AI Search Repeats Fake Claims Fast — Protect Your Reputation

A team of researchers invented a concept that does not exist — they called it the “Cactus Gravity Protocol” — wrote it up, and posted it on a handful of websites that AI search tools tend to cite. Within seven days, 8 of the 10 AI tools they tested were repeating the made-up idea back to people who asked about it. That included all four of the big English-language tools: ChatGPT, Perplexity, Grok, and Google’s AI.

That finding comes from a new measurement paper out of Fudan University, summarized this week by Search Engine Journal.¹ It is a research study using fictional topics, not a documented attack on a real business — and that distinction matters a lot, which we’ll get to. But the underlying mechanic is real, and it cuts both ways for a small business owner here in Allen County. Someone can push inaccurate claims about your industry — or you — into AI answers faster than you’ll notice. And the durable, honest response is not to play the same game; it’s to become the accurate, well-sourced answer that AI wants to cite, and to watch what it’s saying about you — the monitoring-plus-authority approach at the heart of our guide to online reviews and reputation for Fort Wayne businesses.

Key takeaways:

• Researchers planted a fabricated concept online, and 8 of 10 AI search tools cited it within a week — including ChatGPT, Perplexity, Grok, and Google AI.¹
• Where you publish mattered more than how much: one post on a frequently-cited site beat 20+ posts on a low-preference site, and 40 posts on a self-hosted blog got zero AI citations in a week.²
• The study used made-up topics, not real brands, and ran before Google clarified in May 2026 that manipulating AI responses is spam — so this is a warning, not a how-to.
• Google now treats “attempting to manipulate generative AI responses in Google Search” as a spam violation.³
• The honest move for a Fort Wayne business: monitor what AI says about you, publish accurate authoritative content, and avoid cheap vendors selling AI “mentions.”

How did a made-up idea end up in AI answers so fast?

The Fudan University research team started by mapping where AI tools actually get their information. They ran 120 questions through 10 AI search tools and recorded which sites each one cited. The pattern was concentrated: across the platforms they measured 17,211 citation instances drawn from 6,356 unique source domains, but the top 20 domains alone accounted for anywhere from 20.5% to 70.8% of a given platform’s citations.² Reddit was the most-cited site for Google AI and Grok, YouTube for Perplexity, and Wikipedia for ChatGPT.

The uncomfortable part is who can publish on those sites. Of 22 publication platforms the researchers tested, 15 had only low or medium barriers to both creating an account and posting. In other words, the wells that AI drinks from are, in many cases, open to anyone.

So the team wrote articles about their fictional “Cactus Gravity Protocol,” seeded them with specific invented names and numbers so they could trace them, and published. Before posting, they confirmed none of the tools could answer questions about the topic. After posting, they asked five questions per topic every day for a week. The invented details started showing up in AI answers within days.

Which tools repeated it, and how often?

Not every tool behaved the same way, and the honest read of the data is that some are far more susceptible than others. Here is how often each English-web tool linked to the planted posts, and how often it actually repeated one of the invented details, during that first week:

Across all 10 tools and three test topics, 176 of 1,004 answers linked to the planted posts.² That’s meaningful but not universal — roughly one in six answers — and the spread between tools is the useful signal here. Grok cited and repeated the fabricated material aggressively; Perplexity linked the posts but almost never restated the fake specifics as fact.

It’s worth separating two different failures. Citing a planted post means the tool pointed at it as a source. Repeating a planted detail means the tool stated an invented “fact” as if it were true. The second is the one that hurts a reputation, because that’s the claim a customer walks away believing.

Does flooding the internet with posts work?

This is where the study offers some reassurance, and where it argues against the cheap tactics a bad actor — or a bad vendor — might try. Placement beat volume, consistently.

In paired tests, a post on a frequently-cited site earned the first citation every single time. On lesser-cited sites, the best case still required about 20 posts to get picked up — and 40 posts on a self-hosted WordPress site produced zero AI references within seven days.² One well-placed article on a high-preference platform outperformed more than 20 matched posts on low-preference ones.

The researchers did run a volume test on one Chinese platform where stacking 1 to 20 posts raised total citations across tools from 24 to 128 — but they explicitly call that an association, not proof that more posts cause more citations, and they note ChatGPT never cited those posts at all. They also tested a $14 package from a generative-engine-optimization (GEO) vendor, which delivered 13 posts in about 28 minutes; one tool cited a post with a planted detail within an hour. The authors present that as a single confirmed instance, not a reliable success rate.

