ChatGPT sends 0.62% of Google's traffic, three studies find

Marie Haynes pulled twelve months of GA4 data across 100 sites and found that ChatGPT delivered 0.62% of the traffic Google organic did. SEO Works, working a different dataset, put organic ahead of AI assistants by a factor of 39, and ahead on conversions by roughly 30 to one. Per iPullRank, a third study landed in the same week with the same verdict. Three independent teams, no coordination, one number that keeps showing up: AI search barely moves a traffic needle that organic search still dominates by orders of magnitude.

The instinct this data should kill is the one that treats ChatGPT citations like a new SEO channel to be optimised for clicks. It isn't one, and chasing it as one will waste budget that should go elsewhere.

The traffic argument was always the wrong argument

Marketers have spent two years asking "how much traffic does AI search send us," because that is the question SEO taught them to ask. Google organic built entire careers on that metric: rank, click, convert, repeat. ChatGPT was never going to win that contest. It doesn't show ten blue links. It gives one answer, often without a link at all, and the user who gets that answer frequently has no reason to click through to confirm it. Perplexity and Google's AI Overviews behave the same way, to varying degrees. Comparing their referral traffic to organic search's is like comparing a billboard's "traffic" to a search engine's: the billboard was never trying to generate a session.

What the 0.62% figure actually measures is referral volume, a metric built for a channel optimised for href tags. It says almost nothing about whether a brand was named, recommended, or omitted inside the answer itself, which is the thing that actually happens when someone asks ChatGPT "who are the best project management platforms" or "which banks have the strongest ESG disclosure practices." That omission or inclusion carries weight long before any click would have occurred, and in most cases, no click occurs at all.

Citation without a click is still a verdict

This is where the branding frame iPullRank advances earns its keep. A citation in an LLM answer functions less like a search result and more like a mention in a trade publication or an analyst's shortlist: it shapes the consideration set even when nobody follows the footnote. For financial services brands, that means being the custodian ChatGPT names when a prospective client asks about trust and transparency matters more than whether that conversation ever becomes a tracked session. For a multilateral or UN agency, being the source a model cites on refugee financing or disaster-risk reduction is a credibility signal that compounds across every future query on the topic, visible in model behaviour even if invisible in GA4.

Industrial groups have the most to lose from measuring this wrong. A procurement officer asking an AI assistant to shortlist suppliers for industrial automation is not going to click through to five vendor sites before forming an opinion; the model's answer is the shortlist. If a brand isn't in it, the loss shows up nowhere in traffic reports, which is precisely the danger. Philanthropic and policy institutions face the same blind spot when models summarise "leading voices" on a policy question and never surface them at all. Nobody notices the absence in an analytics dashboard, because there was never a session to lose.

What to measure instead

The practical shift this forces is a measurement one. Traffic dashboards should stop being the test of whether AI search matters to a brand, because by that test it almost never will. The test that matters is share of model: how often does ChatGPT, Perplexity, and Google's AI Overview name a brand, in what context, alongside which competitors, and with what framing of its credibility. That is a brand-tracking exercise, closer to what communications teams have always done with earned media and analyst coverage, not a referral-traffic exercise borrowed from the SEO team's dashboard.

The three studies converging on 0.62% and its cousins should end the argument about whether AI search deserves a line item, not because the traffic case is strong but because it isn't, and never was going to be. The case for AI search was never about clicks. It is about who gets named when the asking is done by a machine instead of a person, and for institutions whose value rests on being the trusted, recommended, cited source, that is a battle being fought and won or lost right now, with or without a referral to show for it.