Google Reportedly Tests Removing AI Overview Citation Cards

By Nathan Brooks

Google may be testing the removal of desktop AI Overviews citation cards that normally display links to 2-3 cited domains, based on screenshots Brodie Clark posted on X.

Clark also said AI Overviews appear in more than 80% of searches. That claim would give a broad card change substantial reach, but the report provides no sample size or methodology behind the percentage.

His side-by-side screenshot shows a normal citation card on the left and the card missing on the right. The evidence is a visual interface observation, not a confirmed Google announcement or a named SEO-tool finding.

What changed on desktop?

On desktop searches, Google generally shows an AI Overview with a citation card containing links to the supporting sources. Clark said the visible cards previously featured citations for 2-3 domains alongside the answer.

Clark wrote on X: “Google has made recent changes to AI overviews that are brutal for websites. Previously, they would prominently feature citations for 2-3 domains in visible cards alongside the result.”

He described a “drastic reduction” in their appearance. Clark said website owners may see a significant decline in impressions and visibility during recent days, calling the interface change a likely contributing factor.

That wording describes a possible consequence, not a measured Google ranking decline. The post supplies no impression series, click data, affected-domain list or before-and-after query count.

Is this a test or a rendering bug?

I could not reproduce a persistent missing card after trying for more than a day. In some attempts, the citation card loaded later, leaving open the possibility of a rendering delay or another bug.

That behavior does not settle Google's intent. A delayed card could resemble a test during an early screenshot, while a test could also include an imperfect implementation that loads the module afterward.

On October 29, 2026, Clark said he now saw the behavior consistently across “a lot of test queries.” He did not publish the number of queries, their wording or the percentage of searches affected in his sample.

Clark also said the issue was brutal for websites when AI Overviews appear in more than 80% of searches. The 80% figure remained his statement; Google did not provide an official statistic or methodology in the report.

What did Google say?

I also asked Rajan Patel at Google whether this was a bug, but received no answer by publication. On October 28, 2026, Barry Schwartz separately asked on X: “a bug or a test?”

The screenshots and replies continued in an X forum discussion. Patel did not publicly answer, and Google issued no statement confirming an AI Overviews citation-card test, wider desktop rollout or permanent interface change.

The distinction matters because the screenshots establish an observed layout difference. They do not establish that Google removed citations from the underlying answer, changed rankings or reduced referral clicks.

Which sites and verticals could be affected?

Any site previously visible among the 2-3 cited domains could lose a prominent, clickable route to its pages if the card stays absent. Clark connects that visibility loss to impressions for websites, not to a verified ranking-factor change.

The report names no affected vertical and provides no site-type breakdown. It therefore cannot show whether any particular content category or publisher model faces more exposure than another.

No ranking data or traffic sample supports a segment-level conclusion either. A screenshot taken before the card finishes loading could also overstate a persistent removal, especially given the later loads observed during testing.

What should publishers monitor next?

Repeat desktop queries across multiple days and record whether the card stays missing after delayed rendering. Compare impressions and visibility for domains that commonly appear as AI Overview citations, rather than inferring a ranking loss from one interface state.

Watch for an official Google response to the October 28 question and for replication beyond Clark’s “a lot of test queries.” Until then, the decisive signal is persistent absence after a full load, not the first screenshot.