Did Google's AI answers really stop linking to real websites earlier this year — and if so, why?

This asks about a specific claimed change: that Google's AI answers stopped linking to official websites after February 2026. To be direct, nothing in the collection documents that change, gives it a date, or explains why it would have happened. The most recent material on Google's AI features covers 2025 and earlier. If you saw this claim elsewhere, check it against a primary source before treating it as fact. What the collection does show is that AI answers were already cutting the link between searchers and source sites well before 2026, whether or not the links stayed visible.

The clearest evidence is about behaviour. Pew Research tracked tens of thousands of real Google searches. People clicked a result 8% of the time when an AI summary appeared, compared with 15% when it didn't, and they were more likely to leave without clicking anything (Do AI summaries on Google reduce clicks to actual websites?). A neat natural experiment used Wikipedia as a test case. AI Overviews launched in English before German or French, which let researchers compare the language editions. They estimated a roughly 5% drop in search traffic to English Wikipedia caused by the summaries alone (Does AI search summaries divert traffic away from Wikipedia?). The Wikimedia Foundation's own figures show human pageviews falling about 8% in 2025, though it admits the cause isn't independently confirmed (Is generative AI actually driving Wikipedia traffic declines?).

Publishers see the same pattern. News executives surveyed by the Reuters Institute expect Google referrals to fall 40–43% over three years, because the answer now appears on the search page itself (Will AI Overviews reduce search referral traffic to publishers?). So a link can still be on the page and matter less and less. Researchers call this "answer-layer intermediation": once the summary answers the question, the source becomes a footnote that few people open.

One more idea in the collection helps explain why this matters. People tend to accept fluent AI output without checking it, because checking costs effort and polished prose feels trustworthy. One note calls this "cognitive surrender" (When do users stop checking whether AI output is actually backed?). If readers rarely click through, links to official sites stop doing the job of verification even when they're present. A related argument says search was built to point people to a stable set of existing documents, while AI-generated answers are produced fresh for each query. That shift calls for provenance marking, meaning labels that show where content came from, rather than more links (Why do search tools fail against AI generated content?).

So the honest answer is that the collection can't confirm the February 2026 event. It does suggest a more interesting question: if most people already weren't clicking, would removing the links change much? Measured by clicks, the move away from official sites had largely happened before any links disappeared.