An SEO Published a Fake Google Update. It Ranked on Page One. That Should Terrify All of Us.

We keep coming back to this one because it quietly demolished something we wanted to believe about how search works in 2026.

Jon Goodey, an SEO who runs an AI-assisted content workflow, spotted a hallucination his tools had produced about a Google core update that did not exist. The sensible move would have been to delete it. Instead, he published it on LinkedIn. On purpose. To see whether anyone would notice.

Nobody did. Not the readers, not the other publishers, and not Google.

The article ended up ranking on the first page for "Google March update 2026" and got pulled into Google's AI Overviews as if it were settled fact. We want to walk through exactly what happened, why it worked, and what we'd actually do about it, because the lessons here are bigger than one cheeky experiment.

What actually happened

According to Search Engine Journal's write-up of the test, Goodey published a LinkedIn newsletter describing a "March 2026 Core Update" that had never been announced. There was no such update. The whole premise was invented, the product of an AI hallucination he chose not to correct.

Within a short window, the post was ranking on page one of Google for the obvious query. Then Google's AI Overview, the box at the top that's meant to summarise the best of the web, started repeating the fabricated update as though Google had confirmed its own announcement. Awkward.

It didn't stop there. Other sites picked it up and ran. Search Engine Journal notes that a site called TechBytes published a detailed piece inventing technical specifics for the fake update, including a "Gemini 4.0 Semantic Filter," a "Zero Information Gain" classification, and a "Discover 2.0 Engine." None of those things exist. All of them sound exactly like the kind of thing Google would name.

The big SEO outlets, the ones who report on this stuff for a living, didn't touch it. They know what a real update announcement looks like and where it comes from. The independent sites that amplified it never checked. They just matched the vibe and hit publish.

"Most readers don't fact check. AI overviews and search amplify misinformation. One article is echoed by the Internet with other sites repeating and embellishing on the original false information."

That's Goodey's own summary of the result, and it's hard to argue with when his fake update spread across multiple "independent" sources in days.

Why a made-up update sailed onto page one

We've turned this over plenty in our own campaigns, and we think three forces combined to make it possible. None of them are bugs, exactly. They're how the system is built.

LinkedIn lent it borrowed authority

Google treats LinkedIn as a high-trust domain. Huge user base, constant indexation, a real person with a real profile attached to the post. When the fabricated update went up there, it inherited all of that credibility before anyone read a word of it.

This is the same mechanism that's pushed Reddit threads to the top of so many queries. Platform reputation is being used as a stand-in for whether the specific claim is any good. The two are not the same thing, and this experiment is the cleanest proof we've seen. We dig into how this plays out for brands in our piece on defending your brand narrative across AI search.

AI Overviews pattern-match. They don't fact-check.

This is the heart of it. AI Overviews run on language models that predict the next likely chunk of text. They are extraordinary at sounding authoritative and genuinely poor at deciding whether something is true.

When Google's model pulled from Goodey's article, it wasn't weighing the claim against reality. It recognised the shape of a Google update announcement, the right format, the right terminology, the right structure, and treated the shape as the substance. It saw the outline of truth and assumed the rest.

The scale of that gap is now measurable. An analysis by the AI lab Oumi, published on the Oumi blog, ran AI Overviews against the SimpleQA benchmark of roughly 4,000 hard factual questions. They found that the Gemini 3 generation contained the correct answer about 91% of the time, which sounds reassuring until you read the next number: only around 39% of overviews were fully trustworthy, meaning both correct and actually supported by the sources cited. The rest were either wrong or unverifiable against their own citations. A model can be right most of the time and still be impossible to trust on any single answer, which is precisely the problem.

The web amplifies. It rarely verifies.

Once one source published the fake update, the next sites cited it as confirmation, and the one after that cited those. Each repetition made the fiction look more established. This is the echo chamber doing what it does, and there's academic weight behind why it sticks.

A study by researchers at Cornell, "Trustworthiness Evaluations of Search Results: The Impact of Rank and Misinformation", found that people click higher-ranked results more often but don't actually trust them more, and, more worryingly, that warning banners about unreliable sources backfired by reducing trust in accurate information rather than the misinformation. So the obvious fix, slapping a "this might be wrong" label on things, can make readers more cynical about the good stuff while the bad stuff sails on. That's the kind of finding that should make anyone designing trust systems sweat.

What this exposes about E-E-A-T

Google has spent years telling us that E-E-A-T, short for Experience, Expertise, Authoritativeness and Trustworthiness, sits at the centre of how it judges content. Its own guidance on creating helpful, reliable content is blunt about the hierarchy: "Of these aspects, trust is most important. The others contribute to trust."

And then a fabricated article about a non-existent algorithm update ranked on page one and got recited by AI Overviews. So let's be honest about what E-E-A-T is and isn't.

E-E-A-T is an aspiration, not a live truth detector. Google's systems can spot some signals that correlate with expertise, things like author credentials, site reputation and citation patterns. What they fundamentally cannot do is read a specific sentence and decide whether it's true. The quality rater guidelines describe what Google wants its algorithms to reward. Goodey's experiment shows what they actually reward. The distance between those two is the whole story.

