Google has started replacing news headlines in its traditional search results with AI-generated alternatives, the company confirmed to The Verge this week. Three Google spokespeople described the practice as a "small and narrow" experiment, but refused to say how many queries or publishers are affected. No label tells users when a headline has been rewritten.
The most cited example: The Verge's headline "I used the 'cheat on everything' AI tool and it didn't help me cheat on anything" got trimmed to five words. As Search Engine Land reported, the result read simply "'Cheat on everything' AI tool," which, as Verge senior editor Sean Hollister pointed out, made it sound like the publication was endorsing the product rather than panning it. Another Verge article was rewritten to "Copilot Changes: Marketing Teams at it Again," phrasing the publication never used.
We've seen this before
If the "small experiment" framing sounds familiar, it should. Google ran an identical playbook with its Discover feed. 9to5Google reported in January that what started as a "small UI experiment" in late 2025 quietly graduated to a permanent feature. Google's new language: it "performs well for user satisfaction." The accuracy problems didn't go away. A 9to5Google article about Qi2 charging speeds became "Qi2 slows older Pixels," which was flatly false. An Ars Technica piece explicitly avoiding pricing speculation became "Steam Machine price revealed." A PCMag article about a drone policy was rewritten to say the US had reversed a ban, directly contradicting the story's content.
As Nieman Lab put it in January: Google's pattern is to trial AI integration as an experiment, minimize backlash over accuracy, then expand and cement the change. The AI Overviews "eat rocks" debacle followed this arc too. None of this is subtle.
The bookstore analogy
Hollister's comparison stings: "This is like a bookstore ripping the covers off the books it puts on display and changing their titles," which, harsh as it is, lands pretty close to what's happening. Publishers invest real editorial effort in headlines. They balance accuracy, tone, search visibility, and reader expectations. Google's system apparently scans the page and decides what would make a better title for a given query, which is a polite way of saying it optimizes for clicks on Google's terms, not the publisher's.
The company says this extends beyond news sites too, though it won't specify which categories or how broadly. Louise Frahm, SEO director at ESPN, responded on LinkedIn that headlines are the most prominent element for attracting readers in timely windows, and that misrepresented facts will compromise long-term audience trust. She's right, but there's a harder question underneath that one: if Google controls how your headline appears and your headline is how readers judge your credibility, who actually owns the relationship with your audience?
So what happens next?
Google says any broader launch "may not rely on generative AI," which is a strange thing to promise given that the current test plainly does. The company didn't explain what non-generative alternative it had in mind. Given the Discover precedent, publishers aren't optimistic that "experiment" means temporary. The Verge noted a similar Discover test became permanent in roughly two months.
Multiple Verge staffers spotted rewritten headlines over several months before the practice was reported, meaning this ran unannounced for some time. There's no opt-out mechanism for publishers and no disclosure for users. Google still sends the majority of referral traffic to news sites, which makes complaining about the company's behavior a uniquely uncomfortable exercise. You don't yell at the landlord when you can't afford to move.
For now, publishers can monitor their headlines in search results and document distortions. The News Media Alliance has pushed for transparency and opt-out tools. Whether Google responds to that pressure or simply waits for the news cycle to move on, as it did with Discover, is probably the only question that matters.
