People Aren’t Leaving Google for AI. They’re Using Both

There is a persistent narrative that generative AI is the "Google killer." We see headlines about the death of search and the rise of the chatbot, creating a binary choice: you either use a traditional search engine or you move to an AI interface. But the actual data suggests something far more nuanced and, for those of us in the search industry, far more concerning.

People aren't necessarily switching platforms. Instead, they are layering these tools. They are staying within the Google ecosystem while simultaneously adopting AI, but the way they interact with the web is fundamentally shifting. The danger isn't a loss of users, but a loss of clicks. This connects with SEO Priorities to Rethink when the same signal needs a clearer operating decision.

The Illusion of Audience Stability

If you look at the high level audience data, everything seems fine. Similarweb data from late 2025 through mid 2026 shows that a staggering 95% of ChatGPT users also visit Google. Even as generative AI platforms saw a 70% year over year increase in visits, that overlap remained rock solid. On the surface, it looks like AI is simply an addition to the search toolkit rather than a replacement.

However, we have to be careful with how we interpret "audience overlap." A study from Bocconi University tells a different story, noting a 9.4% drop in traditional search queries for households that gained access to ChatGPT Search. The gap actually widened the longer those households had access to the tool.

The reason both of these facts can be true is that they measure different things. One counts people, and the other counts queries. You can still be a "Google user" while sending a significant portion of your actual intent to an AI. For a business owner or an SEO, the number of people visiting Google is a vanity metric if the number of queries and subsequent clicks is plummeting. The same pattern also shows up in It Comes from Other People, where the practical question is how the signal becomes visible.

Why Being a "Google User" Is a Low Bar

We need to recognize that "using Google" is a very broad category. In the eyes of audience tracking, someone who performs a single search in a month is a Google user. That person might have moved 90% of their deep research to ChatGPT, but they still use Google for a quick map direction, a business's store hours, or a Google login.

Similarweb's research clarifies that overlap doesn't analyze a single research session; it just looks at whether the same person appears in both audiences over a period of time. This means a user can remain loyal to the Google brand while systematically substituting specific types of queries with AI.

these audience metrics often miss the most disruptive shifts. API usage, dedicated desktop apps, and integrated in app AI features are often invisible in these panels. These are the exact environments where search is most likely to be replaced entirely, rather than supplemented.

Expert Interpretation: The tradeoff here is between brand reach and intent capture. You might see your "reach" remain stable in high level reports, but your "intent capture" is leaking. The decision to make here is to stop relying on audience level data and start looking at query level performance and referral traffic. If the audience is the same but the traffic is down, the tool is the culprit. A useful companion note is We Earned 1, because it looks at a nearby part of the same system.

The Decline of the Traditional Query

The Bocconi University paper provides a concrete look at how behavior changes when AI search becomes available. By analyzing U.S. desktop clickstream data, the study found that households gaining access to ChatGPT Search reduced their traditional search queries by an average of 3.14 per week. That is roughly a 9.4% decline initially, which grew to a 17% decline after 20 weeks.

Interestingly, the decline was less severe for those who were already logged in ChatGPT users compared to those who gained access later. This suggests that the "shock" of adoption is different for early adopters versus the general public.

The most critical finding for content creators is where these drops happened. The decline wasn't uniform. Informational categories took the hardest hit. Referrals to academic sites dropped by 32.8%, reference sites by 26.5%, and news sites by 13.4%. Meanwhile, marketplaces and entertainment sites saw almost no change. AI is eating the "informational" web, but it isn't yet replacing the "transactional" or "experiential" web.

Expert Interpretation: This reveals a clear divide in content value. If your site provides "quick answers" or reference data, you are in the direct line of fire. If you provide a marketplace or a unique entertainment experience, you are safer. The strategic move is to pivot informational content toward "opinionated" or "experiential" content that an LLM cannot synthesize from a reference page.

The Hybrid Search Experience

Not all research points to a decline. Some data suggests that AI can actually act as a catalyst for more web exploration. A working paper from SSRN found that users who started using LLMs actually visited more unique websites per week. There was even a temporary spike in traditional search activity before it eventually leveled off.

