YouTube is putting an AI answer box closer to the thumbs of US mobile users — and closer to the familiar problem of AI tools sounding confident when they are wrong.

The company began testing its conversational “Ask YouTube” search experience in April, initially limiting it to US YouTube Premium subscribers aged 18 and over. The experiment was designed to make search “feel more like a conversation,” combining written answers with long-form videos, Shorts and suggested follow-up questions.¹

In early testing, the feature could produce compact explainers on subjects such as Apollo 11, then organize related clips into categories including mission footage and videos about the lunar surface. It also returned background cards for Neil Armstrong, Buzz Aldrin and Michael Collins when asked who flew the mission. But some prompts, including one on Apollo 11 conspiracy theories, still led to a conventional list of search results rather than an AI-generated response.¹

That unevenness became more consequential in a test involving Valve’s new Steam Controller. Ask YouTube supplied a broadly accurate overview and relevant videos, yet incorrectly said the discontinued original Steam Controller had no joysticks. It does have one. The test’s conclusion was blunt: users “need to do your due diligence to make sure they’re accurate.”¹

Now YouTube is widening the experiment to its mobile app. Signed-in US users aged 13 and older can access it by tapping the “Ask” option in the search bar, a notably broader audience than the original Premium-only trial.²

The expansion reflects Google’s larger push to turn search into a conversational product, not merely a directory of links. Yet the early factual slip captures the trade-off: Ask YouTube may speed up discovery, but it has not removed the burden of checking what it tells viewers.