What we checked
We set out to answer three questions. What does Whoop tell its members about using their data to train AI, and where? When did that change? And do Whoop’s own figures for the size of its data hold together?
The first two come from Whoop’s own documents: its privacy policy, its terms of use, its legal policies page, its help pages and its blog, read as they are now and as the Internet Archive recorded them over the years. The third comes from Whoop’s published figures and its chief executive’s statements, checked against each other with a short script.
Whoop’s documents today
On November 4, 2026, we read the current versions of:
- the full privacy policy, dated August 3, 2026;
- the terms of use, dated August 3, 2026;
- the legal policies page, with Whoop’s privacy principles, its definitions and its consumer health data notice, also dated August 3, 2026;
- the help article on Whoop’s AI coach and the blog post on Whoop AI.
We kept a dated copy of each page’s text and screenshots of the passages the story quotes. The current terms of use and legal policies page are also preserved by the Internet Archive.
How the documents changed
The Internet Archive has copied Whoop’s documents many times. We took one copy a month of the privacy policy (from 2015), the legal policies page (from 2024) and the terms of use (from 2023), and used the 112 that hold the document’s text. Our script searched each for phrases such as “to train any AI Technology models,” “research and development” and “zero training,” and recorded the first copy containing each one. The readme lists every phrase.
A phrase search can miss reworded text, so we then read the copies on either side of every change the story reports. Three things limit what this can show:
- Monthly copies date a version, not a day. A change can have been made up to a month before the first copy that shows it. Where the documents print a “Last Updated” date, we give that date.
- The terms before August 2023 aren’t in our analysis. The Archive holds earlier copies; we didn’t analyze them, so the story says only what the terms said from 2023.
- One copy doesn’t fit. The Archive’s copy of the terms from April 1, 2026, is headed “Last Updated: October 9, 2025,” but it carries the license from before that date, without the training clause that every other copy of the October 9 version has. We don’t know why; it may be a version shown to some readers or a fault in the copy. It doesn’t change what the October 9 version says.
The blog post’s history comes the same way: the Archive’s full copies run from February 8 to June 13, 2026, and none contains the sentence about model training. There is no usable copy after June 13, so we can say only that the sentence was added after that date. The Archive has no copy of the current version, so we publish ours, saved on November 4, 2026, with screenshots of its date and of the sentence, in the downloads below.
The podcast
Will Ahmed’s words come from the All-In podcast interview published on October 29, 2026. We transcribed the episode, and our editor checked every passage the story quotes, and who said it, against the audio. The two clips in the story are trimmed from the episode’s audio, with nothing removed or rearranged inside them, and each links to the full episode at that point.
Whoop’s numbers
whoop_figures.csv lists every figure we found that Whoop or Ahmed has published about members, hours of data, wear time and Whoop’s research, with the exact wording and the source. check_figures.py does the arithmetic from that file:
- Hours of data. Whoop said “more than 24 billion hours” on March 31, 2026, and Ahmed “just over 50 billion” on October 29, 212 days later. Adding 26 billion hours in that time means 122.6 million hours of wear a day: 5.1 million people wearing a band 24 hours a day, every day. Whoop reported more than 2.5 million members in March and more than 3 million in July. Growing in a straight line through those figures, members wearing a band around the clock would have added about 15.5 billion hours; even 4 million members for the whole period would have added 20.4 billion, for a total of about 44 billion.
- Members. Whoop’s own figures agree with each other: a straight line from about 2 million in December 2025 to 3 million in July 2026 passes through 2.54 million on March 31, against the “over 2.5 million” Whoop reported that day. Ahmed’s “around 4 million” is what he expects at the end of the year.
- Wear time. Ahmed didn’t say how “23.6 hours a day” is measured. Whoop’s research page gives “99.1% Data captured on active days,” which counts only days a band is worn. The one published study of wear by Whoop’s scientists that we found counted days, not hours: of 11,914 members who joined in 2024, 4,351 (36.5%) wore the band every day, as the study defined it, through their first 12 weeks. That study included only members who still had data in their twelfth week.
What we didn’t do
- We didn’t use the Whoop app or open an account, so we haven’t seen its settings screens. The story relies on what Whoop’s documents and help pages say the settings do.
- We didn’t ask Whoop for comment.
- We didn’t test whether Whoop’s de-identified data could be traced back to people, and we don’t know what Whoop removes from it beyond its own definition.
- We didn’t republish Whoop’s privacy policy or terms. The phrase checks, with links to every Archive copy, are in policy_history.csv, and the Archive holds the current terms.
Downloads
| File | What it is | Size |
|---|---|---|
| whoop_figures.csv | Every figure Whoop or its chief executive has published about its members, its data and its research that we found: 33 rows, each with the exact wording and its source. | 6 KB |
| figures_check.txt | The arithmetic behind the story’s numbers section: member counts, the gap between 24 billion and 50 billion hours, and the wear-time figures. | 3 KB |
| policy_history.csv | One row per monthly Internet Archive copy of Whoop’s privacy policy, legal policies page and terms of use that we analysed (143 rows), with the phrases each contains. | 36 KB |
| policy_history_summary.txt | For each phrase, the first Archive copy that contains it and the copy before it that doesn’t. | 4 KB |
| whoop-blog-post-2026-11-04.txt | Our copy of Whoop’s blog post “How WHOOP AI guidance turns your health data into personalized coaching,” as it appeared on November 4, 2026: the article’s text, including the sentence about model training. | 9 KB |
| whoop-blog-post-date-2026-11-04.png | Screenshot of the post’s title and its April 30, 2026 date, taken November 4, 2026. | 71 KB |
| whoop-blog-post-faq-2026-11-04.png | Screenshot of the post’s answer that “WHOOP uses anonymized member data for AI model training,” taken November 4, 2026. | 74 KB |
| readme.md | What each file and column means, what each phrase check matches, our sources, and how to rerun everything. | 10 KB |
| data.zip | All of the data files above and the readme, in one download. | 13 KB |
| code.zip | The Python scripts that check the figures, trace the documents through the Archive and build this package, with the readme and licences. | 15 KB |
SHA-256 checksums
To confirm a download is the file we published, compare its SHA-256 checksum with this list.
3999319ab96fedb2e5f26e5e0af91466c514b7a46140666b9e64b473e7d327e9whoop_figures.csv1fee31e98e0b9c67b25d20f74c44dec5540584445eac3b484a2d2b59e729896efigures_check.txt61660f568f3fa79e92b4008d732fd1721b5e71e975c03753ba82961d5a520a50policy_history.csv08cb4e07eff8c8ec4a3c793cd17c6fb5d191556ba7c372f7f3fab29d3441968apolicy_history_summary.txteaf4d0862ec42aa6ff969784bb464a2a3dc1eacb795746003cddeac1e4cb2098whoop-blog-post-2026-11-04.txt271d6df2ebc6c5ffd3cc5ee6a589ef5f365c4395b85f2d681b35380733418aa2whoop-blog-post-date-2026-11-04.pngcd65517aecc178461141ce2e2ba986d270abfe6123197bb643611d9befcb56ebwhoop-blog-post-faq-2026-11-04.png25e9a0277b62e54b06fe3134079902f4b0ed831fa5adb3c9fc66899c5d61d76breadme.mdf1d279cf23af381e7db453b7cb1b0e3f73f4a6fd7acc87ae026f767871a114dcdata.zip49eb15be3f53446848bbbb1de1bb4b25039a1c847b89a6720155c1883c6ee07dcode.zip