Google Calls Disabled Discover Source Clicks a Bug, Ends Experiment

Google has turned off a Discover experiment that disabled taps on sources in AI summary cards, after a vice president confirmed the problem was "a bug in a Discover experiment."

Rajan Patel, VP of Engineering for Search at Google, posted the confirmation on X on Monday evening. He was replying to Lily Ray, who had quote-posted a report from Damien Andell, a researcher of Google Discover, on the morning of Oct. 23. Patel wrote: "Hey Lily – this was a bug in a Discover experiment. We've turned off the experiment, and the experience works as intended again."

Patel's reply did not say what the experiment was testing, how long it ran, or how many people saw the disabled taps.

What did the original report find?

Andell reported on Oct. 23 that Discover was testing an AI summary change that disabled clicks on sources. He found the test through a flag called "disable_aio_sources_button_tap." According to Andell, this flag stopped taps on the sources at the bottom of a summary from opening the article feed.

Before the test, tapping a summary's sources opened a feed of articles from those sources — one path from an AI summary to the sites it cites.

Andell's two screenshots show a Fox 9 video from YouTube that appeared in his Discover feed. Each screenshot shows a brief summary underneath the video and a series of icon links below it. The second image circles the icon links and labels them "Sources."

What happened after Google's reply?

On Oct. 27, the morning after Patel's reply, Andell checked the code again and posted an update on X: "I just checked and the experiment did indeed remove and replace it with another one called 'aio tap falls through to hero'."

His post did not explain how the new experiment works or where it is running. The disabled-tap flag is gone; an unexplained replacement has taken its place.

What does this change for publishers?

Publishers can track this surface through the Discover generative AI report in Search Console, which reports how many times users see links to their pages in Discover through generative AI capabilities. Each source page is listed by URL in that report.

The only metric described in Search Console Help for this report is Impression. An impression counts when someone sees a link to a website via generative AI in Discover — not a source tap. Google, in other words, measures visibility in AI summaries but does not report whether users actually tap through to cited articles.

What should practitioners do now?

Until Google or Andell explains the replacement experiment, a direct check takes seconds: tap the sources on an AI summary card in a Discover feed and see whether the article feed opens. If a publisher's Discover traffic shifts in coming weeks, this evolving tap behavior is one variable worth ruling out before blaming content performance.