Google Is Testing AI-Generated Headlines, but Things Aren’t Going Well.
Take a look at the beginning of this article. See that headline? If it’s different from the one you clicked to reach this page, congratulations: Google may have chosen you to participate in its latest AI experiment: rewriting news headlines for some users in Google Discover.
Evidence of these new efforts was first discovered by The Verge , as the update apparently affected writer Sean Hollister. Here’s what’s happening: when you swipe right on the home screen of a Pixel or Galaxy (or scroll down in the Google app on an iPhone, or open a new Chrome browser window with Google as your homepage), there’s now a chance that the article previews you see in Google Discover are actually generated by artificial intelligence, rather than reflecting headlines and/or descriptions hand-written by the authors and editors of those articles.
Sometimes AI headlines are simply clumsy or vague: one AI headline introduced another Verge article about specific Microsoft AI initiatives as “Microsoft Developers Embrace AI,” which doesn’t say much, especially in the current tech landscape.
AI headlines can’t be trusted
But even more dangerously, these headlines can distort the facts. In Hollister’s case, Google Discover gave him the headline “Steam Machine Price Revealed,” while the original Ars Technica article simply stated, “Valve’s Steam Machine Looks Like a Console, But Don’t Expect a Console Price.” Clicking the link leads to an article with quotes from a Valve developer implying that the upcoming PC-home console hybrid won’t have a subsidized price like most home consoles, which is quite different from the official price.
Another headline Hollister saw stated, “Qi2 Slows Down Older Pixel Models,” implying that using a Qi2 charger could reduce the phone’s performance. The original article simply stated that older Pixel models wouldn’t be able to fully utilize Qi2’s fast charging capabilities.
Sure, errors in headlines about consumer technology are likely to cause only momentary frustration or confusion, and perhaps even a missed opportunity to buy the best phone charger. But imagine if this misinformation concerned something more serious, like the Luigi Mangione case. Given previous attempts by other companies to aggregate news using AI, this is entirely possible .
Perhaps worst of all, these AI headlines seem to be able to cast aspersions on anyone, posing a risk of libel. PCGamer recently wrote a cheeky article about Baldur’s Gate 3 , reporting on gamers who discovered they could use the Polymorph and Dominate Beast spells to recruit child NPCs who, thanks to the laws of real-life Germany, cannot die. One can imagine how this would be useful in the game, but hey, it’s all fiction, right? Unfortunately, Google’s AI headline decided to change PCGamer’s original headline, “Child Labor Is Unbeatable,” to “BG3 Players Exploit Children.” Horrible.
What’s going on with the headlines about Google AI?
Hollister and I reached out to Google for comment and received the same response: the new headlines are part of a “small user experience experiment for a subset of Discover users” and a continuation of similar AI-powered previews introduced in Google Discover in October. These previews were short, AI-generated summaries of articles that users could expand to see more information (and even the AI-powered headline), but did not replace existing human-written headlines.
The new experiment “changes the layout of existing headlines to make it easier to digest topic information,” which appears to be code for the AI headlines, which are now placed at the top, where you’d expect real headlines to be. I personally didn’t participate in the user interface experiment, but Hollister reported that he couldn’t see the real headlines until he navigated to real articles.
How to tell if your Google Discover headline was written by AI
Clearly, this test has a number of problems. AI-generated headlines can distort the essence of news, as happened in the Hollister case, or make false accusations. And unfortunately, since they appear right where real headlines used to appear, readers may well believe they were approved by the article’s authors or editors. If a Discover headline seems suspicious, there are three ways to determine whether it was written by AI.
-
Google’s AI is obsessed with making headlines shorter. All the AI-generated headlines Hollister saw were four words or less, and while we at Lifehacker also favor brevity, I can tell you from experience that real journalists and editors tend to write headlines a bit longer.
-
It seems that not a single headline generated by Google’s AI capitalizes anything but the first word. This is a striking departure from the style guidelines of most websites. For example, at Lifehacker, we use the AP style, which capitalizes almost every word except articles like “the.”
-
You can click “Learn More” below the Discover preview and check for a tag indicating the article is “Created with AI, which may make mistakes.” Articles with real headlines won’t even have a “Learn More” button.
Unfortunately, there doesn’t seem to be an option to opt out of these AI headlines, as Google didn’t provide one when I asked, instead simply repeating that it’s a “small UI experiment.” This means that not everyone is seeing them yet, at least.
For me, as someone who used Google Discover frequently before switching to an iPhone, this is still a major drawback. It used to be a convenient way to find out important news without scrolling through social media or checking multiple homepages, but I can imagine having to delve into every headline to determine whether it’s accurate would make things significantly more difficult.
It’s also not good for journalists who rely on Google Discover to drive traffic and could bear the brunt of user dissatisfaction with inaccurate content from readers who don’t realize it was machine-generated. Currently, I think the latter option is more likely. But even if Google eventually addresses the issues with AI-generated headlines, it could still negatively impact traffic, potentially removing the incentive to click on ads, which is an integral part of any good headline. Google will continue to use external content to keep users on its platform, but the creators of that content will see it less often.
(Of course, as always, if you want the most accurate idea of what an article says, it’s best to read it carefully before forming an opinion.)