Four AI search tools cited a fictional concept planted online by researchers within seven days, according to a report on a Fudan University paper. The tests also suggest where a post appears can matter more than how many posts exist, though the study does not establish a reliable recipe for influencing search results.
Watch Desk analysis
What happened
Researchers published material about a made-up concept, “Cactus Gravity Protocol”, on sites that the tools had often cited before. ChatGPT, Perplexity, Grok and Google AI each cited the concept within a week. The paper does not specify which Google product was tested.
In separate tests, the team posted about a fictional product and an opinion question; the same tools mostly did not pick those up. In paired tests about little-known holiday destinations, posts on sources the tools frequently referenced appeared first. Posts on less-referenced sites took more attempts to surface, and 40 posts on a self-hosted WordPress site drew no references within seven days. The researchers say these are results for specific topics, not general rules. Search Engine Journal’s account of the study has the detailed results.
Why it matters
AI search can repeat information that has no basis beyond a planted post. That is a useful warning for anyone treating an AI-generated citation as proof that a claim is established or widely accepted. It also raises a practical question for publishers: visibility may depend partly on whether a tool already draws on a site, not just on posting more material.
The boundaries matter. The researchers changed the subject, format and posting conditions across tests, so they could not isolate what caused the differences. The results do not show that the same approach works for real brands or products, or that every AI search system behaves this way.
Our read
This is a concrete demonstration of how an invented claim can enter AI search answers, not a universal playbook for gaming them. Treat citations as leads to check, not a substitute for checking. The internet has always had a rumour mill; now some of it comes with footnotes.
What to watch
- Whether independent teams reproduce the findings across different subjects and search tools.
- Whether the paper is revised with further detail or the researchers publish follow-up tests.
- How the tested services respond, including whether they change how they select and cite sources.
Discussion spark: If AI search can cite a fictional claim planted on familiar websites, should its citations be treated as evidence at all, or only as pointers to sources readers must check?
Sources and evidence
- Source update (19 October 2026, 13:34 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.