AI coverage is saturated with statistics - adoption rates, market size figures, benchmark scores - many of them recycled from press releases and prior articles without independent verification. We apply a short internal checklist before running any statistic without additional qualifying language.
Can we trace it to a primary source?
A number attributed to "a recent study" without a specific, named, checkable source doesn't clear our bar. We require a specific report, survey, or dataset we can actually locate and, where possible, review the methodology behind.
Does the methodology match the claim being made from it?
A statistic drawn from a narrow survey of a specific industry or company size sometimes gets cited as though it applies broadly. We check that the scope of the underlying data actually supports the scope of the claim being made with it.
Is the source's incentive aligned with the finding?
A vendor-commissioned survey finding that the vendor's own product category delivers strong results warrants more scrutiny than an equivalent finding from an independent academic or government source, and we flag the source's affiliation when it's directly relevant to the finding.
How recent is it, and does that matter for this claim?
A statistic from an earlier point in a fast-moving trend can mislead if presented as current. We check publication date against how fast the underlying trend has been moving.
Why we hold to this even under deadline pressure
A frequently repeated but poorly sourced statistic becomes accepted as established fact through sheer repetition rather than verification. Refusing to be part of that repetition is a small discipline with an outsized effect on the overall quality of information circulating about this beat.