A peer-reviewed study just measured what ChatGPT is doing to Google search.

Everyone in search has assumed ChatGPT is pulling queries away from Google, but assumption is not measurement, and until this month the number was a matter of opinion. Then Search Engine Journal published a Bocconi University study on July 13 that put a peer-reviewed figure on it, and the figure travelled fast, because broader access to ChatGPT was tied to a 9.4% weekly reduction in traditional search that deepened the longer people used it, reaching around 17% after twenty weeks.

The 9% headline is now everywhere, which is fair, but the more useful information is in the parts that did not fit in a headline. What makes this study worth reading rather than just citing is that it separates the loss by query type and by the kind of site that gains or loses, and that is where a marketing team can actually make decisions, because "search is down 9%" tells you to worry while "here is exactly which searches are leaving" tells you what to do.

The core finding, and how deep it goes

The top-line comes from comparing groups with broader versus narrower ChatGPT access, and the Bocconi study reported in Search Engine Journal found the 9.4% weekly reduction growing to roughly 17% after twenty weeks of use. When the researchers looked specifically at people who were already ChatGPT users, comparing two groups, the reduction averaged about 4.9% and rose to around 8.2% after the same twenty-week window. The consistent shape across every cut of the data is that the effect compounds, so this is a behaviour change rather than a one-time novelty dip that reverses once the shine wears off, and a habit that deepens over twenty weeks is exactly the kind of thing that wrecks a flat-line traffic forecast built on last year's volumes.

The searches that are actually leaving

The averages hide a sharp split, and the split is the most actionable thing in the paper. The declines are concentrated in informational queries, with academic research searches down 32.8% and reference-style searches down 26.5%, while transactional and recreational searches barely moved. People are taking their "explain this to me" and "what is the background on this" questions to a chatbot, and keeping their "buy this" and "find me something fun" questions on Google.

For content strategy that distinction is close to a map. If your traffic leans heavily on top-of-funnel explainers, definitions and reference material, you sit in the exposed zone and should expect erosion, so the answer is either to make that content genuinely worth the click or to earn citation inside the AI answers now absorbing those queries. If your value is transactional, comparative or experiential, you are on firmer ground for now.

Most real businesses are a mix, and the exercise worth doing is sorting your own pages into those buckets so you know which part of your library is under pressure and which part is safe to keep investing in. It is a dull afternoon of work and it will change how you spend the next two quarters.

Fewer clicks, but a different kind of click

The study also measured where sessions send people, and the contrast is stark, because ChatGPT sessions directed users to external websites only 5.2% of the time against 31.1% for Google sessions, which confirms the obvious point that a chatbot keeps far more journeys inside itself. The less obvious point is what kind of traffic survives that filter, and it is not evenly distributed, since the clicks ChatGPT does pass along skew away from ad-supported, viral and popular-destination sites and toward reference sites, SaaS tools, academic sources, developer sites and non-profits. Ad-supported sites make up around 27.6 percentage points less of ChatGPT referral traffic than they do of Google's, which suggests these referrals are higher-intent and harder to monetise through display advertising, but potentially better-converting for specialised business-to-business and tool-led sites. If you sell software, run a genuinely useful tool, or publish reference material people act on, the ChatGPT referral you get may be smaller in volume and better in quality than the equivalent Google visit, which is a very different conclusion from "AI is killing our traffic" and worth checking against your own analytics before you decide which camp you are in.

The trust gap nobody has resolved

Running alongside all this is a finding that complicates the picture in a useful way. A YouGov survey from July found that only 28% of Americans trust AI search answers, and that low-trust number sits against usage that keeps climbing anyway, which is a familiar human pattern and also a genuine opening.

The gap suggests there is real value in content that positions itself as the verifiable, original, human-checked source behind an answer, because a meaningful share of users want to confirm what the AI told them, and being the place people land when they go to check the claim is a legitimate strategy. It rewards primary research, clear sourcing and the kind of first-hand expertise a model cannot manufacture, and since the trust gap will not stay open forever, the brands that build that reputation while scepticism is still high are the ones likely to hold it as trust rises.

How to spend against this

The wrong response is to panic-cut your SEO budget, because the study does not say search is dying, it says a specific slice of it is moving, so the right move is to reallocate toward what still works. Sort your content by query type, defend or upgrade the informational material that is most exposed, lean into the transactional and comparative pages that are holding, and start tracking ChatGPT referral traffic as its own line so you can see what it converts at rather than guessing, because the volume is small today and rising and you want a baseline before it matters.

Underneath all of it, the durable move is to become the source worth citing and worth verifying, since that is what earns both the AI citation and the human who clicks through to check it. Deciding which pages to defend and which to let go is difficult and easy to get wrong alone, and it is where our SEO engagements tend to begin, so your budget follows where the demand actually went instead of where it used to be.