Google Discover AI Breaks 3 Publisher Headlines

Update (April 2026): The headline-rewriting experiment continued through Q1 2026 and was followed by Google's late-December 2025 Discover update, which hit publishers with 90%+ traffic drops on many sites. We documented the wider Discover collapse in Google December Update Kills Discover Traffic. The AI-rewritten-headline behaviour described below has not been rolled back at the time of writing. The structural recommendations on multi-channel diversification still hold, only more so.

Sometime last week, Google started quietly testing something that's making publishers want to throw their laptops out the window. The company's Discover feed, that personalised news stream on millions of Android phones and Chrome new tabs, is now replacing carefully crafted headlines with AI-generated alternatives. These AI versions are typically four words or fewer, often vaguer than the originals, and sometimes flat-out wrong.

And I don't mean "slightly misleading" wrong. We're talking headlines that promise information the article doesn't contain, turn nuanced reporting into clickbait, and occasionally create implications so disturbing they could damage a publisher's reputation permanently.

Google confirmed to The Verge this is a "small UI experiment for a subset of Discover users" meant to help people "digest" content more easily. But when your AI turns a gaming article into "BG3 players exploit children," you've got a problem that goes well beyond user experience.

If you rely on Google Discover traffic for your WordPress site, this affects you right now.

What Google is testing right now

Google Discover has always shown publisher headlines alongside article previews. You write a headline, Google shows your headline, people decide whether to click. Simple contract between publisher and platform.

Not anymore. For a subset of users, Discover is now replacing your original headline with an AI-generated version, typically compressed to four words. The AI attempts to summarise your article into an ultra-short phrase. It does this without asking permission, without giving publishers any control, and often without any visible label showing the headline was rewritten.

Your carefully crafted, SEO-optimised, brand-consistent headline is invisible until someone taps "See more" on the preview. Until then, Google's four-word AI interpretation is all readers see.

According to Dataconomy, this started appearing in early December 2025. Multiple journalists and publishers have documented examples across different topics. The results range from merely annoying to genuinely harmful.

Three examples that went viral

The "child exploitation" disaster

PC Gamer published an article about a Baldur's Gate 3 game mechanic where players clone non-player character children to exploit a loophole. The headline was playful and nerdy: "'Child labor is unbeatable': Baldur's Gate 3 players discover how to build an army of unkillable kids through the power of polymorph and German media laws."

Google's AI shortened this to: "BG3 players exploit children."

Without context, without the playful tone, without any indication this is about a video game, that headline suggests something horrific. And it appeared with PC Gamer's name right underneath it.

The fake price announcement

Ars Technica wrote: "Valve's Steam Machine looks like a console, but don't expect it to be priced like one." The entire point was that Valve hadn't announced pricing yet, and when they do, it'll probably be expensive.

Google's AI version? "Steam Machine price revealed."

That's not a simplification. It's the opposite of what the article says. Anyone clicking expecting official pricing would be confused and probably blame Ars Technica, even though the problem is entirely Google's.

The technical lie

9to5Google published practical advice: "Don't buy a Qi2 25W wireless charger hoping for faster speeds, just get the 'slower' one instead." The article explains why the expensive charger doesn't actually charge Pixel phones faster.

Google's AI rewrote this as: "Qi2 slows down older Pixels."

That's not what the article says. At all. But anyone seeing that AI headline might avoid Qi2 chargers entirely, thinking they'll damage their phone. Misinformation, courtesy of Google's automation.

Why this matters for WordPress publishers

When you write a headline, you're making dozens of editorial choices: framing the story, setting the tone, managing reader expectations, representing your brand. Google's AI strips all of that away and replaces it with algorithmic interpretation. When the algorithm gets it wrong, the publisher takes the reputational damage.

The irony? Google's own Discover policies explicitly warn publishers against using clickbait and misleading headlines. Now Google's AI is doing exactly that, creating vague, sensationalist, sometimes false headlines, and attaching them to your content.

"Clickbait: Headlines that manipulate readers into clicking, using exaggerated language or withholding critical context."

Google, Discover content policies

Reading Google's own content policies while watching their AI break every rule in the document is an experience. I've been running websites since before Google Discover existed, and I've seen the company contradict itself before. But this one is bolder than most. They're warning publishers not to write misleading headlines while simultaneously generating misleading headlines with their own AI.

