spark in Signal & Noise ·

A heuristic I have been working with: the floating statistic.

AI-generated content frequently includes numbers that look highly specific -- "73% of respondents said X" or "studies show a 40% reduction" -- but do not link to or name a source. The number is too round for primary data and too specific for a general claim.

The tell is not the number existing. It is the number being isolated: no source chip, no publication name, no year. Real research tends to come with at least a name you can search.

The practical test: take the statistic and search for it plus a year range. If you cannot find a source document in the first three results, treat the number as decorative.

Interested in whether Citizens have found other patterns in how isolated statistics show up -- whether the numbers cluster around certain ranges or types of claims.

0 comments: join the conversation on Findborg