The Day AI Defeated Google (As Its Own Owner)

A tale of dog probiotics, digital detectives, and the search giant’s algorithmic betrayal

It began, as many modern catastrophes do, with a dog.

Our Havanese — a creature whose primary contributions to society include looking adorable and developing mysterious itching episodes — needed probiotics. Not the fancy kind. Not the Instagram-influencer-approved kind. Just something to settle his stomach and stop him from scratching like he’s auditioning for a flea circus. We’d previously tried a German supplement from love2dogs.de, and the little guy ate it. Not with the enthusiasm of a Labrador discovering a forgotten sandwich, mind you, but he ate it. Progress.

So there we were, ready to make the purchase again. Simple, right?

AVG antivirus had other ideas.

My wife’s computer lit up like a Christmas tree of paranoia. “Threat secured,” it proclaimed, with all the smug confidence of a mall security guard who finally caught someone shoplifting. I laughed it off. My PC runs NOD32 — a different antivirus with different opinions — and it said nothing. Silence. Complete digital indifference.

But here’s the thing about being a computer nerd: we don’t just ignore digital alarm bells. We investigate them. We pull at threads until the entire sweater unravels into a pile of concerning yarn.

And boy, did this sweater come apart fast.

The Investigation Begins (Cue Noir Soundtrack)

The first red flag wasn’t a flag so much as a red carpet rolled out by someone who clearly hoped nobody would look too closely at the stains. Love2dogs.de is hosted on AWS — which sounds legitimate until you realize it’s the web hosting equivalent of renting a storage unit: anonymous, cheap, and perfect for people who don’t want questions asked.

WHOIS lookup? Completely redacted. The domain registration information reads like a witness protection profile. No names. No addresses. Just digital tumbleweeds rolling through empty fields where accountability should be.

The company behind the operation? G&V Investment s.r.o. — a Czech entity whose name practically screams “we are definitely in the dog supplement business and not at all primarily interested in moving money around.” Investment companies typically invest in things. Dogs, historically, have not been one of those things.

To their credit, there is contact information on the website. A full address. Phone numbers. All the trappings of legitimacy. The only problem? According to every customer who’s tried, those phone numbers are about as useful as a chocolate teapot. They ring. Nobody answers. The calls disappear into the void, presumably joining all those missing socks and half of every Tupperware lid.

TrustPilot: Where Dreams Go to Leave One-Star Reviews

Let’s talk about the reviews.

TrustPilot paints a picture so vivid you can almost smell the disappointment. The one-star reviews tell a story as old as e-commerce itself: customer pays money, customer receives nothing, customer’s emails vanish into the digital ether, customer begins questioning their life choices.

The common themes read like a Greek tragedy performed by people who just wanted supplements for their pets:

“Ordered three weeks ago. No product. No response. No refund.”

“Promised 1–3 day delivery. It’s been ten days. My dog has moved on to new digestive problems.”

“PayPal won’t help. The company won’t respond. My faith in humanity is at an all-time low.”

Meanwhile, the five-star reviews exhibit all the authenticity of a robot trying to pass a captcha test. Repetitive phrasing. Generic praise. The kind of enthusiasm that only algorithms can muster at 3 AM. Gemini — and we’ll get to Gemini’s heroic performance shortly — analyzed these and returned a verdict that surprised absolutely no one: probably auto-generated.

Enter Gemini: The AI That Actually Did Its Job

Here’s where we pause to appreciate something genuinely impressive.

When I asked Google’s Gemini to investigate Love2dogs.de’s claims of “4.8 stars on 7,800+ verified reviews,” it didn’t just nod along and regurgitate the marketing copy. It went to work. Actual work. The kind that involves checking sources, cross-referencing data, and having what can only be described as healthy skepticism.

Gemini found exactly ONE TrustPilot presence for the company. Not 7,800 reviews. Not even close. 1,273 reviews, averaging 3.9 out of 5 stars, with a full quarter of customers leaving negative feedback. The “4.8 rating” existed only in the website’s self-congratulatory imagination.

When I asked it to check for Google reviews — thinking maybe the company had cultivated a loyal following somewhere else — Gemini came back empty-handed but thorough. No Google Business presence. No Map reviews. And where the company did have local presence with actual reviews? Those reviews were, as the data showed, definitively negative.

This is AI doing what AI should do: synthesize information, fact-check claims, and deliver findings that help humans make better decisions. Gemini earned its keep that day. It cross-referenced multiple domains, analyzed review authenticity patterns, and delivered a report that would make any digital forensics specialist nod approvingly.

One AI model, given a clear task, systematically dismantling a facade built on false claims. Beautiful. Almost poetic.

