What did the study find?
When an AI Overview appeared at the top of Google, organic clicks to websites fell by 39.8% and searches that ended with no click at all rose by 34.5%. Those are the headline figures from a randomised field experiment, which its authors describe as the first causal evidence on the question, run by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University.
The method is what makes it worth reading. Earlier figures came from comparing traffic before and after Overviews launched, which mixes in every other change. This one recruited 1,065 desktop Chrome users, put a browser extension on their machines for two weeks in January and February 2026, and hid Overviews for one group while leaving them on for another. Across 68,089 searches, the groups were otherwise balanced. The users were in the US and on desktop; India is mobile-first, so treat the size of the effect as indicative. The full write-up links to the paper.
Which searches lost clicks, and which did not?
This changes the answer for a business site.
| Search type | Share of searches | Effect on clicks |
|---|---|---|
| Informational ("what is a ball valve", "how is GFRP rebar made") | 71% | Large, significant drop |
| Navigational (a company or site name) | 18% | No measurable drop |
| Transactional (looking to buy or enquire) | 12% | No measurable drop |
The authors note that the navigational and transactional estimates are noisier, so "no measurable drop" is not proof of zero. Even so, the damage is clearly concentrated where people want an explanation, not a supplier.
Position mattered too. More than 87% of Overviews sat in the top slot, and almost the whole effect came from those. The biggest loss in absolute terms fell on whoever ranked first organically.
Did people prefer the AI answer?
No measurable difference. Users with Overviews hidden reported the same satisfaction, information quality and ease of finding things as users who saw them. Their behaviour on the pages they visited did not change either. The study suggests position, not quality, drives the loss.
What does this mean for a B2B or manufacturing website?
Split your site into two halves and the picture is clearer.
- Product, capability and contact pages. These are mostly reached by people searching your name or searching to buy. The study suggests Overviews barely touch that traffic. If enquiries from these pages are falling, look first at rankings, speed and the page itself.
- Blog and knowledge articles. This is where most of the loss falls. A post explaining a standard or a process in general terms is exactly what an Overview now answers in place.
The response is not to stop publishing. It is to stop publishing what anyone could write. Google's own guidance on AI search says original, first-hand content is likely to matter most. Test data from your own plant, a selection table with real numbers, an application note for a specific industry: that is what a purchase engineer clicks through to read.
What should I check on my own site this month?
Open Search Console, filter the Performance report to queries containing "what", "how" and "why", and compare clicks for the last three months against the same months last year. Then do the same for queries containing your company name. If the first group has fallen and the second has not, you are seeing the pattern the study describes, and your enquiry pipeline is safer than the blog traffic suggests. Search remains the foundation here; the SEO section covers the rest.
Questions founders ask
How much traffic do AI Overviews take?
In a randomised field study by researchers at the Indian School of Business and Carnegie Mellon, organic clicks fell 39.8% on searches where an AI Overview appeared, and searches ending with no click rose 34.5%. Searches without an Overview showed no change.
Do AI Overviews hurt B2B enquiries?
Less than headlines suggest. The study found the click loss concentrated on informational searches. Navigational searches, such as a company name, and transactional searches showed no statistically measurable drop, though those estimates were noisier. Enquiry-driving pages are mostly reached through those searches.
Did people find AI Overviews more useful than links?
Not measurably. Users who had AI Overviews hidden reported the same satisfaction, information quality and ease of finding answers as users who saw them, and their behaviour on the pages they visited did not differ.
Should I stop writing blog posts because of AI Overviews?
Stop writing generic ones. A post that restates common knowledge is the type an Overview replaces. Posts with your own data, specifications and worked examples give an Overview something to cite and a buyer a reason to click through.