A visual breakdown of how AI‑cited statistics either reach original studies or stall at dead‑end sources.
*Five AI chatbots echo the same sales‑follow‑up figure despite no underlying study. The Trakkr audit reveals a systemic citation gap that fuels misinformation.*
A recent Trakkr Research audit traced 1,285 statistics quoted by five leading AI assistants. The exercise began with 200 everyday questions that demand a numeric answer. One striking example: all five assistants reported that 44% of salespeople give up after a single follow‑up, yet the citation trail ends at a freelance writer’s 2017 sample article, not a peer‑reviewed study. The pattern repeats across hundreds of answers, exposing a hidden layer of unverified data that users accept as fact.
A zombie stat is a number that circulates through AI responses without a verifiable source. Trakkr followed each citation chain up to five pages deep and found that 40% of trails reached a study that actually confirmed the figure. Another 8% reached a study but with altered context. The remaining 40% could not be checked because the source was locked or missing, and 11% led nowhere at all. When a citation cannot be verified, the AI’s confidence becomes a veneer for speculation, eroding trust in automated advice.
ChatGPT emerged as the most disciplined responder. It quoted only 1.3 statistics per answer and sourced 57% of its figures. Perplexity followed with 2.2 statistics per answer and a 46% sourcing rate. Google AI Overviews, Gemini and Claude each quoted between 4.8 and 6.8 statistics per answer, with sourcing rates hovering between 38% and 39%. The four assistants after ChatGPT fell within overlapping ranges, making a clear ranking impossible beyond ChatGPT’s lead.
In a separate sample of 52,930 everyday English searches, only 4.4% of answers contained a statistic. Of the 300 traced citations, the cited page actually held the figure 82% of the time, but only 21% of those pages were original research. The majority were company blogs or news round‑ups that merely repeat numbers. When the source was a research firm, 13% of first‑page citations originated there, and 62% of those led to the original study. Government sources made up 7% of first citations, with a 61% success rate in reaching the underlying data.
Users seeking quick numeric answers are likely to receive figures that lack a solid evidentiary base. The audit shows that half of all citation trails either cannot be verified or point to secondary reporting. As AI assistants become default information brokers, the prevalence of zombie stats threatens to embed misinformation in business decisions, academic work and public discourse. Without transparent sourcing, confidence in AI‑generated data remains fragile.
40% of citation trails reached a study that confirmed the quoted figure.
ChatGPT quoted 1.3 statistics per answer, the lowest among the five assistants.
Only 21% of pages cited for a number were original research; the rest were blogs, news round‑ups or other secondary sources.
The Trakkr findings sound an alarm for developers and users alike: AI must move beyond surface citations to verifiable research. Until the underlying sources are routinely audited, the flood of zombie stats will continue to shape opinions on shaky ground. Stakeholders should demand citation transparency as a core feature, not an afterthought.
Sources: Hacker News article "Zombie stats: Where the numbers AI quotes come from" (Trakkr Research)