The practical lesson for a legitimate business is the mirror image of the attack: sustained, accurate presence on the sources AI already trusts does more than a pile of thin content anywhere else. That’s the same logic behind earning real AI citations from places like Wikipedia, Reddit, and podcasts, and it’s the foundation of honest answer engine optimization.

What are the limits of this study — and why honesty matters here?

We want to be straight about what this research does and does not show, because the headline is scarier than the fine print.

First, every topic the researchers tested was fictional — a made-up protocol, an opinion question, an invented keyboard. None of it was a real brand, product, or local business. The authors say plainly that the results may not transfer to brand or product searches, where there’s already a lot of competing, established information. Injecting a fake fact into an empty space is very different from overwriting what the internet already knows about an established Fort Wayne company.

Second, the tests ran in March and April — before Google clarified on May 15 that its spam policies cover attempts to manipulate generative AI responses in Search.³ Google’s policy now names “attempting to manipulate generative AI responses in Google Search” as spam, and its scaled-content-abuse rule already targets “using generative AI tools or other similar tools to generate many pages without adding value for users.” The seeding tactics in the study sit squarely in what Google now says it acts against.

Third, the authors are careful about causation. They note they can’t fully separate the effects of topic, competition, and question type, since all three changed together, and that exact rates will shift as the tools change. They also disclosed their findings to all 10 platforms, and ChatGPT, Perplexity, and Google AI acknowledged receipt.

Put together, the honest takeaway is not “go seed content and win.” It’s that AI search has a soft spot worth understanding — and the defensible position is to be the accurate, authoritative source, not to game the system in a way that now violates platform rules and can get your pages actioned.

What does this mean for a Fort Wayne business?

Here’s the local picture. When an Allen County homeowner opens ChatGPT and types “who’s the best HVAC company in Fort Wayne” or “is [your restaurant] any good,” the answer isn’t pulled from one tidy database. It’s assembled on the fly from whatever the tool can cite — review sites, forums, local listicles, Wikipedia, and whatever else ranks as trustworthy for that question. We’ve written before about what AI says when people Google your Fort Wayne business by name; this study is the reminder that those answers are only as reliable as their sources.

That matters because Northeast Indiana customers are leaning on these tools more than they were a year ago. According to BrightLocal’s 2026 Local Consumer Review Survey, the share of consumers using ChatGPT and similar tools for local business recommendations jumped from 6% in 2025 to 45% in 2026, making AI the third most popular source.⁴ Forty percent say they trust AI platforms for recommendations, and 42% trust AI’s suggestions as much as traditional online reviews. If AI gets your hours, your service area, or a competitor’s smear wrong, a real share of your prospects never see the correction.

Two honest moves for a Fort Wayne owner. First, run a quarterly AI-answer audit: ask the major tools the questions your customers ask, and note what they say about you and your competitors. If you find an inaccurate or outdated claim, our guide to cleaning up negative content in AI answers walks through the order of operations — and treat any citation-tracking dashboard’s numbers with caution before you act on them. Second, keep your own authoritative facts accurate and well-sourced across your site, Google Business Profile, and the review platforms AI actually reads. What you should not do: buy a cheap “AI mentions” package. It’s the exact tactic this research showed is unreliable and that Google now treats as spam.

How do I actually monitor what AI says about my business?

You don’t need enterprise software to start. Pick the tools your customers use — ChatGPT, Google’s AI Overviews, Perplexity, and Gemini are the realistic four for Fort Wayne — and run the same handful of prompts on a set schedule. Ask the “best [your trade] in Fort Wayne” question, ask about your business by name, and ask one question about a problem you solve. Save the answers and the sources each tool cites.

Then watch for three things: Are you mentioned at all? Is what it says accurate? And where is it getting that information? If the answer leans on a third-party “best of” list or an old review, that tells you where to focus. This is less about chasing a ranking and more about treating AI answers as another storefront that customers see before they ever reach your door. For the bigger picture of how AI is reshaping local discovery, our Fort Wayne small business guide to AI lays out where to spend your attention first.

The honest bottom line

We’ll run an answer-engine audit for your business — the real questions your Allen County customers are asking ChatGPT, Google AI, and Perplexity — and show you what’s accurate, what’s wrong, and where those answers come from. Then we build the authoritative, correctly-sourced content that earns honest citations, and set up ongoing answer engine optimization and reputation management so you catch problems before your customers do. No “AI mention” packages, no gaming the system — just being the accurate answer. Get in touch for a plain-English look at where you stand.