The short version
- Sending every search to AI Mode cut clicks to outside websites by 18.8 percentage points per search and by about 2.3 clicks per person per day, in a US experiment with 1,100 people (Wang and colleagues, 2026).
- That loss comes on top of today’s Google, which already includes AI Overviews; hiding AI Overviews in the same experiment raised outside clicks by 8.8 points.
- AI Mode answered all 400 searches in our test, while AI Overviews appear on 36.0% to 96.0% of searches depending on the industry, so legal, healthcare, local and multi-location businesses would see the most searches newly answered by AI.
- The loss was broad: fewer people clicked through to news sites, Reddit and Wikipedia, and opinion and advice communities lost most of the extra engagement AI Overviews had sent them.
- The figures come from a forced one-week switch, a sample younger and more educated than the US population, and an AI Mode with no ads yet. Measure your own exposure before you plan around them.
What AI Mode is
AI Mode is Google’s conversational search. Powered by Gemini, it answers a search directly and takes follow-up questions. It is separate from AI Overviews, the AI summary at the top of a normal results page.
| Date | Event |
|---|---|
| March 2026 | Wang and colleagues test AI Mode as the default for 1,100 US users |
| August 21, 2025 | AI Mode rolls out globally |
| May 2025 | AI Mode launches, with early access in the United States |
| May 2024 | Google launches AI Overviews |
What the experiment tested
Researchers at the University of Pennsylvania and Northeastern University ran a preregistered field experiment on Google Search in March 2026 (Wang, Gleason, Bart, Wilson and Metaxa (opens in a new tab)). 1,100 US adults who search with Google in Chrome installed a browser extension. After three ordinary days, each was randomly assigned to one of three versions of Google for seven days.
| Group | What changed |
|---|---|
| No AI Search | AI Overviews hidden; AI Mode searches sent to the normal results |
| Current Search | Nothing: Google as it is |
| AI Mode Search | Every search redirected to AI Mode |
The AI Mode group is the closest test yet of AI Mode as the default. 94.7% of that group’s searches reached AI Mode. Before the switch, participants used AI Mode for only 0.6% of their searches, and 36% of their searches showed an AI Overview.
What happened to clicks
Compared with people on normal Google, the AI Mode group changed as follows.
| Measure | Change with AI Mode as default | 95% interval |
|---|---|---|
| Outside clicks per search (click-through rate) | −18.8 points | −22.2 to −15.3 |
| Outside clicks per person per day | −2.3 | −2.8 to −1.8 |
| Search sessions per day | −0.92 | −1.30 to −0.55 |
| Share of people who clicked a news site | −12.5 points | −18.7 to −6.3 |
| Share who clicked Reddit | −21.2 points | −27.5 to −14.8 |
| Share who clicked Wikipedia | −9.9 points | −14.7 to −5.0 |
| Share who searched on Bing, DuckDuckGo or Yahoo | +11.2 points | +6.4 to +16.0 |
| Minutes per session | +0.43 | +0.20 to +0.66 |
People spent longer inside Google and visited the rest of the web less. They also trusted what they found less: trust fell 0.34 points on a seven-point scale, and satisfaction, usefulness and sense of control all dropped. The fall in daily sessions was larger for heavy Google users.
What it means for your traffic
The study defines click-through rate as outside clicks divided by Google searches, so −18.8 points means about 19 fewer visits to other websites for every 100 searches. Each follow-up question in an AI Mode conversation counted as a search, which pulls the rate per search down; the per-day count does not have that problem.
The per-day count is the clearest traffic number. Before the switch, participants averaged 4.1 outside clicks and 9.2 searches a day. A drop of about 2.3 clicks a day is roughly 55% of that average, or 43% to 68% across the study’s interval. That share is our arithmetic, not a figure the authors report, and it assumes the comparison group stayed near its pre-study average.
