Sponsored Clicks Unchanged, Organic Fell
A randomized field experiment by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon found that outbound organic clicks fall roughly 38 per cent on queries where Google’s AI summaries appear. Users were randomly split into groups; one saw the summaries, the other had them stripped out in real time by a purpose-built browser extension.
That is the figure the headlines carried. Two findings from the same study cut deeper and appeared almost nowhere: sponsored clicks did not change at all, and neither did user satisfaction.
The first tells you where the loss is absorbed: from organic results rather than advertising. The second questions the rationale — when summaries were removed, participants’ reported satisfaction, perceived information quality and ease of finding what they needed showed no measurable change.
How the Study Worked
BU BÖLÜMÜN ÖZETİ
- Users were randomly assigned
- Earlier research was observational
- Position turned out to be decisive
- Published but not peer-reviewed
The method is what separates this from earlier work.
Users were randomly assigned
1,065 desktop Chrome users were split into three groups: standard search, a group with summaries removed in real time, and a conversational interface group. Data collection ran from 7 January to 10 February 2026.
Earlier research was observational
Previous estimates compared traffic before and after the feature launched, or queries that triggered summaries against those that did not. Neither design could rule out other factors moving at the same time. Randomization solves precisely that.
Position turned out to be decisive
More than 87 per cent of summary appearances sat at the very top of the page, above every organic listing. Removing the summary from that slot produced a marked rise in outbound clicks, while no measurable effect appeared when it sat lower down.
Published but not peer-reviewed
The paper was posted to SSRN on 3 April 2026 and revised in June. The authors describe it as a working draft to which further results may be added.
What the Numbers Say
BU BÖLÜMÜN ÖZETİ
- Clicks fall by close to a third
- Zero-click searches grow
- Advertising is untouched
- Percentages vary by revision
Four findings reinforce each other.
Clicks fall by close to a third
On queries where a summary appears, outbound organic clicks drop around 38 per cent. Outbound clicks per search rise from 0.38 to 0.61 when the summary is removed.
Zero-click searches grow
The share of searches ending without any click rises from 54 to 72 per cent when a summary appears. Roughly three in four searches complete without reaching a website.
Advertising is untouched
No change was measured in clicks on sponsored results. The loss concentrates in organic traffic, and overall search frequency does not change either.
Percentages vary by revision
Some reports state 38 per cent, others 39.8; both come from different versions of the paper. For decision-making, the direction and magnitude matter more than the specific figure.
Whose Problem Is This?
BU BÖLÜMÜN ÖZETİ
- Those sitting comfortably
- Those under pressure
- Those outside it
- How it compounds
Exposure follows which question types drive your traffic.
Those sitting comfortably
Businesses whose traffic comes mainly from brand searches and direct visits. Summaries answer definitional and general-knowledge questions; someone searching for a specific brand still arrives.
Those under pressure
Content sites and blog-led strategies drawing traffic from general information queries. “What is” and “how to” questions are where summaries are strongest, and that traffic is not expected to return.
Those outside it
Businesses drawing traffic from advertising are not directly affected; the study found no effect on sponsored clicks. Where they also have organic traffic, they feel it there.
How it compounds
Organic traffic falls, the business cuts content investment, visibility declines further and brand searches stop being fed. A trimmed content budget eventually erodes the traffic that remained.
What to Do About It
BU BÖLÜMÜN ÖZETİ
- Split your traffic by question type
- Measure beyond the click
- Produce content that cannot be summarised
- Grow your own channels
All four can be done this month and none waits for traffic to recover.
Split your traffic by question type
Which pages arrive through general-knowledge questions and which through purchase intent? The decline in the first group is structural; the second is better protected.
Measure beyond the click
A user may read the summary and then arrive via your brand name or a typed address. Those visits are not attributed to organic search, making digital’s contribution look smaller than it is.
Produce content that cannot be summarised
Content resting on your own data, your own measurement and knowledge from the field does not compress into a few sentences. Compiled information is exactly what a summary does well.
Grow your own channels
Email lists, customer portals and direct contact operate independently of search engines. Reducing dependence means building another channel.
The Digital Side
BU BÖLÜMÜN ÖZETİ
- Credit lands in the wrong channel
- Impressions and clicks diverge
- Content returns get underestimated
- The satisfaction finding matters separately
The hardest part of this shift is not lost traffic but lost measurement.
Credit lands in the wrong channel
A user who reads a summary then searches your brand gets recorded as brand search or direct traffic, though the journey began with a general query.
Impressions and clicks diverge
Impressions can rise while clicks fall. Read separately, what looks like lost visibility is actually a change in clicking behaviour.
Content returns get underestimated
A calculation looking only at organic sessions never sees the brand-awareness value of appearing inside a summary. It is why we rebuild measurement first in digital consulting work.
The satisfaction finding matters separately
Satisfaction staying unchanged when summaries are removed opens the user-benefit justification to question — a point also being watched on the regulatory side.
A Solid Digital Foundation
BU BÖLÜMÜN ÖZETİ
- Brand searches need reading separately
- Direct traffic spikes are a signal
- Compare by query group
- Measurement is set up once
When measurement breaks, the first job is separating channels correctly.
Brand searches need reading separately
Searches using your brand name behave differently from category searches; combined, the real change disappears. Google Search Central describes how search performance data can be split by query. A rise in brand searches may be the trace of lost general traffic.
Direct traffic spikes are a signal
An unexplained rise in direct visits may indicate users seeing you elsewhere and typing the address themselves.
Compare by query group
Comparing query groups rather than total traffic shows which question types lost ground. The overall total conceals it.
Measurement is set up once
Channel separation and query breakdown, defined once, generate data every period; we handle the setup within process and e-commerce consulting.
Frequently Asked Questions
Sık Sorulan Sorular
No. It applies only to queries where an AI summary appears; in the study summaries showed on roughly 42 per cent of queries.
The method is strong: randomization supports causal inference. It is nonetheless an unreviewed draft based on 1,065 US desktop users.
The figures shift slightly between versions of the paper. For decisions, magnitude and direction matter more than the exact number.
The study found no effect on sponsored clicks, meaning that channel is unaffected. Advertising is paid traffic, however, and does not replace organic visibility.
The opposite. What is affected is compiled general knowledge; content built on your own data is both harder to summarise and more likely to be cited.
By separating query types. Comparing your click rate on general-knowledge queries against brand queries shows where the loss sits.
Source: Field experiment by Saharsh Agarwal (Indian School of Business) and Ananya Sen (Carnegie Mellon University), published on SSRN 3 April 2026 and revised in June. Not peer-reviewed.