Google's AI Overviews Are a Mess: Here's What Happened and What to Do

If you've been on the internet at all in the past week, you've probably seen the screenshots. Someone asks Google "how to make cheese stick to pizza" and the AI Overview—Google's fancy new AI-generated answer that appears right at the top of search results—suggests adding a quarter cup of non-toxic glue. Another user asks about geology and the AI cheerfully recommends jumping off the Golden Gate Bridge for a better view. I'm not making this up.

Google rolled out AI Overviews to hundreds of millions of users on May 14, 2026, and within hours, the internet did what the internet does best: it found every single way to break it. By May 16, the company was scrambling to pull back some of the most egregiously wrong answers, but the damage was already done. The question on everyone's mind is: can we trust Google Search anymore?

I've been following this rollout closely, partly because I'm a tech nerd who cares about this stuff, but mostly because I use Google Search every single day—probably hundreds of times. And honestly? This feels like a bigger deal than most people realize. Let me walk through what happened, why it matters, and what you can actually do about it.

The Glue Incident and Other Horrors

The most viral example came from a user who searched for "how to make cheese stick to pizza better." Google's AI Overview pulled from a Reddit comment—not a recipe site, not a cooking blog, but a Reddit comment—that jokingly suggested adding glue. The AI treated it as a legitimate instruction. It then served that up to millions of people as authoritative advice.

This isn't just a funny meme. Think about the implications. Someone with a food allergy could follow AI-generated advice that's completely wrong. A parent could be told to give their child a dangerous home remedy. A student could be taught incorrect historical facts. The stakes are genuinely high.

Other highlights from the AI Overview disaster include: telling a user that cats have been on the moon (they haven't), claiming that President Obama was born in Kenya (he wasn't), and suggesting that you can cure depression by drinking bleach (you absolutely cannot). Each of these was generated by the AI pulling from unreliable sources—forum posts, satirical articles, and straight-up misinformation—without any kind of quality filter.

How Did Google Let This Happen?

This is the part that really gets me. Google has been working on AI for over a decade. They have some of the brightest minds in the world. They own the largest search engine on the planet. And yet, they rolled out a feature that was clearly not ready.

I think the answer is panic. Ever since OpenAI launched ChatGPT in late 2022, Google has been in a defensive crouch. The company that once defined innovation was suddenly seen as the old guard, the dinosaur that got caught sleeping. Microsoft was integrating GPT into Bing. People were talking about the "death of search." Google needed to show it was still relevant.

So they rushed. Google CEO Sundar Pichai personally approved the accelerated timeline for AI Overviews, according to internal reports from The Verge. The team was told to launch by Google I/O in May, regardless of whether the system was fully ready. And now we're seeing the consequences of that decision.

The irony is that Google actually has a lot of the pieces to make this work. They have a massive dataset from years of Search Quality Raters. They have a powerful language model in Gemini. But they skipped the part where you rigorously test for edge cases and safety. They optimized for speed over accuracy, and users are paying the price.

What This Means for Regular Users

Here's where I get practical. If you're someone who just wants to find information online—whether it's a recipe, a news article, or a fact-check—how do you navigate this mess?

First, stop trusting the AI Overview by default. I know it's tempting because it's right there at the top of the page, formatted nicely with a little sparkle icon. But as we've seen, it could be completely wrong. Treat it like a suggestion from a well-meaning but clueless friend—interesting, but verify everything.

Second, learn how to turn it off. There's no official toggle in Google's settings (yet), but you can add &udm=14 to the end of any search URL to bypass the AI Overview and go straight to the old-style 10 blue links. I've set this as a custom search engine in my browser so I never have to see the AI stuff unless I want to.

Third, diversify your search tools. I've personally started using DuckDuckGo more. They've committed to not using AI-generated answers in search results. Also, Perplexity.ai has been doing AI search for a while, but they cite sources more transparently. And honestly? Sometimes I just go straight to Wikipedia. It's not perfect, but at least it's written by humans who have to cite their sources.

The Bigger Problem: AI Hallucinations

What happened with Google's AI Overviews isn't unique to Google. Every major language model hallucinates—that's the technical term for when AI confidently makes stuff up. ChatGPT does it. Claude does it. Gemini does it. The difference is that those tools are clearly labeled as experimental. You know going in that you can't fully trust them.

Google's sin was presenting AI-generated answers as if they were authoritative. The AI Overview box looks like a search result. It has the same design language. It appears in the same spot where Google used to show curated knowledge panels. The average user has no way of knowing that this information was generated by a statistical model that doesn't actually understand truth or falsehood—it just predicts the next most likely word.

Until this changes, we're all playing a game of trust roulette every time we search. Is this answer real? Is it hallucinated? Is it from a reputable source or a Reddit thread? Google made the wrong bet, and now we're the ones paying the price.

What Google Needs to Do Now

I'll give credit where it's due: Google did acknowledge the problem. On May 17, a spokesperson told reporters that they were working on fixes and had already removed some of the most egregious responses. But that's not enough.

What Google needs to do is three things. First, implement a quality filter that prevents the AI from pulling from unreliable sources like Reddit, Twitter, or satire sites. Second, add a clear disclaimer to every AI Overview that says "This answer was generated by AI and may be inaccurate." Third, give users the option to opt out entirely. Right now, you can't do that without workarounds.

If Google doesn't do this, they risk eroding the trust that's made them the dominant search engine for over two decades. And in search, trust is everything. Once users start going elsewhere, it's really hard to win them back.

I'm not writing off Google Search entirely. I think they'll fix this eventually. But for now, I'm treating every AI Overview with extreme skepticism. And honestly? That's probably the smartest thing any of us can do.