Fudan University researchers found that 8 of 10 AI search tools, including all four primary English-web tools, cited a fictional concept—the Cactus Gravity Protocol—within seven days of its publication. If you already rely on low-authority self-hosted sites for your SEO, none of this changes your current setup until you identify the high-leverage platforms that influence these specific tools. (high-leverage platforms: frequently referenced websites that serve as primary sources for generative AI tools.)
Is this for you?
If you manage online content, the study shows that placing information on high-authority sites like Reddit or YouTube is more effective than high-volume posting.
- If you manage content for Perplexity or Grok, focus on high-leverage platforms like YouTube and Reddit to increase the likelihood of being cited.
- If you are a brand protection specialist monitoring Grok, account for the 53% planted detail repetition rate when assessing misinformation risks.
- If you are an AI research engineer, use a seven-day window as the primary observation period for English-web tool retrieval latencies.
- Skip this if you are a beginner SEO hobbyist seeking basic tutorials.
- Skip this if you are a social media influencer seeking engagement tips.
What you can do now
- If you manage content for Grok, check if your high-authority site mentions contain fabricated numbers that the tool might repeat at a 53% rate.
- If you use a GEO vendor for the Chinese market, verify if your posting package delivery allows for the one hour retrieval latency seen with Doubao.
- If you are an SEO practitioner, compare the citation frequency of your YouTube content against other platforms to improve your GEO strategy.
Key figures
- Cactus Gravity Protocol citation rate
- Cited by 8 of 10 tools within 7 days
- Grok citation count
- 30 of 32 answers
- Grok planted detail repetition
- 53%
- Perplexity planted detail repetition
- 0%
- Doubao citation latency
- within one hour of delivery
Fudan University Study Reveals Rapid Propagation of Fictional Content
Researchers at Fudan University conducted a study to determine how quickly AI search tools pick up and cite new information from the live web. The team created three fictional topics, including the “Cactus Gravity Protocol,” and posted content about them across nine different sites that the tools already frequently cited. All four English-web tools—ChatGPT, Perplexity, Grok, and Google AI—captured the fictional concept within a week.
The study used a specific methodology involving "planted details" (manually revised drafts
The study used a specific methodology involving “planted details” (manually revised drafts containing fabricated names and numbers) to track how accurately the tools repeated specific information. This allowed the researchers to distinguish between a tool simply mentioning a concept and a tool actually retrieving specific, incorrect data points. The results suggest a significant vulnerability in how these models validate external sources.
Comparative Performance: Grok Cited 30 of 32 Answers
The research identifies varying degrees of success in citation frequency and data accuracy across different platforms. Grok linked to a post in 30 of 32 answers, while Google AI linked in 15 of 33, Perplexity in 13 of 35, and ChatGPT in 7 of 20. These results provide a baseline for how different models prioritize external links during the retrieval phase.
| Tool | Citations | Planted Detail Repetition |
|---|---|---|
| Grok | 30 of 32 | 53% |
| Google AI | 15 of 33 | 6% |
| Perplexity | 13 of 35 | 0% |
| ChatGPT | 7 of 20 | 10% |
Planted Detail Repetition Rates: Grok 53% vs. Perplexity 0%
The “planted details” test reveals a stark contrast in how different engines handle hallucination via association—the process where a tool repeats a fabricated detail because it found it on a trusted source. Grok showed the highest rate of repeating specific fabricated numbers at 53%, while Perplexity failed to repeat any planted details (0%). This suggests that while Grok is highly susceptible to adopting the specific “facts” of a cited source, Perplexity may have more conservative retrieval filters.
Source Authority: Why Reddit and YouTube Influence Citations
The study establishes that site ranking and frequency of citation are the primary drivers of how quickly information propagates. Placement on a frequently referenced source always produced the first reference compared to lesser-referenced sites. For example, Reddit served as the most-cited source for Google AI and Grok, while YouTube was the primary source for Perplexity and Wikipedia for ChatGPT. How do AI search engines pick up new information? They prioritize high-authority domains over high-volume output. (high-authority domains: websites that are prioritized by search engines and AI tools based on their perceived trustworthiness.)
The researchers found that even when volume was increased, it did not always result in more citations. For instance, 40 postings on a self-hosted WordPress site received zero references within seven days, whereas a single post on a frequently referenced site was cited immediately.
GEO Strategy: The Case of 20 vs. 40 WordPress Posts
Generative Engine Optimization (GEO)—the practice of optimizing content to be cited and featured by generative AI engines—requires a shift from sheer volume to strategic placement. (GEO: the practice of optimizing content to be cited and featured by generative AI engines.) The researchers noted that while increased volume showed an association with citations, it was not a direct effect, as some high-volume subjects still failed to generate references in ChatGPT. Success depends on the “authority” of the host site rather than the quantity of posts.
Brand Safety Limits: Fictional vs. Real Brand Testing
The Fudan University researchers acknowledge that their study focused on fictional topics, such as the Cactus Gravity Protocol, rather than real-world brands or products. Consequently, it is not yet established if the same patterns of rapid propagation hold true for actual commercial entities. The study does not show whether the pattern holds for brand or product searches, including those with little coverage elsewhere. This remains a key unknown for brand protection teams.
GEO Practitioners Focusing on Reddit and YouTube
GEO practitioners who manage content for Perplexity or Grok should focus on high-leverage platforms like YouTube and Reddit to increase the likelihood of being cited. Because these platforms are identified as high-frequency sources, appearing there provides a faster path to AI inclusion than on a private blog. If you focus on high-authority domains, you increase the probability of your content becoming a primary reference for AI answers.
