Google is testing AI-generated micro-headlines in Discover that replace publishers’ original headlines. Unsurprisingly (given Google’s messy experience with AI-generated Search summaries last year), the results range from misleading to nonsensical. The experiment involves replacing the headlines that writers and editors actually crafted with ultra-short, AI-generated titles, each four words long.
For example, Google’s AI rewrote an Ars Technica story about Valve’s Steam Machine as “Steam Machine price revealed,” but Valve hasn’t disclosed a price yet. In another example, it turned a PC Gamer article into “Schedule 1 farming backup,” which doesn’t make any sense at all without context. A 9to5Google piece’s title got rewritten to “Qi2 slows older Pixels,” though the original simply warned readers not to expect faster speeds, The Verge reports.
Real results in Google Discover.
Credit: The Verge
Google spokesperson Mallory Deleon described this as “a small UI experiment for a subset of Discover users,” claiming the company is testing a new design to make topics easier to digest before readers click through. The AI headlines technically feature a small label that reads, “Generated with AI, which can make mistakes,” but it’s only visible after the user taps “See more.” Since the publication’s name sits right beside the AI headline, readers can easily assume the publication itself is responsible for AI-generated clickbait or slop.
This experiment is part of a clear pattern of AI-related abuse against the open web. Today, Google aims to keep users within its own interface with AI summaries and Discover snippets, rather than sending traffic to the sites that host the original content. Many online publications say this harms their numbers, both financially and traffic-wise—something that research supports—but Google insists it isn’t a problem.