The treadmill said 15.1 km/h. My Garmin said 12.3 km/h. Apparently, we’d been doing two completely different runs on the same machine.
I was there. I remember running it. I’m fairly sure my legs do too.
And that, dear reader, is the ridiculous little problem that has taken me a year and a half, more apps than I care to remember, and a rather expensive collection of experiments to solve.
From one watch to another… and another
I’ve been fighting this since April 2025. Yep.
From Life Fitness treadmills to a Mi Band 9, Apple Health and MapMyRun. From a chest-strap heart-rate monitor that worked beautifully on the treadmill screen to exported workouts that apparently forgot my heart existed.
The treadmill knew my speed and distance. The chest strap knew my heart rate. But every app had its own interpretation of the same session. Getting those measurements into one reliable workout became a surprisingly ambitious project for a man who’d merely gone out for a run.
And it didn’t end there either. This year, I graduated to the older, stubbornly offline treadmills at Fitness First. We tried iSmoothRun, more recording combinations, and eventually an Apple Watch Series 11. Roughly S$1,200 later, I returned the watch.
It was a perfectly lovely watch. I just wanted something rather exotic from it: an indoor run I could actually correct.
So we started looking at things to attach to shoes. Stryd, COROS POD 2, even NPE RUNN, which attaches to the treadmill itself. Then came the smart-shoe archaeology: Nike+, Adidas miCoach and various other attempts to convince footwear to become an information system.
Yes. We considered buying a computer for my shoe because my watch and the gym treadmill couldn’t agree on how far I’d run. Sisyphus, but with Bluetooth.
18 May. Two records. And a watch with opinions.
Running log · 18 May 2026
Two personal records
The run that mattered.
Two records, more than two decades after the ones I set at eighteen. And Garmin had its own version of both.
★ Personal record / 1 km
4:0015 km/h
Treadmill1.00 km
Garmin≈ 0.91 km
★ Personal record / 3.03 km
12:55elapsed
Actual average pace4:16/km
Garmin recorded4:48/km
A 32 sec/km disagreement. On the same run.
The first comparison is distance; the second is average pace. Both came from the same day’s treadmill sessions.
On 18 May 2026, I ran 1 km in 4:00, an average of 15 km/h. I also ran 3.03 km in 12:55, averaging about 4:16/km. Two personal records, surpassing what I’d achieved when I was eighteen. I was in my 40s. You can imagine what those runs meant.
Garmin’s recording of that very same one-kilometre effort? About 0.91 km. It also recorded 4:48/km for the 3.03 km run, rather than the 4:16/km I actually ran. That’s 32 seconds per kilometre missing from a personal record that meant a great deal to me. And in October, another treadmill sprint topped out at 15.1 km/h while Garmin gave me a maximum of 12.3.
Garmin was already the best place for me to keep the running data: pace, laps, heart rate and the rest. But I wanted the correction made there, not in yet another spreadsheet sitting alongside the original mistake. Once edited, the corrected Garmin activity should be the source of truth.
So I gave ChatGPT the keys
We built Garmin MCP, an integration that lets ChatGPT and other AI assistants work directly with Garmin Connect. One Garmin sign-in, a verification code when required, and the assistant can finally see the runs we’ve been arguing about.
At first, we could read activities and health data. That meant actual laps, heart-rate samples, personal records, effort and feeling, even activity photos where Garmin had them. We could also ask about sleep, HRV, stress, steps, weight and Body Battery, where the data existed. We could chart distance, duration, frequency and pace over time instead of squinting at screenshots and trying to remember whether Tuesday was better than Thursday.
Then we discovered the other half: writing back. Correct an activity’s supported details, log a manual run, import a FIT, GPX or TCX file, or have the assistant build a structured running or cycling workout. Preview it, approve it, and only then save it to Garmin. Changes require separate permission and confirmation; the assistant isn’t quietly rewriting your watch’s sensor history. Deletes get their own explicit confirmation, because we’ve spent rather too long collecting those runs to lose one to an enthusiastic assistant.
And the best bit? It’s free. Reading and writing, with separate permissions for changes. No subscription required. We can even bring in data from other sources, including Google Sheets, when a run needs a second opinion.
The treadmill photograph is still our favourite use case: show the console, compare it with Garmin, confirm the supported corrections. But this has become far more than a treadmill fix. The full feature set, feedback channel and updates live at garmin.geekist.co.
Come run with me
If your watch has ever told you a different story from the treadmill, I’d love to hear it. Tell me about the run that sent you down your own recording rabbit hole.
And please use the comments. Tell me what watch and treadmill you use, which figures go missing, and what you would change. Those reports are far more useful for shaping the next features than another list of things an app might do.
I’d rather be running. 🙂 Garmin MCP Your Garmin data, in conversation Your watch has recorded the details. Connect Garmin to your AI assistant and start asking better questions about them. What you can do 01 Explore activities, laps and daily health 02 Compare running trends and personal records 03 Preview and approve running or cycling workouts 04 Correct supported activities and import files Reading is free. Changes require separate permission and confirmation. Independently developed by Geekist. Not affiliated with or endorsed by Garmin.
Get more from every run.