The additional outbound clicks that websites get when Google’s AI previews are removed appear to be of similar quality to clicks with AIOs still included. This result comes from an updated randomized field experiment, which showed a 39.8% decrease in organic clicks when summaries are displayed.
Researchers Saharsh Agarwal and Ananya Sen have revised their working document to include new analytics on click quality, processing switches, and query types. The newspaper reported a 38% reduction when SEJ covered it for the first time in April.
The revised numbers show an increase in outbound clicks per search and non-click searches when AI previews are shown. Since AI previews were triggered on approximately 41% of queries, this equates to a measurable reduction in outbound clicks across all searches.
What click quality data shows
For clicks on queries where an AI preview would have appeared, the authors compare how often people click the back button to return to the results page, how often a visit ends within 10 seconds without interaction, and the total time spent on the destination site.
None of the three showed a statistically significant difference between the groups. About 4 out of 10 clicks on the same tab returned to the results page in both conditions, about 18% of visits ended within 10 seconds, and time spent on site was statistically indistinguishable.
Vice President of Search at GoogleLiz Reid, said that AI insights reduce “bounce clicks,” but have not released data to support the claim. The experience comes back to the same question on the other side. If summaries were absorbing mostly low-value visits, the extra clicks in the no-preview group should look worse, but they don’t.
The authors write that the result is “contradictory to the idea that AIOs primarily eliminate visits to low-engagement websites.”
Disclosure: The revised document cites two Search Engine Journal articles in its discussion of Google’s claims.
Reversed behavior when groups changed
After the initial two-week period, the researchers alternated tasks.
Participants who received AI previews saw their external clicks per search decrease, while those who no longer received them saw their external clicks per search increase. Interestingly, zero click rates were reflected.
Information requests have the effect
When the authors looked at the data by query type, they found that losses are concentrated in information searches, while removing AI previews increased the number of clicks per search. Navigational and transactional queries showed no measurable change, although these samples are smaller.
AI insights were triggered or intended to trigger on 53% of information queries, compared to 15% of navigation queries and 6% of transactional queries. A breakdown of positions adds that the three top-ranked results generated the most clicks when a top-of-page preview was removed, with the top position nearly doubling.
Why it matters
Google’s response to questions about traffic loss relied on improving the quality of clicks. This experiment tested the claim and found no measurable difference.
Looking to the future
The authors write that overall traffic losses could increase if AI insights appear in more queries over time. The document is a working draft of the SSRN and has not yet been peer-reviewed.
Featured Image: they are/Shutterstock