And Then Google Search Absolutely Betrayed Everyone

If Gemini was the hero of this story, Google Search’s AI Overview feature was the villain who showed up wearing the hero’s costume and immediately started setting things on fire.

Remember all those facts we just uncovered? The fake review counts? The poor customer experiences? The phone numbers that apparently route to an alternate dimension?

I searched Google for reviews of this company. The AI Overview — that little box at the top of search results that Google insists on showing everyone whether they asked for it or not — cheerfully informed me:

“Love2Dogs generally receives very positive Google and Trustpilot reviews (around 4.8/5 stars), with customers praising quick delivery, excellent customer service, and noticeable improvements in their dogs’ skin, coat, joint health, and reduced scratching/licking…”

I read it twice. Then a third time, because surely I was misunderstanding something.

I was not misunderstanding.

Google’s AI Overview had looked at a website that openly claims to have 4.8-star ratings — a claim that anyone with access to Gemini (Google’s own AI!) can debunk in approximately thirty seconds — and decided to present those claims as verified fact. At the top of the search results. In a position of authority. With all the gravitas of a news anchor delivering the evening headlines.

Quick delivery? Twenty-five percent of reviewers would like a word.

Excellent customer service? The phones that no one answers might disagree.

The AI Overview didn’t just fail to verify. It actively amplified false information and positioned it above every other result, including legitimate competitors who presumably pay good money to advertise on Google and now have to compete with Google’s own AI vouching for questionable operations.

The Implications (Or: We Need to Talk About This)

Here’s what keeps me up at night about this.

Google AI Overview isn’t some beta feature hidden in a developer console. It’s front and center, the very first thing millions of people see when they search for something. It comes with Google’s implicit endorsement — the assumption that if Google is summarizing information this prominently, that information must be reasonably accurate.

But AI Overview doesn’t fact-check. It summarizes. And when the primary source of information is a website making claims about itself, AI Overview becomes an unwitting megaphone for whatever narrative the website wants to project.

The businesses who play by the rules — who maintain honest reviews, who answer their phones, who actually ship products — now have to compete with AI-generated summaries that give equal or greater weight to self-reported statistics. AdWords customers who pay for positioning are displayed below an AI summary that’s essentially free advertising for anyone willing to put impressive-sounding numbers on their website.

It’s not just a Google problem, either. Every search engine is racing to implement similar features. Bing has them. DuckDuckGo has them. The entire search industry has decided that what users want is quick answers, and they’re not wrong. Users do want quick answers. They just also want those answers to be true.

The Bright Side (Yes, There Is One)

Here’s the thing that makes this story not entirely depressing: the AI tools that performed the investigation worked brilliantly.

Gemini, when given a specific analytical task, absolutely delivered. It didn’t hallucinate review counts. It didn’t invent customer testimonials. It went to the actual sources, counted the actual reviews, computed the actual averages, and reported findings that were accurate, useful, and actionable.

VirusTotal’s AI-powered analysis flagged the website with one vendor marking it as malicious and another as suspicious — not overwhelming evidence, but signal in the noise.

The problem isn’t that AI can’t verify information. The problem is that AI Overview isn’t trying to verify. It’s trying to summarize, which is a fundamentally different task. One requires checking claims against reality. The other just requires finding claims and restating them more concisely.

When AI is deployed as an analytical tool — as a partner in investigation, as a cross-referencer of data — it can be genuinely remarkable. Gemini’s ability to check multiple domains, analyze review patterns, and synthesize findings into a coherent report saved me hours of manual work.

When AI is deployed as a summarization layer on top of unverified content, it becomes a liability. A confident, authoritative-sounding liability that sits at the top of search results and misleads people who have every reason to trust it.

What You Should Actually Do

If there’s a practical takeaway from this adventure, it’s this: don’t trust AI-generated summaries in search results. At least not yet. Maybe not ever, unless search engines fundamentally change how they approach verification.

Use AI tools like Gemini or Claude as investigative partners. Give them specific tasks: “Check if this company’s claimed ratings match reality.” “Find reviews across multiple platforms.” “Analyze whether these testimonials look authentic.” They’re remarkably good at that kind of work.

But when you search for something and see that confident little AI Overview box at the top? Treat it like you’d treat a stranger’s recommendation. Interesting, potentially useful, but absolutely requiring verification before you hand over your credit card.

Our Havanese is still itching, by the way. We found a different supplier. One that answers their phone. One whose reviews didn’t require an AI-powered forensic investigation to verify.

Sometimes the old ways are better.

The author primarily uses AI tools to make computers do things they probably shouldn’t. They maintain a blog at ai-404.medium.com, where the 404 error code feels increasingly appropriate given the state of AI-generated search results.