what you should fix.
The contrarian numbers
Roughly 15% of Google's clicks vanished in a single year. But ChatGPT accounts for only around 1–2% of search traffic. Google still processes on the order of 9 billion searches a day — no chatbot came close to displacing that volume. The clicks didn't migrate to a rival. They evaporated inside Google itself: AI Overviews answer the question at the top of the results page, and users read the summary and never click through. In sensitive categories the effect is brutal — medical queries are down roughly 30%, a third of askers getting their answer from the synthesis and bouncing.
So the first correction to the standard story: your biggest AI-visibility problem is probably Google's own answer layer, not a chatbot.
Here's the second correction. The chatbots matter more than their traffic share suggests, because of who is using them and how. An Adobe survey found that 77% of ChatGPT users use it as a search engine. And the sessions happening there aren't idle ones — they're the high-intent, high-consideration queries: vendor comparisons, procurement questions, "compare A vs B for a 500-person company." Small share of traffic; outsized share of decisions. Referral counts undersell it further still, because an assistant can shape a shortlist without ever generating a click your analytics can attribute.
What this means for where you optimize
The tempting conclusion is that you now have two separate optimization problems: one for AI Overviews, one for assistants. You don't. Both surfaces reward the same machine-legibility signals, because both are doing the same job — extracting facts from your pages and synthesizing an answer:
- Structured data. AI Overviews are assembled from content Google can parse confidently; assistants ground their answers the same way. The JSON-LD that earns citations serves both.
- Atomic, quotable claims. A synthesis engine — Google's or OpenAI's — lifts short, self-contained, verifiable sentences. The playbook for getting cited by ChatGPT and Perplexity is the AI Overviews playbook too.
- Entity consistency. Neither surface confidently names a brand the web disagrees about.
- Machine access. If fetchers can't reach fast, render-free content, you're absent from both.
The strategic shift is accepting that the click is no longer the unit of success. If a growing share of questions get answered on the results page or in a chat window, the question becomes: when the answer is synthesized, are you in it, and are you cited as the source? That's a different game than ranking — we've mapped the full contrast in GEO vs SEO.
The practical read
Don't reallocate your effort based on referral traffic; it's a lagging, undercounting indicator of AI-mediated demand. Do the work once — structure, citability, entity, access — and it compounds across Google's answer layer today and whichever assistant your buyers prefer next quarter. The brands treating these as one machine-legibility problem are already showing up in both places; the ones optimizing blue links are optimizing a shrinking surface.
Where do you stand on those shared signals right now? Run a free Legible report — it scores exactly the dimensions both surfaces read, and shows you which gap is keeping you out of the answers.