For WordPress sites, this is particularly concerning. We've seen Discover traffic surge past 50,000 daily views for sites producing timely, well-optimised content. But all of that traffic depends on people clicking your headline. When an AI compresses "WordPress 6.9 Caching Bug Crashes Servers Across UK" into "WordPress bug fixed," the urgency disappears and so do your clicks.

And here's what really bothers me: you can't control this. You can write the perfect headline, optimise it for search intent, and do everything right, and Google can still replace it with algorithmic nonsense before anyone sees it.

Writing AI-resilient headlines that survive

You can't stop Google from rewriting your headlines. But you can write headlines that still make sense even when an AI tries to compress them into four words.

Front-load the important information. Put the subject and action right at the start. "WordPress 6.9 Breaks Three Plugins" survives compression better than "Three Common Plugins Stop Working After Latest WordPress Update."

Use specificity. Numbers, version numbers, dates, and proper nouns are AI-resistant. "PHP 8.5 Speeds Up WordPress by 23%" is more resilient than "New PHP Version Improves WordPress Performance." Our PHP 8.5 + WordPress 6.9 benchmarks title worked because the specific numbers made it hard for AI to misrepresent.

Avoid clever wordplay. PC Gamer's headline about "child labor" being "unbeatable" worked perfectly in context. Strip away that context and you've got a disaster. Straightforward beats clever.

Keep it under 60 characters. A concise, specific headline is harder to ruin than a long, elaborate one.

Consider your featured image. Even if Discover mangles your title, your image can still convey the story. When we wrote about WordPress 6.9 breaking three plugins, the featured image communicated urgency visually, which helps when the headline gets mangled.

And for the love of all that's holy, don't rely on irony or sarcasm in your headlines anymore. Not in a world where AI is doing the interpreting.

Technical steps for WordPress sites

Check your metadata. Make sure your Open Graph og:title and schema headline properties match your actual H1 title. Inconsistent metadata gives AI more room to improvise. Our free Meta Tag Checker shows you exactly how Google and social platforms read your page's meta tags.

Use proper schema. BlogPosting or NewsArticle schema with full author and publisher information makes it harder for AI rewrites to obscure your editorial identity.

Monitor Search Console. Watch for sudden drops in CTR even when impressions stay stable. That's your sign that headlines might be getting mangled.

Maintain performance. Google favours fast-loading content with good Core Web Vitals in Discover. Our WordPress hosting includes built-in optimisations for Discover traffic, and the CDN handles global load times so you don't have to think about it.

When a post gets 50,000 Discover impressions in an afternoon (and it happens more often than you'd think), you need infrastructure that won't buckle. The last thing you want is to nail the headline, dodge the AI rewrite, and then have your site crash under the traffic spike. Our guide to the WordPress 6.9 caching bug shows what happens when hosting can't keep up.

Should you panic and rewrite everything?

No. Don't go nuclear and rewrite your entire archive because of this experiment.

Google's calling this a "small UI test" which suggests it's not rolled out to everyone. Most Discover users probably aren't seeing AI-rewritten headlines yet. And there's a decent chance Google will either kill this experiment entirely or modify it based on the brutal feedback they've received from publishers.

"AI Mode intelligently organises information with simple breakdowns and helpful links to explore more on the web."

Google, AI Mode in Search announcement

"Intelligently organises information" is a generous description when the same technology just told the world that PC Gamer publishes articles about exploiting children. That disconnect between Google's marketing language and the reality of what their AI produces is something every publisher needs to keep in mind. The technology is powerful. It's also reckless.

Apply these principles to new content going forward. Write headlines assuming an AI might compress them. Front-load the important bits, be specific rather than clever, and make sure your featured image carries some of the narrative load.

Watch your existing top performers in Discover. If posts that consistently generate thousands of impressions suddenly show CTR drops, that could indicate AI rewrites are affecting them.

Your best defence is producing content that solves real problems. Write about breaking news when you can add expertise. Cover security vulnerabilities with specific version numbers and clear remediation steps. Publish performance benchmarks with real data. That kind of content performs well in Discover regardless of headline shenanigans.

If you're producing that kind of content consistently, you need hosting that can handle the traffic when Discover does its thing. Our Turbo Hosting is built for sites that need to maintain performance under sudden traffic spikes. Google's sister site 365i Web Design covered how Google's Discover-only core update changed which sites get featured, and fast hosting is a consistent factor.

Focus on what you can control: headline structure, technical optimisation, content quality, and site performance. Those fundamentals matter regardless of what Google does with AI.

Frequently Asked Questions