AI Governance Institute
← News
Research2026-10-29

Only 40% of AI-Cited Stats Are Verifiable, Trakkr Research Finds

What happened

Trakkr Research published Zombie stats: Where the numbers AI quotes come from, tracing 1,285 statistics cited by ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. Only 40% of the citations led back to a study that confirmed the same figure. Another 8% reached a source but with a changed number or altered context. The remaining 51% either pointed to an uncheckable intermediate source or produced a dead end with no verifiable origin. All five systems presented their citations with consistent, uniform confidence regardless of whether the underlying figure could be verified. The study covered a global sample and was not limited to any single subject area or industry.

Why it matters

  • ·Regulatory and investor submissions that include AI-sourced statistics may contain unverifiable or fabricated figures. Presenting false data in filings can trigger enforcement action under securities, consumer protection, or sector-specific disclosure rules, regardless of whether an AI tool was used.
  • ·Compliance teams that rely on AI assistants for policy research, benchmarking, or risk quantification face a systematic accuracy gap. Because all five tools studied present uncertain and verified citations with the same tone, standard human review may not catch errors without active source verification.
  • ·The NIST AI Risk Management Framework (AI RMF 1.0) and Playbook and the EU AI Act (Regulation (EU) 2024/1689) both require organizations to assess and manage AI output quality in high-stakes contexts. A documented 60% citation failure rate strengthens the case for mandatory source-verification steps before any AI-generated statistic enters a formal document.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Identify all internal workflows where AI assistants are used to produce statistics, benchmarks, or cited research, including board reports, regulatory filings, investor communications, and policy documents.
  • ☐Add a mandatory source-verification step to any AI-assisted drafting workflow: require that every statistic be traced to its original source before the document is finalized, not merely reviewed for plausibility.
  • ☐Update AI acceptable-use policies to state explicitly that AI-generated statistics must be independently verified before use in external communications, regulatory submissions, or legal proceedings.
  • ☐Brief legal and compliance reviewers on the Trakkr findings so they understand that confident-sounding AI citations may be unverifiable, and that the confidence of the citation is not a signal of its accuracy.
  • ☐Ask your AI governance committee to assess whether existing hallucination and output validation controls cover citation integrity specifically, and whether your AI model registry records which tools are approved for research use.

What to watch next

Regulators focused on AI-generated content accuracy are likely to cite research like this when issuing guidance or initiating enforcement. The EU AI Act (Regulation (EU) 2024/1689) already requires providers of high-risk systems to document output reliability, and enforcement by the EU AI Office is expanding. The prior incident in which fabricated statistics survived seven months in ChatGPT, reaching academic and investment research shows regulators and courts are already encountering this problem in practice. Compliance teams should watch for sector-specific guidance from financial and healthcare regulators on AI-assisted research workflows, where citation errors carry the highest liability exposure.

Related Coverage

Research2026-11-12

AI Chatbots Surfaced Fake Retail Sites in up to 15% of Shopping Queries

IP protection firm Podqi tested ChatGPT, Gemini, and Perplexity across 6,795 shopping-related queries covering 151 brands. All three chatbots surfaced risky URLs in a material share of responses. Risky sites included fake impersonator pages, knockoff retailers, and expired domains reregistered by third parties. Perplexity showed risky sites in 15 percent of answers, ChatGPT in 8 percent, and Gemini in 4 percent.

Research2026-11-15

Forescout AI Found 66 New IoT Flaws, Breaking Legacy Patch Timelines

Forescout researchers used an AI harness built on Claude models to scan open-source TCP/IP and IoT software stacks, finding 66 previously unknown vulnerabilities and rediscovering 33 known ones. The AI approach was more effective at detecting memory-safety defects than logic or design flaws. The findings signal that AI-accelerated vulnerability discovery is compressing the time between flaw identification and potential exploitation, creating pressure on existing patch governance programs.

Corporate Policy2026-11-14

Apple's Foundation Models Framework Triggers EU AI Act Downstream Obligations

The release includes Acceptable Use Requirements. For the AFM 3 Cloud model, Apple also published documentation under Article 53(1)(b) of the [EU AI Act (Regulation (EU) 2024/1689)](/policy/eu-ai-act) to support downstream providers. Organizations building or procuring apps on this framework must assess their own compliance obligations as AI system deployers under applicable law.