Coverage
Disclosure statistics cover papers with primary arXiv category math.RT or math.RA from January 2026. Historical submission counts cover 2016–2026. Counts for all mathematics categories begin in January 2026. Submission statistics reuse the disclosure census for math.RT and math.RA over the same period. Other mathematics categories are collected separately. The mathematics data record collection dates by category and month.
Counting
Each arXiv identifier is counted once, by its primary category at collection and month of first submission (UTC). Cross-listings do not contribute additional counts. The mathematics total is the sum over primary math.* categories; papers cross-listed from other subjects are excluded.
Only completed months are published. The current month is excluded from all counts and downloads. Unavailable periods are distinguished from periods with zero papers. Historical comparisons use the same completed months of each year by default.
Classification
Declarations are identified automatically from PDF text and arXiv comments. Candidate passages are classified using OpenAI Codex. Supporting passages and source references are retained privately.
- Use declared
- An explicit statement of generative-AI assistance with the paper.
- Non-use declared
- An explicit statement of non-use, possibly restricted to part of the work. Papers declaring both use and non-use are classified under use.
- No declaration found
- No relevant statement was detected in the checked sources.
- Unclassified
- Text was unavailable or insufficient, or classification remains unresolved.
Percentages use all papers in the selected category and period, including unclassified papers. Papers without a detected declaration may still have used AI. Automated processing may miss or misclassify statements, particularly in scanned text or languages not covered adequately by the screening procedure.
References to AI research, proof assistants, or conventional mathematical software do not by themselves constitute declarations of generative-AI assistance.
Counts may change after revisions or category corrections. Paper records and supporting passages are retained privately. Only aggregate statistics are published.
Lean formalisation
The Lean count includes papers stating that at least some of their results or proofs have been formalised in Lean. Citations, proposed future work, and formalisation in other proof assistants are excluded. This statistic overlaps the disclosure categories. It follows the selected category and period and is excluded from the disclosure percentages and pie chart. The statements are classified automatically; linked code and its correctness are not independently verified. Incomplete checks are reported separately.
Data
Maintained by Gustavo Jasso. The website and processing software were developed with OpenAI Codex. Classification and extraction are automated and have not been independently audited.
Inspired by Louis-Simon Cyr’s statistics on AI disclosures in math.SG.
This website is not affiliated with arXiv.