How much of that reaches a given site depends on two things we measured in our AI Mode and AI Overviews study:
- Coverage. On 400 US searches run on September 26, 2026, an AI Overview appeared on 63.2%, while AI Mode answered all 400 and cited a source on 94.0%. As the default, AI Mode would also answer the searches that show no AI Overview today.
- Ranking does not carry over. Of pages in the organic top 10, AI Overviews cited 29.7% and AI Mode 16.3%; for positions 1 to 3, 42.3% against 26.0%. On the 147 searches with no AI Overview, 69.3% of AI Mode’s citations were Google’s own pages.
Being cited also does not mean being visited. A Pew Research Center panel (opens in a new tab) found users clicked a source cited in an AI Overview on 1% of visits to pages that showed one. For chat-style tools beyond Google, see how answer engines change website clicks.
Which industries have the most new exposure
AI Mode answered all 400 of our test searches, 50 in each of eight industries. AI Overviews do not appear that evenly. In our AI Overview frequency study of 800 US commercial keywords, searched on September 26, 2026, the share that showed an AI Overview ranged from 96.0% to 36.0% by industry. The lower the share today, the more of an industry’s searches AI Mode as the default would newly answer with AI.
| Industry | Keywords with an AI Overview today | 95% interval |
|---|---|---|
| B2B software and technology | 96.0% | 85.9% to 100.0% |
| Financial services and insurance | 88.0% | 79.5% to 92.5% |
| Retail and ecommerce | 68.0% | 39.5% to 85.5% |
| Hospitality and travel | 62.0% | 41.4% to 90.9% |
| Home and local services | 53.0% | 34.0% to 69.8% |
| Healthcare and dental | 43.0% | 26.9% to 75.8% |
| Legal and professional services | 40.0% | 25.9% to 56.6% |
| Franchises and multi-location brands | 36.0% | 18.6% to 64.6% |
The biggest driver is the local pack, the map with three businesses. Holding industry, intent and other factors constant, a search with a local pack had a 30.7% predicted chance of an AI Overview, against 78.6% without one. Local and multi-location businesses are among the most sheltered from AI answers today, so they would see the largest share of their searches newly answered by AI. Shares are by keyword, not weighted by search volume.
Who loses most
Every kind of site the experiment tracked lost visitors: fewer people clicked through to news sites, Reddit and Wikipedia. The authors conclude that the impact “generalizes well beyond Wikipedia and news.”
Even people who knew where they were going had trouble. Asked about their week with AI Mode, 15.3% of participants said it was harder to reach a specific website, and 13.4% mentioned fewer links or less source diversity.
Opinion and advice content loses a cushion it had. Zhang, Cui and Zhang (opens in a new tab) found that AI Overviews raised daily comments by 12.0% in Reddit communities they can cite, relative to communities they cannot, mostly in discussions of advice and personal experience. After AI Mode launched globally on August 21, 2025, that extra engagement largely disappeared: for comments it went from 9.54 a day to −0.17, and for comment authors it fell by 59%. The authors read this as a conversational interface absorbing the depth that a static summary used to send onward.
What Google says
Google disputes that its AI features cost websites traffic. As the experiment’s authors summarize it, Google argues that AI features are complements that have kept overall referral traffic steady, improved click quality and opened new chances to be surfaced through longer, more complex queries. Google has also said that AI features increase search engagement.
The experiment tested part of this. Hiding AI Overviews raised outside clicks, AI Mode reduced them, and AI Mode cut daily search sessions by 0.92 rather than adding to them. Click quality was not measured, there or in any other study cited here, so that claim remains open.
What it means for paid search
The share of people who clicked an ad fell by 42.7 points in the AI Mode group, but only because AI Mode showed no ads at the time. The authors call their results a pre-monetization baseline and expect satisfaction and other measures to shift once Google adds advertising, sponsored results or other mechanisms. For a paid search budget, how AI Mode carries ads is the open question: this experiment cannot say whether paid clicks would hold up.
