Google is testing AI to rewrite newsroom headlines in its Discover feed. The test has spread into a live feature that changes the titles users see. The outcomes are haphazard and even misleading. The readers observe a headline that is not in line with the article. Publishers lose traffic and trust.
Why This Experiment is Worse Than a Botched Joke
AI is great at speed and summaries. AI is not great at context and truth when it hallucinates. The Discover tool rewrites publisher headlines and sometimes turns nuance into nonsense. A story about a game exploit became a headline that sounded like abuse. A technical post about charging became a claim that older phones slow down. Those are not small slips. They change what people believe.
Google says the feature helps users get context fast. Google also says it is a feature, not just an experiment. That is the problem. When a platform decides to rewrite headlines at scale, it bends the information flow. The feed is huge. Millions of people use Discover each day. When a big engine changes headlines, it can shift public view, and it can steal clicks from real journalists.
Publishers are not small players in this story. Many news sites rely on referral traffic from Google to survive. When Discover shows an AI title that is wrong, the reader may blame the outlet. That turns a careful piece of reporting into a reputation risk. Some outlets have already raised alarms. They say the feature erodes editorial control and damages brands that spent years building trust.
The change also hides the AI label from most users. Discover does not show that an AI made the headline unless the user digs into a menu. That lack of transparency is a trust trap. Readers assume the headline came from the reporter or the editor. That assumption will break trust both ways. Publishers will be blamed, and platforms will look like they do not care about accuracy.
This is not the first time a major tech company shipped a messy AI news feature. Apple paused a similar alert system after it spread false claims about events. The lesson is clear. Fast AI features can amplify errors and then do real damage. Regulators should notice this pattern now before the damage grows deeper.
What Regulators Publishers and Platforms Must do Next
First, regulators must require clear labels on AI rewritten headlines. A user must know when an official feed shows an AI version instead of the original headline. Labels must be visible at first glance, not hidden in menus. This is a rule about truth in advertising. It is also a rule about civic safety.
Second, platforms must pause wide rollouts until they prove accuracy with independent audits. The audits must test real content across topics and languages. They must show error rates and the kinds of mistakes the models make. Platforms must publish those results. Publishers must be allowed to opt out if they do not want automated headlines on their links.
Third, publishers must demand control or better transparency agreements. Newsrooms should be able to map which links get AI titles. They should get a clear appeals path when a headline harms their brand. Google must offer a fix that returns the original headline when the AI text misleads. This is basic fairness for creators who deliver the reporting.
Fourth, regulators must update liability rules when official feeds alter public information. The law should make it clear who answers for false or misleading headlines published on a major platform. A single wrong headline can affect markets, votes, and reputations. That is not a small risk. Lawmakers must set guardrails for how AI is used in public-facing channels.
Finally, readers can act now. Switch off suspicious Discover cards. Send pictures with the publisher in case of a headline error. And insist that sites should display the original headline in case they change the headline. A policy memo cannot work as fast as public pressure. It has the capability of driving improved practice in the lab and safer releases.
This episode is a challenge of whether big tech can create tools that support the masses or tools that defraud the masses. Artificial intelligence can be helpful in cases where it helps editors to write better summaries. The fear of AI is that it will recreate and take credit away from reporters. Publishers must blow out. Regulators must intervene. The reader must not be duped by a feed that puts a higher price on the clicks than on truth.