Let's start with a concrete number: a study out of Bocconi University measured what happens to your Google searches once you get expanded access to ChatGPT. The result: weekly search queries drop by 9.4% on average in the first weeks, and that drop reaches 17% after twenty weeks of steady use. If you work in SEO for AI (or you simply live off organic traffic, like most of my client roster), this is the kind of study worth reading slowly before you react to it.
The researchers did not guess at this. They used Comscore clickstream data, the same source used to measure real desktop browsing behavior in the United States, and compared how the same users searched before and after gaining wider access to ChatGPT. It is real behavior, measured through actual clicks, not an intent survey or a consultancy projection.
What did the study actually find?
The headline finding: expanded ChatGPT access cut weekly search queries by 9.4% on average. Among people who already used ChatGPT before the study, the drop started more moderate at 4.9%, but after twenty weeks of use it climbed to 8.2%. For the group that got newly expanded access, the cumulative drop at twenty weeks reached 17%.
And here is the number that matters most to me as a consultant: the drop was not even across search types. Informational queries, the ones looking for a direct answer instead of a purchase or a specific site, took the biggest hit. Reference-type searches fell 26.5%. Academic research searches fell 32.8%. Meanwhile, transactional and recreational searches, the ones that typically end in a purchase, a booking or a video, barely moved.
That lines up with what ChatGPT does best: answering direct questions without making you sift through ten tabs. It also lines up with something I have been watching in client Search Console accounts for months: glossary pages, definitions and introductory guides are starting to lose impressions before product or service pages do.
A quick methodological note: the data comes from the United States, where ChatGPT adoption has had more time to mature than in Chile or Peru. But the trend the study describes is not a US-only phenomenon. Google is already showing AI Overviews in Spanish in both markets, and generative AI adoption across Latin America has been picking up fast over the past two years. If you are not seeing this drop in your own numbers yet, it is probably a matter of time, not a matter of whether it happens at all.
Are paid ads picking up the slack?
A fair question: if organic traffic is shifting toward AI answers, are paid ads at least holding up there? Data from SE Ranking on Google's AI Mode gives a partial answer. Text ads show up in 29.45% of commercial keywords inside AI Mode. That sounds encouraging, until you see the second number: the advertiser's domain was not among the sources AI Mode cited in 88% of the keywords where ads appeared.
In other words, you paid to show up, you showed up, and in the vast majority of cases the answer the AI built for the user did not even mention you as a source. The ad and the citation are two different games, with different rules, and winning one does not guarantee anything in the other.
That undercuts an idea I have been repeating in client meetings since last year: AI visibility is not purchased the same way traditional search visibility is. You can pay to show up. You still cannot pay, at least not directly, to be the source the AI decides to cite.
What does this mean for your content strategy?
This is where most marketing teams get the first reaction wrong. They read a study like this and assume the answer is to produce less informational content, or to abandon classic SEO altogether. I think it is the opposite. If generic informational searches are the ones AI is absorbing, the opportunity sits in informational content AI cannot summarize well without citing you, because it holds a data point, an original analysis, or information that is not repeated across a hundred other pages.
This connects directly to an Ahrefs study I bring up often with clients: what correlates most with showing up in AI answers is not traditional backlinks, it is brand mentions in relevant context. Generic content, even when it is technically well optimized, competes for space against thousands of nearly identical pages. Content with a genuine point of view, built on data you generated or interpreted yourself, has nothing to compete against because there is no equivalent source.
In practice, this means reviewing two things in your content inventory.
First, which of your pages are purely definitional or reference material, think glossaries, "what is X" pages, generic lists, versus which pages carry your own data, case examples or point-of-view analysis. The first group carries the most risk of losing clicks even while holding its ranking. The second group is likely to gain ground.
Second, check whether you are only measuring organic sessions, or whether you are also tracking impressions, position and mentions outside Google Analytics. If your monthly report still reads "traffic went up" or "traffic went down" without breaking it down by search type, you are missing exactly the signal this study points to.
How can you start preparing this week?
Without any extra budget, there are three things you can check today.
1. Segment your pages by intent type
Open Google Search Console, filter your highest-traffic pages, and quickly sort them into two buckets: generic informational content versus content with data, case studies or original analysis. You do not need an elaborate spreadsheet. A list of twenty pages with a single "has its own angle: yes or no" column is enough to start prioritizing.
2. Check impressions versus clicks on your informational pages
If your glossary or definition pages are holding impressions but losing clicks month over month, that is likely exactly what the study describes: people no longer need to click through because the answer already reached them. The fix there is adding something only you can offer: your own data point, a case example, an interpretation nobody else has on hand.
3. Check your traffic sources outside Google
If you have access to Google Analytics 4, check the referral channel and look for traffic coming from chatgpt.com, perplexity.ai or similar sources. It is still a small percentage for most sites, but the month-over-month growth tells you how fast behavior is shifting for your specific audience, beyond whatever any general study says.
Wrapping up
No study, however well built, tells you exactly what will happen to your site. But this one gives a fairly clear map of where the risk sits: generic, informational content with no point of view of its own. And it also points to the way out: keep writing informational content, but stop treating it as interchangeable with everyone else's.
This week's exercise is simple: take your ten highest-traffic informational pages, and for each one ask what it holds that an AI could not simply summarize without needing to cite you. If the answer is "nothing," you already know where to start.