AI
An experiment with fake statistics exposes how AI search can launder false evidence
A Peec AI researcher says a site containing 50,000 fabricated statistics was largely buried by Google but continued to be retrieved by ChatGPT, generating referrals and citations, including in academic and financial material. The account argues that AI retrieval systems reward relevance, specificity and apparent freshness without routinely checking truth, allowing false numbers to gain credibility through repeated citation. The experiment is self-reported and its results are not independently audited, but it highlights a serious verification gap in AI-assisted research.
Editorial summary · Publisher article
What matters
- The site cost about $50 to create and reportedly received more than 72,000 ChatGPT-User requests in one recent month.
- The author distinguishes retrieval from Google ranking: a page can be poorly ranked by Google yet remain available to an AI system’s search and grounding process.
- The proposed safeguards include source and year beside every statistic, independent verification, accessible pages and monitoring for AI-bot retrieval.
Why false numbers can look authoritative
The author argues that models often seek statistics because a precise number is easy to quote and appears evidentiary. Retrieval systems may score relevance and apparent authority rather than independently verify truth, while later citations can make an invented number look increasingly credible.
Context & caveats
- The traffic, citation and visibility figures come from the author’s own analytics and commercial research context; the packet does not provide independent replication.
Why this made the edition
The author documents a large controlled-in-spirit experiment with fake statistics and reports retrieval, citation and referral observations, offering a consequential warning about AI search. The findings are self-reported and not independently verified, but the packet contains enough detail for a caveated analysis.
Source material