Google Tests AI-Generated Headlines in Search Results – Publishers React

Google is testing artificial intelligence-generated headlines in its traditional search results – the familiar “ten blue links” – replacing headlines originally written by journalists with AI-created versions. The experiment, first reported by The Verge, has sparked a backlash from news organizations and SEO professionals, who view the practice as a breach of an implicit contract that has underpinned the relationship between Google and publishers for over 25 years: that the content appearing in search results reflects what the publisher originally published.

Sean Hollister, senior editor at The Verge, discovered that Google had reduced his headline “I used the ‘cheat on everything’ AI tool and it didn’t facilitate me cheat on anything” to just five words: “‘Cheat on everything’ AI tool.” The result, Hollister reported, made it appear as though The Verge was recommending a product it actually cautioned against. Another headline was rewritten as “Copilot Changes: Marketing Teams at it Again,” a phrase that did not appear in the original article.

Hollister likened the practice to “a bookstore ripping the covers off the books it displays and changing the titles.”

Google acknowledges the tests are a “small and limited” experiment not yet approved for wider rollout. According to three company spokespeople cited by The Verge, the AI identifies the content of a page and generates a headline it deems “helpful and relevant to the user’s query,” with the goal of improving the match between search and result and increasing engagement.

The experiment isn’t limited to news outlets; it affects websites generally. Google has not disclosed how many users are seeing the rewritten headlines, but confirmed it runs “tens of thousands” of live traffic experiments simultaneously. This builds on earlier tests, reported by The Verge in January, where Google began replacing headlines in Google Discover, its mobile content feed. Google confirmed in February that this feature had moved beyond the testing phase.

The examples documented by The Verge illustrate the concerns of publishers. In Google Discover, AI-generated headlines have already produced problematic results, including “BG3 players exploit children” (referencing a Baldur’s Gate 3 trick unrelated to child exploitation) and “Steam Machine price revealed” (when no price had been announced). The AI appears to prioritize condensing complex headlines to four words or less, often losing nuance, tone, and editorial context. An ironic headline, when reduced to a literal phrase, can invert its meaning. A headline discouraging a product, shortened to appear as a recommendation, can compromise the credibility of the publisher.

For SEO professionals, the impact is twofold. First, SEO headline strategies become less relevant if Google is rewriting titles regardless. Second, performance data becomes unreliable: if click-through rates decline, is it due to a poor headline or a Google alteration without notification?

Louisa Frahm, Director of SEO at ESPN, expressed the stakes on LinkedIn: “After 10+ years in news SEO, I’ve reach to believe the headline is the most prominent element to attract readers in short windows of time. If that view is altered and facts are misrepresented, long-term public trust will be eroded.”

The experiment comes as publishers are already experiencing organic traffic declines of 30-60% due to AI Overviews, Google’s feature that directly answers queries without requiring clicks. Now, in addition to losing traffic, publishers are losing control over how their content is presented to potential readers.

Google stated that if it were to launch a feature based on this experiment, it would not be a generative AI model creating the headlines. Though, the company did not explain how the alternative would function. The Verge noted that a similar “experiment” in Discover ultimately became a permanent feature, casting doubt on Google’s assurances.