Every number here has a source

Which sources we use, and how we grade them

Every figure on this site links to the source it comes from, and each is recorded in our claim ledger with the source, the date we checked it and a grade.

Grade A

Probability samples, experiments and official records, such as Census Bureau surveys and companies' SEC filings.

Grade B

Weighted panels, usage logs, payroll data, executive surveys and working papers.

Grade C

Opt-in and vendor surveys, news, single cases, and our own estimates.

Each figure also carries a mark:

Primary data

Read directly from the source that collected it.

Cited secondary

Reported by someone other than the people who collected it, or from data we can't inspect.

Estimate

A forecast, a planning range or our own calculation that goes beyond adding up reported figures.

TODO for Robb: confirm these one-line definitions of the three marks; they paraphrase how the existing pages use them.

When sources disagree

We don't pick a winner quietly. When two good sources disagree, the page shows both, says what each measures and explains the gap, as on Do companies that pay for AI actually use it? and Daily use: two different questions.

Revisions

Trackers restate their history when a source revises it. Each page shows the date its sources were last checked; the pages in Spend and tokens were checked on October 25, 2026.

TODO for Robb: decide how revisions are logged publicly (a changelog per tracker, or a version number in each data file).

How to cite

See How to cite How We Use AI.

Conflicts of interest

Read our conflict of interest statement. Where a source has a stake in its own figures, the page that uses them says so.