It gets less comfortable. Back in January 2025, Google's president of global affairs, Kent Walker, told the European Commission the company would not adopt the fact-checking measures in the EU's Disinformation Code of Practice, arguing the approach "simply isn't appropriate or effective" for Google's services, as reported by Fox Business. So it isn't only that the algorithm can't fact-check. Google has explicitly said it won't be building dedicated systems to do it either. Our take: that's a defensible engineering position and a worrying public one at the same time, and businesses need to plan around it rather than wait for it to change.

The AI Overviews problem is bigger than one fake update

This is the part we find genuinely uncomfortable, and it has knock-on effects for every site trying to earn traffic.

AI Overviews are sold as a time-saver: we've read the web so you don't have to, here's the answer. When that answer is a fabrication, the feature has actively misled the user and removed the step where they might have caught it. Old-school blue links at least forced a click and a moment of judgement. The summary hands over a conclusion and hopes you don't look closer.

Most people don't look closer. Pew Research Center tracked the browsing behaviour of 900 US adults and found that when an AI summary appeared, users clicked a link to a website just 8% of the time, compared with 15% when no summary was shown. Clicks on the citations inside the summary itself happened on a measly 1% of visits. The same study, also covered by Search Engine Journal across nearly 69,000 queries, found AI summaries appeared on about 18% of searches in March 2025. So a feature that's wrong a meaningful share of the time is increasingly the only thing most searchers ever read, and almost nobody clicks through to check it.

For any business that relies on accurate information showing up in search, that's a live risk. If a competitor publishes a confident, false claim about your industry, that claim can land in an AI Overview as established fact before you even know it exists. We wrote about why this matters for measurable traffic in our breakdown of the drop in click-through rates from AI Overviews.

What we'd actually take away from this

Doom is easy. Useful is harder. Here's the practical version of what this experiment changes about how we work, in the order we'd prioritise it.

1. Treat original reporting as your moat. If AI can rank and amplify content that merely sounds right, then the only durable strategy is content that proves genuine, verifiable expertise. Share data you actually collected. Reference experiments you ran. Link to primary sources. Show your working. "Sounds plausible" was never enough, but now we have the receipt.
2. Be sceptical of industry news, including ours. We've shared SEO updates without forensically checking every line. Most practitioners have. The fix is a habit: before you change strategy off the back of something you read, ask what the primary source is, whether you can verify it independently, or whether you're just trusting that someone else did. If the only "source" is another blog citing a blog, you don't have a source.
3. Make your expertise visible, even though it won't save you alone. Real author bios with checkable credentials, citations to primary research, links to your actual results. This won't stop you being outranked by confident nonsense in the short term. It builds trust with the humans who read you, and that's the asset that compounds.
4. Don't outsource your thinking to an AI summary. Whether it's AI Overviews, ChatGPT or Gemini, these tools are brilliant for synthesis and terrible for verification. Every AI-generated claim is a starting point for research, never the conclusion.

There's also a defensive angle that's easy to miss. Google's spam policies documentation explicitly names "using generative AI tools or other similar tools to generate many pages without adding value for users" as scaled content abuse. The same system that amplified Goodey's fiction will eventually penalise sites that mass-produce that kind of content, which is exactly why we keep pushing clients toward fewer, deeper, genuinely useful pages. Our content pruning playbook covers how to clear out the thin pages that put you in the firing line.

How to fact-check an algorithm update before you panic

Since this whole saga started with a fake update, here's the checklist we run internally whenever a new "Google update" starts circulating. It takes about five minutes and would have caught Goodey's experiment instantly.

• Check Google's own channels first. Confirmed core and spam updates appear in the Google Search Status Dashboard and via the official Google Search Central accounts. No primary announcement, no confirmed update.
• See whether the established outlets covered it. If Search Engine Journal, Search Engine Land and Search Engine Roundtable are silent on a supposedly major update, that silence is the signal. They report fast and they report everything real.
• Trace the claim to its origin. If every article ultimately links back to a single LinkedIn post or one unfamiliar blog, you've found a rumour wearing a suit, not a confirmed event.
• Look at your own data. Real volatility shows up in your rankings and analytics, not just in headlines. We always cross-check against what we actually see in client accounts before we believe a word of it. If you want a second opinion on whether your traffic moved because of an update, our analysis of the March 2026 core update volatility is a useful reference point.

Why we're still betting on real expertise

We know how this reads. Google can't fact-check, AI amplifies fiction, E-E-A-T is aspirational, misinformation ranks. Grim.

But Goodey's experiment proved something hopeful too. The sites that didn't fall for it were the ones with actual editorial standards. They have processes. They have reputations built on being right rather than being first, and they protected those reputations by saying nothing about a story they couldn't verify.

Over time, the sites that embellished the fiction will lose trust, and the ones that held the line will gain it. Algorithms can be fooled today. Reader trust compounds quietly in the background, and it's the one ranking signal that can't be faked into existence. This is exactly the foundation we build on when we run an SEO programme for a client: real expertise, primary data, and content that holds up when someone actually checks it.

The experiment exposed a genuine weakness in how search works right now. The right response isn't to panic or to game it. It's to be the source people reach for when everything else turns out to be noise. If you want help building that kind of authority into your search and AI visibility, get in touch with our team and we'll tell you honestly where you stand.