Daily Marketing News
AI Search Cites Fabricated Statistics in Peec AI Experiment, Raising Questions About Source Credibility
A reported experiment highlights a challenge for brands pursuing AI search visibility: a citation can help people discover content without establishing that its claims are accurate.
What the experiment reported
In an October 7 account published by Peec AI, researcher Metehan Yesilyurt described launching a website containing 50,000 fabricated statistics. He reported more than 10,000 ChatGPT referral sessions and over 330 referring domains after approximately seven months. He also said the invented figures appeared in an academic paper and a bank’s investment research.
According to Yesilyurt, ChatGPT continued retrieving and citing the site after its Google visibility fell. These are the author’s reported observations from one experiment, rather than an independent audit or a measurement of every AI search platform.
Read the original Peec AI account for the setup and supporting material.
Three metrics that should not be confused
The case brings several different measures into the same conversation. Keeping them separate is essential to interpreting the result.
- Referral sessions: visits attributed to a source. A session total is not a count of AI answers or citations.
- Referring domains: distinct websites linking to a destination. Multiple links can repeat the same unsupported claim.
- Evidence quality: whether a claim is supported by an identifiable source, appropriate methodology, and a conclusion the evidence actually warrants.
None of these measures can substitute for the others. A marketing dashboard that reports reach without examining evidence can reward content that is widely distributed but unreliable.
What this means for SEO and AI search visibility
For a brand, appearing in an AI answer is a discovery opportunity. It is also a representation of the business that deserves scrutiny. Teams should examine what an answer says, which page it references, and whether the page supports the wording used.
Traditional search performance remains useful context. However, a Google ranking report alone cannot tell a team how a particular AI answer represents its brand. AI citation monitoring should therefore complement existing search analytics. It should not be treated as a replacement for useful content, technical accessibility, or established measurement.
The practical question is how to make credible information easy to locate, interpret, and verify. The experiment does not establish a guaranteed recipe for earning citations, nor does it demonstrate that publishing more statistics will improve commercial outcomes.
A credible content workflow for marketing teams
Start with a real audience question
Choose a question your customers actually need answered. A focused topic gives the research a clear purpose and makes it easier to judge whether the resulting page is useful. Keep the scope narrow enough that a reader can understand exactly what the findings cover.
Give every number a source and a boundary
Place the original source near the claim. Include the data year, population, sample size, and method when relevant. Explain exclusions and limitations. For internal research, describe how the information was collected and who reviewed it before publication.
Make the conclusion easy to verify
Use descriptive headings and plain sentences. Distinguish a measured result from a forecast or an opinion. Avoid turning a narrow finding into a universal statement, and ensure that a reader following the source link can locate the supporting evidence.
Assign ownership after publication
Give important research pages an owner and a review date. Check sources for changes, update outdated figures, and publish corrections where needed. A visible publication date is useful, but it should not imply that old data has been refreshed when it has not.
Measure the quality of visibility
Build a small, repeatable set of customer questions to review across the AI platforms relevant to your audience. Record the platform, date, prompt, response, cited URL, and whether the response accurately reflects the source. Results may vary, so retain examples rather than treating a single answer as a stable ranking.
Pair that review with referral traffic and meaningful website actions, such as qualified inquiries or engaged visits. Investigate whether citations send readers to the right page and whether the answer preserves important qualifications. A misleading citation is a correction task, even when it generates traffic.
Douce’s perspective: build a source worth checking
For brands investing in AI visibility, editorial standards belong inside the growth strategy. Research, content production, distribution, and measurement need shared definitions of what counts as a supported claim.
A useful first step is to audit a handful of high-value pages: identify unsupported numbers, replace secondary references with original sources where possible, and add clear limitations. Then review how those pages are represented in AI answers. This creates a manageable improvement cycle grounded in both audience needs and evidence.
Explore Douce’s earlier analysis of how CMOs are approaching AI platform visibility, or contact Douce to discuss your brand’s content and search strategy.