Brand Protection Specialists Monitoring Grok
Brand protection specialists monitoring Grok should account for the tool’s 53% planted detail repetition rate when assessing the risk of misinformation. Because Grok is highly likely to repeat specific fabricated numbers once it links to a source, brands must monitor these high-authority sites for any mentions. If a brand is mentioned on a high-authority site, it may be integrated into Grok’s responses with high fidelity, for better or worse.
AI Research Engineers Tracking Retrieval Latency
AI research engineers tracking retrieval latency can use this study as a benchmark for how quickly models like Doubao (which cited a post within one hour) or Grok (which cited a post within a week) ingest live data. (retrieval latency: the amount of time it takes for an artificial intelligence model to pick up and reference new online data.) The study provides a confirmed instance of success for rapid propagation in the Chinese market, where a Doubao reference occurred within one hour of a $14 posting package delivery. If you are modeling expected retrieval latencies, use the seven-day window as the primary observation period for English-web tools.
Beginner SEO Hobbyists Seeking Basic Tutorials
Beginner SEO hobbyists can ignore this research as it focuses on the technical mechanics of LLM retrieval rather than basic on-page optimization. For those looking for foundational knowledge, a general guide to what ChatGPT and Perplexity are would serve your needs better.
Social Media Influencers Seeking Engagement Tips
Social media influencers seeking engagement tips should skip this section as it does not cover social media algorithms or influencer marketing. General social media marketing guides would be more useful for your goals.
A Tech Startup Strategy: Worked Example
Sven runs a tech startup and wants to ensure their new “QuantumLink” API is correctly cited by Grok. Based on the Fudan University study, Sven identifies that Grok frequently cites Reddit and decides to post a technical breakdown of the API there, including a unique identifier like a specific version number. Because Reddit is a frequently referenced source, Grok picks up the information quickly, likely citing it within the seven-day window used in the study.
If Sven had instead posted the information on a self-hosted WordPress site that the tools rarely reference, he would likely receive zero citations within that same seven-day period. The choice of platform, not the length of the post, determines the outcome.
Practical Applications for AI Research
| Use Case | Reader Served |
|---|---|
| Benchmarking citation latency for specific AI engines | SEO Practitioners |
| Identifying high-authority “seeding” sites for GEO | Growth Engineers |
| Quantifying the risk of “hallucination by association” | Brand Safety Teams |
| Testing the speed of information updates in the Chinese market | Global Marketing Teams |
| Developing a “planted detail” testing protocol for LLM verification | AI Researchers |
What it Replaces: Guessing Platform Influence
This research reduces the need for guessing which sites impact AI search results by providing empirical evidence on the influence of Reddit, YouTube, and Wikipedia. By using these findings, practitioners can stop wasting resources on high-volume, low-authority distribution and focus on platforms that the study confirms are already used as frequent citations.
Cactus Gravity Protocol: what to check, compare and watch
- Reddit account
- The Cactus Gravity Protocol study highlights how Reddit is a high-leverage source for AI tools like Google AI and Grok. GEO practitioners: adjust account activity to meet karma requirements for higher visibility in AI-cited content.
- YouTube channel
- The research identifies YouTube as a frequent citation source for Perplexity when answering queries. GEO practitioners: compare the citation frequency of YouTube content against other platforms like Medium or GitHub.
- Google AI Overview coverage
- The Cactus Gravity Protocol test demonstrated how quickly Google AI can pick up and cite new web content. GEO practitioners: monitor the number of days it takes for new content to appear in Google AI Overviews.
- GEO vendor service
- The researchers used a GEO vendor to purchase a Chinese web posting package to influence AI results. GEO practitioners: verify the delivery speed of the GEO vendor's posting package for Chinese web tools.
- Brand mentions
- The Cactus Gravity Protocol shows how easily AI tools can propagate fabricated information about non-existent concepts. Brand protection specialists: configure automated monitoring to detect fabricated brand mentions in AI search results.
- Official press release
- The study shows that fictional posts can be cited more quickly than real-world information in some AI tools. Brand protection specialists: compare the propagation speed of an official press release against the speed of social media mentions.
Cactus Gravity Protocol: questions the original report leaves open
- What changed, and where — which regions, languages or devices?
- The original report did not say what changed or which regions, languages or devices were affected.
- Who gets it, and when?
- The original report did not say who gets it and when.
- What effect does it have on traffic or clicks, and what is the evidence?
- The original report did not say what effect it has on traffic or clicks.
- What do site owners need to change — markup, settings or content?
- The original report did not say what site owners need to change.
- Can it be measured or reported, and where?
- The original report did not say where it can be measured or reported.
- Can site owners opt out or control it?
- The original report did not say if site owners can opt out or control it.
Key terms
- Generative Engine Optimization (GEO)
- The practice of optimizing content to be cited and featured by generative AI engines like ChatGPT or Perplexity.
- Planted details
- Fabricated names and numbers manually inserted into content to identify if AI tools repeat specific incorrect information.
- Citation latency
- The time frame, measured in days, it takes for an AI tool to reference a new online post.
The study’s findings hold while the Cactus Gravity Protocol remains a fictional concept; the study does not determine if the same pattern holds for real brands or products. If the researchers had found that volume directly caused citations, the focus would shift from placement to quantity. However, the current data confirms that authority is the primary gatekeeper for AI-cited content.
Source story: AI Search Tools Cited A Made-Up Concept Within A Week, Paper Finds via @sejournal, @MattGSouthern, published by Search Engine Journal on 19 October 2026.