A 2025 field study of Chrome users supported this, showing that sessions utilizing both an LLM and a search engine lasted longer and involved more sites than sessions that used an LLM alone. In those cases, the AI seemed to act as a jumping off point, giving users the confidence or the keywords to dive deeper into the traditional web.

This suggests a "hybrid" model of search. Some users use AI to narrow down a complex topic, and then use Google to verify the facts or find the original source. In this scenario, AI doesn't replace the search engine; it optimizes the user's path to the website.

The Blur Between Google and AI

The conversation becomes more complicated when we realize that Google is no longer just a list of blue links. With the rollout of AI Overviews and "AI Mode," the distinction between "using Google" and "using AI" has vanished.

Sundar Pichai has noted that AI Overviews have over 2.5 billion monthly active users, and AI Mode has surpassed 1 billion. From Google's perspective, these features make people use Search more. But from a publisher's perspective, this is a "black box." When a user engages with AI Mode, they are technically using Google, but they may never actually see the traditional results page.

A recent experiment published on arXiv highlighted this. When users were routed into AI Mode for the majority of their searches, the experience changed fundamentally. The user is still within the Google ecosystem, but the traditional "click through" mechanism is bypassed.

Expert Interpretation: This is the most dangerous shift because it happens inside the "safe" walls of Google. The tradeoff is user convenience versus publisher visibility. Google wins by keeping the user on the page; the publisher loses the visit. You must inspect your "zero click" search rate. If your impressions are high but clicks are falling, you are likely being synthesized into an AI Overview.

The Real World Impact on Traffic

The disconnect between "audience size" and "referral traffic" is where the real pain is felt. A news site might see its Google impressions remain steady, because the AI is citing them or using their data to generate an answer, but the actual referral traffic drops.

The Bocconi study noted a 13.4% decrease in search referrals to news sites for those using ChatGPT Search. Similarly, the Wang study found a 12.5-point reduction in users clicking news links when using AI Mode. The pattern is consistent: the user is still searching, the search engine is still providing the answer, but the user no longer feels the need to click through to the source.

This effect is most pronounced in reference and academic sites, where referrals dropped by more than 25%. For these sites, the AI isn't just a tool; it's a filter that removes the need for the original visit.

The Blind Spots in Current Data

It is important to acknowledge that we are operating with incomplete information. None of the current datasets track a single individual across Google, Gemini, ChatGPT, and AI Mode at a granular, task by task level over a long period.

Similarweb focuses on audience overlap, not session level intent. The Bocconi and Gholami studies rely on desktop clickstream data, which misses the massive shift toward mobile and app based AI. Google's own claims of "record query volumes" are vague; they don't define what constitutes a "query" in AI Mode or provide a baseline for comparison.

We don't yet know exactly which types of searches are migrating to AI and which are staying in the traditional search box. We are seeing the symptoms, lower clicks, stable audiences, but we don't have the full map of the user's journey.

What to Watch Moving Forward

The landscape is shifting toward monetization. Early experiments with AI Mode often lacked ads, but that is changing. Google is already testing new ad formats created with Gemini and expanding "Direct Offers" into AI generated responses.

As ads are integrated into AI Mode, we will likely see a shift in click metrics. The introduction of commercial incentives usually changes how users interact with a page. If Google can monetize the AI response directly, the incentive to send users to external websites may decrease even further.

For those tracking this, the next major data point will likely come from Alphabet's earnings calls. The key is to look past the "growth" claims and look for baselines. We need to know if "more searches" means more people searching, or simply more queries per person within a closed AI loop that never leads to a website.

Expert Interpretation: The final decision for any digital strategist is to diversify. If your entire lead generation engine relies on the "search and click" model, you are vulnerable to a system that is designed to eliminate the click. The goal should be to build direct relationships with your audience, via newsletters, communities, or proprietary tools, so that you are no longer dependent on whether a user chooses a search box or a chat prompt.