The share of math preprints on arXiv that acknowledge AI has risen rapidly, from 4% in April to 25% in August. This increase was not due to an influx of new AI-assisted authors: If we restrict to papers with at least one author who published regularly before 2023, the rate of AI acknowledgments stays within 1.5 percentage points of the overall figure each month.
| Month | Math preprints examined | Acknowledge AI use (any purpose) | Acknowledge AI use for substantial research | Share acknowledging AI use (%) | Share acknowledging AI use for substantial research (%) |
|---|---|---|---|---|---|
| 2025-01 | 3,127 | 11 | 0 | 0.35 | 0.00 |
| 2025-02 | 3,048 | 15 | 1 | 0.49 | 0.03 |
| 2025-03 | 3,536 | 8 | 2 | 0.23 | 0.06 |
| 2025-04 | 3,394 | 26 | 0 | 0.77 | 0.00 |
| 2025-05 | 3,384 | 27 | 0 | 0.80 | 0.00 |
| 2025-06 | 3,249 | 30 | 2 | 0.92 | 0.06 |
| 2025-07 | 3,585 | 31 | 2 | 0.86 | 0.06 |
| 2025-08 | 3,297 | 37 | 2 | 1.12 | 0.06 |
| 2025-09 | 3,783 | 65 | 6 | 1.72 | 0.16 |
| 2025-10 | 4,008 | 43 | 6 | 1.07 | 0.15 |
| 2025-11 | 3,678 | 50 | 7 | 1.36 | 0.19 |
| 2025-12 | 3,766 | 72 | 4 | 1.91 | 0.11 |
| 2026-01 | 3,618 | 70 | 12 | 1.93 | 0.33 |
| 2026-02 | 3,516 | 84 | 23 | 2.39 | 0.65 |
| 2026-03 | 4,290 | 158 | 20 | 3.68 | 0.47 |
| 2026-04 | 4,058 | 158 | 29 | 3.89 | 0.71 |
| 2026-05 | 4,474 | 297 | 75 | 6.64 | 1.68 |
| 2026-06 | 4,769 | 402 | 90 | 8.43 | 1.89 |
| 2026-07 | 5,199 | 785 | 182 | 15.10 | 3.50 |
| 2026-08 | 5,754 | 1,453 | 337 | 25.25 | 5.86 |
To track how mathematicians are using AI, we categorized AI use acknowledgments into six distinct types: substantial research, research assistance, writing, code & computation, literature review, and formalization. In August, 6% of all papers explicitly acknowledged that AI made a substantial research contribution. Papers with at least one author who published regularly before 2023 acknowledge AI for substantial research at roughly the same rate as the overall pool of papers in each month.
Epoch's work is free to use, distribute, and reproduce provided the source and authors are credited under the Creative Commons BY license.
Learn more about this graph
This graph is based on analysis of every math arXiv preprint from January 2025 to August 2026. We identify AI use acknowledgments in preprints by filtering for papers that contain an AI keyword and then asking GPT-5.6 Sol to classify any AI use acknowledgments, using Claude Fable 5 as a validator. The substantial research category consists of acknowledgments that credit AI with research contributions that are approximately at the level of a coauthor. A human review of 200 random acknowledgments that were classified as either substantial research or research assistance showed 93% human agreement with the classifications.
This graph is based on our AI Use in Math Research explorer. Read the full data and methodology here.
Assumptions and limitations
We measure voluntary disclosures of AI use in math preprints, so the trends reflect changes in disclosure behavior in addition to changes in AI use.
Judging the extent of an AI contribution to a research result from an AI use acknowledgment is difficult and subject to error. In particular, the boundary between substantial research and research assistance is fuzzy. We instructed the AI classifier to count acknowledgments in which the extent of AI contributions is vague or unclear (e.g., “the proof was found with the assistance of AI”) in the research assistance category. The instructions specify that to qualify as substantial research, an acknowledgment must describe the AI contribution as “essential,” “key,” “crucial,” “substantial,” or something similar. For each AI use classification, we store a passage from the paper justifying the classification. See the full methodology for more details.
Download this data
Explore this data
Our estimates of the number and types of AI use acknowledgments in math research papers submitted to arXiv, obtained by having AI models analyze the full text of preprints that mention AI.