What the evidence does not tell you yet
- Not a real default. The experiment forced every search into AI Mode for seven days. The authors note that gradual, opt-in adoption may behave differently, and that people may adapt over a longer period.
- Not a representative sample. Participants skewed younger, more educated and more left-leaning than the US population.
- Before ads. AI Mode carried no ads during the study, and the authors expect satisfaction and other measures to shift once Google monetizes it.
- Not by industry. The study measured everyday searches together, not your category, and did not separate high-stakes topics such as health, legal or financial questions.
- One market, one engine. US users on Google in March 2026.
How to estimate your own exposure
- Export 90 days of Google organic clicks by query from Search Console.
- Mark the queries that show an AI Overview today; part of the loss is already priced into those.
- For the rest, assume an AI answer will appear: AI Mode answered every search in our test.
- Check whether AI Mode cites your pages on your most valuable queries, and repeat the check: in our tests, when an AI Overview appeared on two days, the URLs it cited two days apart had a median overlap of only 0.43.
- Apply the study’s range as a stress test, not a forecast: about 19 fewer outside clicks per 100 searches, or roughly half of daily outside clicks, and see what that does to pipeline.
- Treat queries where people want to land on your site, such as brand, login or pricing searches, separately; in the study, people found it harder to reach a specific website in AI Mode.
If the stress test hurts, the work is to become a source that AI answers use and name, alongside protecting the demand that comes to you directly. That is what generative engine optimization is for.
Frequently asked questions
Will Google make AI Mode the default?
None of the research cited here says it will. The experiment’s authors describe Google as positioning AI Mode “as the future of search,” and AI Mode rolled out globally on August 21, 2025. This guide treats AI Mode as the default as a scenario to plan for, not a prediction.
How much traffic would we lose if AI Mode became the default?
In the only randomized test so far, outside clicks fell by 18.8 percentage points per search and by about 2.3 clicks per person per day, against a pre-study average of 4.1. Your own loss depends on how many of your queries already show AI Overviews and whether AI Mode cites you.
Does being cited in AI Mode make up for lost clicks?
Not on current evidence. Pew Research Center found users clicked a source cited in an AI Overview on 1% of visits to pages that showed one, and a study of Wikipedia found that traffic from AI Overview citations did not offset the decline AI Overviews caused.
Are AI Overviews already costing us traffic?
Probably some. In the same experiment, hiding AI Overviews raised outside clicks by 8.8 points per search. Pew found that users clicked a search result on 8% of visits to pages with an AI Overview, against 15% on pages without one.
Which industries would be hit hardest if AI Mode became the default?
The ones AI Overviews reach least today. In our sample, AI Overviews appeared on 36.0% of franchise and multi-location keywords, 40.0% of legal and 43.0% of healthcare keywords, against 96.0% in B2B software. AI Mode answered every search, so those industries would see the most searches newly answered by AI.
Which websites lose the most in AI Mode?
The experiment found losses across news sites, Reddit and Wikipedia, and the authors expect the effect to reach well beyond them. Research on Reddit found that advice and opinion communities lost most of the extra engagement AI Overviews had given them once AI Mode launched.
Sources
- Wang, Gleason, Bart, Wilson and Metaxa (2026), AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence (opens in a new tab), arXiv:2608.18352.
- Zhang, Cui and Zhang (2026), The Impact of AI Search on the Online Content Ecosystem: Evidence from Google and Reddit (opens in a new tab), arXiv:2605.16428.
- Chapekis, Lieb, Shah and Smith, Pew Research Center (2026), Investigating Click Behaviors On Google Search Result Pages That Produce an AI Overview (opens in a new tab), arXiv:2608.04831.
- Khosravi and Yoganarasimhan (2026), Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia (opens in a new tab), arXiv:2602.18455.
- Underneath (2026), AI Mode vs AI Overviews: how different are the sources?
- Underneath (2026), When does Google show an AI Overview? 1,248 US searches