AI & Automation Policy
Last reviewed: September 16, 2026
NumberResearch.org uses automation, including AI systems, in producing its datasets and research. We disclose this openly: we will not pretend the work is done entirely by hand, and we will not hand automation the final word.
The division of labor
Automated systems collect, normalize, validate, and draft. Material changes and research conclusions are reviewed by a named editor.
- Automation does: retrieve source files on schedule, store snapshots, parse and normalize records, compute diffs and counts, run validation checks, flag anomalies, and draft text that describes data.
- Humans do: review sensitive changes before publication, review and approve research conclusions, decide what to publish, and take responsibility for errors. The review step is described in our Methodology.
What we commit to
- No hidden automation. We do not present automated output as purely human work.
- No unreviewed conclusions. A research conclusion — any claim that goes beyond restating source data — is not published without human review and approval, consistent with the byline on our work: prepared by the NumberResearch.org Data Desk, reviewed and approved by the Publisher (see Editorial Standards).
- Same sourcing bar. Automated drafting does not lower the sourcing standard. Whether a sentence was drafted by software or a person, its factual claims must trace to Tier A sources.
- Accountability stays human. If an automated step produces an error that we publish, that is our error, and it is handled under the Corrections Policy. "The system did it" is not a defense we use.
Why we work this way
Numbering data changes daily across many source files. Automation is the only honest way to keep snapshots complete and diffs accurate at that scale; unaided manual transcription would introduce more errors, not fewer. Human review is concentrated where judgment matters: sensitive changes and conclusions.
Questions
Questions about how automation was used on any specific page: research@numberresearch.org.