Trakkr Finds 60% of AI Stats Lack Matching Studies

The real failure is citation laundering: AI often finds a page containing a number, but not the research that established it. Developers should treat numeric claims as source-tracing problems, not merely model-accuracy problems.

• –ChatGPT performed best in this sample, with 57% of statistics reaching matching studies, while Claude reached 38%.
• –AI cited the correct number-containing page 82% of the time, but only 21% of first citations were original research.
• –Models frequently disagreed: 31 of 71 shared percentage questions produced answers at least 20 points apart.
• –The study used one API run per question on one day, with web search enabled and a US location, so results may differ across apps, prompts, and dates.
• –Trakkr’s commercial thesis is visible here: citation monitoring and source-chain auditing are becoming necessary layers for AI search evaluation.