Google Is Testing AI-Generated Headlines, and It's Not Going Well

Take a look at the top of this article. See that headline? If it looks different than what you clicked on to get to this page, congratulations: Google might have chosen you to participate in its latest AI experiment: rewriting news headlines for some users in Google Discover.

Evidence of the new effort was first spotted by The Verge, as it seems writer Sean Hollister was affected by the update. Here's what's going on: When you swipe right on your Pixel or Galaxy home screen (or scroll down in the Google app on iPhone, or open up a new Chrome browser window with Google as your homepage), there's now a chance the article previews you'll see from Google Discover were actually generated by AI, rather than mirroring the headlines and/or descriptions handwritten by those articles' actual authors and editors.

Sometimes, these AI headlines are just clunky or vague—one AI headline introduced another Verge story about specific AI initiatives within Microsoft as "Microsoft developers using AI," which doesn't tell you much, especially in the current tech landscape.

You can't trust AI headlines

But more dangerously, these headlines can also get the facts of the story wrong. In Hollister's case, his Google Discover fed him a headline saying "Steam Machine price revealed," whereas the original article from Ars Technica simply said "Valve's Steam Machine looks like a console, but don't expect it to be priced like one." Clicking through leads to an article with quotes from a Valve designer hinting that the upcoming PC/home console hybrid won't have a subsidized price like most home consoles, which is not at all the same thing as an official price reveal.

Another headline Hollister saw said "Qi2 slows older Pixels," which implies using a Qi2 charger on your phone could hurt its performance. The original article simply said that older pixels won't be able to use the full extent of a Qi2 charger's fast-charging.

Granted, mistakes with consumer tech headlines will probably only cause some momentary disappointment or confusion, or maybe a missed opportunity to buy the best charger for your phone. But imagine that misinformation applied to a story about something more serious, like the Luigi Mangione case. Considering previous attempts other companies have made to summarize the news with AI, it's hardly unlikely.

Perhaps worst of all, it also seems these AI headlines can throw shade when it wasn't intended, introducing a risk of libel. Recently, PCGamer wrote a cheeky story about Baldur's Gate 3, covering gamers who discovered that they can use the Polymorph and Dominate Beast spells to recruit child NPCs to their cause who, thanks to real-world German laws, can't die. You can imagine how that would be useful in a game, and hey, it's all fiction, right? Unfortunately, Google's AI headline chose to change PCGamer's original "Child labor is unbeatable" into "BG3 players exploit children." Yikes.

What's going on with these Google AI headlines?

Both Hollister and I reached out to Google for comment, and were given the same response: The new headlines are part of a "small UI experiment for a subset of Discover users," and follow up on similar AI previews introduced into Google Discover in October. Those previews featured short AI summaries of articles that users could expand to see more information (and even an AI headline), but didn't outright replace existing, author-written headlines.

The new experiment "changes the placement of existing headlines to make topic details easier to digest," which seems to be code for the AI headlines now being placed up-top, where you would expect the real headlines to be. I'm personally not part of the UI experiment, but Hollister reported he wasn't able to see the actual headlines until he clicked through to the real articles.

How to tell if that Google Discover headline was written by AI

Obviously, there's a number of problems with this test. The AI headlines could misreport the news, as they already have in Hollister's case, or make false accusations. And unfortunately, since they're right where actual headlines have been shown in the past, it's totally understandable for a reader to think they were approved by the articles' authors or editors. If a Discover headline looks fishy to you, there are three ways to identify whether it was written by AI.

• Google's AI is obsessed with making headlines shorter. All of the AI headlines Hollister saw were four words or less, and while we like to be concise here at Lifehacker too, I can say from experience that actual journalists and editors usually write headlines that are a bit longer than that.

• None of Google's AI headlines seem to capitalize anything but the first word. That's a stark difference from most websites' style guides. At Lifehacker, for instance, we use A.P. style, which capitalizes most words expect for articles like "the."

• You can tap "See more" under the Discover preview and check for a tag saying that it was "Generated with AI, which can make mistakes." Articles using actual headlines won't even have a "See more" button.

Unfortunately, there does not seem to be a way to opt out of these AI headlines, as Google did not provide me with one when I asked, instead simply reiterating that this is a "small UI experiment." That means not everyone is seeing these for now., at least

As someone who made frequent use of Google Discover back before I moved to an iPhone, that's still a major bummer. In the past, it's been a convenient way to catch up on stories that were relevant to me without having to scroll social media or check multiple homepages, but I can imagine that having to scrutinize every headline to know whether or not it's real will make things a lot rougher.

It's also not great for journalists, who both rely on Google Discover for traffic, and could take the brunt of user ire about inaccurate headlines from readers who don't realize a machine created them. As it is, I think the latter is the more likely outcome. But even if Google eventually works out the kinks with AI headlines, they could still hurt web traffic, potentially removing the incentive to click that is part of all good headline writing. Google will continue to use outside content to keep people on its platform, but the people behind that content will get fewer eyes on it.

(Of course, as always, if you want to get the most accurate idea of what an article says, it's best to read it thoroughly before forming an opinion.)

View the full article