Strava just quietly dropped one of its most consequential updates in years: 4.45 million activities have been purged from the platform’s leaderboards, the result of a machine learning system built to flag “improper” uploads.
Most of these deletions come from activities logged under the wrong sport, from car rides posing as bike commutes, to GPS glitches masquerading as record-setting sprints, to, in some cases, full-on fabricated performances.
The model was introduced in February without fanfare, but the numbers speak loudly. And for a platform where segment crowns and PRs still carry weight, it marks a serious shift.
“Leaderboards feel like they mean something again,” one user posted on Reddit, a sentiment echoed across forums where longtime Strava athletes have grown jaded by the ease of cheating.
Strava’s cleanup effort is part of a larger slate of updates aimed at returning the platform to its core promises: accurate data, real training, and honest competition. And if the message wasn’t clear before, it is now: Strava is done letting the fakes slide.
What Gets You Flagged?
Strava hasn’t publicly disclosed the full criteria its machine learning model uses to identify suspect uploads, but it’s trained to detect patterns across mislabeled activities.
Obvious tells include sustained car-level speeds recorded as runs or rides, missing heart rate data on efforts that seem too good to be true, or implausibly short times on known KOM/QOM segments.
And it’s not just about accidents.
While some flagged activities are clearly user error (forgetting to switch your watch from driving to biking, for example), others fall squarely into the “gaming the system” category.
This kind of manipulation isn’t new.
FakeMy.Run, a site that lets users generate realistic-looking GPS workout files with just a few clicks. Plot a route, set a pace, download the file, voilà, a custom .gpx that can be uploaded to Strava or any fitness platform.
Its creator, Arthur Bouffard, insisted it was almost satirical. “To challenge the culture shift,” he wrote. “To prove the good old saying that you shouldn’t trust what you see on the internet.”
But his app struck a nerve.
Because while most people won’t fake a marathon, some definitely will fake a segment attempt, or a weekly mileage badge.
Strava, for its part, responded to our request for comment on that piece with a clear line: “Accounts found violating the Terms of Service, including through manipulating or fabricating activity data, will be suspended.”
The latest 4.45 million deletions show they weren’t bluffing.
A Smarter Summer Update
Alongside its leaderboard sweep, Strava rolled out several new features aimed at making the platform more useful, particularly for paying subscribers.
The Routes tab in the Maps section now taps into Strava’s Global Heatmap, an enormous dataset drawn from billions of real user uploads, and pairs it with AI to suggest smarter paths. Instead of guessing which roads are runnable or rideable, users now get suggestions that reflect how people actually move on the ground.
In June, Strava will launch Tappable Points of Interest, which lets users tap any café, trailhead, or lookout on the map to instantly generate a route that includes it.
In July, Point-to-Point Routing will let users select a start and end point, then generate the best activity-specific route between them, again powered by the heatmap and AI.
Live Segments also got a boost. Strava has doubled the number of real-time segments you can see on your device mid-effort and added more comparison tools to help you track PRs, segment rankings, and pacing side-by-side.
None of these features directly combat fakes, but together they signal a platform trying to recalibrate around real-world use. While the tech has long leaned social, these tools feel built for the athlete, not the audience.
