How accurate is your running app's GPS or treadmill distance?
Distance error depends on where you are, and a treadmill's error comes from a different source than GPS entirely. An instrument validation study that ran eight GPS sport watches through matched test routes found errors from 0.9% to 4.1% (overestimating distance) on an open running track, worsening to a combined -3.5% to -8.9% (underestimating) across forest and city test areas grouped together (Gilgen-Ammann et al., 2020, JMIR mHealth and uHealth). A treadmill involves no satellite signal at all: a 2021 motion-capture study found belt speed itself running about 98% to 103% of the console's target during a stride, dipping further, by 7.3% to 16.6% below target, for a moment during the braking phase right after footstrike (Willwacher et al., 2021, Scientific Reports).
Error by setting
This table is Runked's own, built by lining up the two studies' figures and Apple's and Garmin's own support pages by setting; none of the four sources publish a table shaped like this on its own.
| Setting | Typical error in the cited source | Why | What to do |
|---|---|---|---|
| Open sky (clear view, no obstruction) | About 3 m (10 ft) per GPS point under clear sky, per Garmin's GPS accuracy FAQ. This is a position-accuracy figure; the rows below instead use cumulative distance error (MAPE), a separate measure. | Direct, unobstructed line of sight to satellites, with minimal signal reflection. | Baseline condition. The rows below show how far accuracy moves from here. |
| Running track (open stadium, 400 m) | 0.9% to 4.1% MAPE, with distance overestimated on the track per the study's Bland-Altman analysis (Gilgen-Ammann et al., 2020). | The study's authors suggest manufacturers may auto-correct for the underestimation expected in obstructed areas, which can overshoot into overestimation once the obstruction is gone. | Closest to reliable among the settings this study measured. Still expect a few percent of drift here too. |
| City (tall buildings) | Part of the same study's combined forest-and-urban figure: -3.5% to -8.9% MAPE, underestimated in seven of eight watches. The study did not separate city results from tree-cover results. | Garmin's FAQ describes an "urban canyon" effect: buildings block part of the sky and reflect signals before they reach the device. | Treat a city-route distance as an underestimate. Garmin's FAQ recommends calibrating in an open area away from buildings. |
| Trees (dense canopy) | Same combined range as the city row above: -3.5% to -8.9% MAPE, underestimated (Gilgen-Ammann et al., 2020). | Garmin's FAQ attributes this to dense foliage interfering with the signal, a different obstruction with the same net effect as tall buildings. | Same as the city row: expect underestimation under trees as well. |
| Treadmill | Belt speed averaging about 98% to 103% of the console's target during a stride, dipping to 7.3% below target on a low-cost consumer treadmill and 7.8% on a higher-cost one, during the braking phase right after footstrike (Willwacher et al., 2021). | No GPS involved. The number comes from the belt motor and control system reacting to the foot's impact and loading right after landing, a mechanical response with nothing to do with a satellite signal. | Garmin's indoor-tracking FAQ recommends calibrating via a Treadmill activity of at least 1 to 1.5 miles, or a GPS-tracked outdoor mile, with arms swinging naturally and without gripping the handrails. |
What the evidence says
Rahel Gilgen-Ammann, Theresa Schweizer and Thomas Wyss tested eight GPS-enabled sport watches, from Apple, Coros, Garmin, Polar and Suunto, in a 2020 instrument validation study published in JMIR mHealth and uHealth. One participant, a 26-year-old woman, walked, ran and cycled a fixed course through three test areas, urban, forest and an open running track, with each area's true reference distance (404 to 4,297 meters) measured by trundle wheel, producing 36 recorded distances per watch. Across all eight watches, overall mean absolute percentage error (MAPE) ranged from 3.2% (the Polar V800, the best performer) to 6.1% (the Suunto 9 Baro, the worst); only the study's three Polar watches kept an overall MAPE under 5%. Broken down by area, the study grouped forest and urban readings into one combined figure, where MAPE ran from -3.5% to -8.9%, with distance underestimated in seven of the eight watches, while the open running track alone posted MAPE of 0.9% to 4.1%, with every watch overestimating distance there instead, per the study's Bland-Altman analysis. The study's own authors offer an explanation for the reversal: manufacturers may auto-correct for the underestimation the algorithm expects in obstructed areas, which can then overshoot into overestimation once the obstruction is gone. The authors are explicit that their design used a single subject performing every activity, which limits how far these specific numbers generalize to runners of different body types or running form.
Steffen Willwacher and colleagues took a different approach in a 2021 study in Scientific Reports: they measured the treadmill belt itself. Seven participants (six men, one woman; 60.7 to 108.2 kg; ages 18 to 35) walked and ran at speeds from 1.0 to 4.5 meters per second on four treadmills: two ordinary, non-instrumented models, a lower-cost one (Sole F85) and a higher-cost one (h/p/cosmos Stellar med), plus two research-grade, force-instrumented treadmills built for gait analysis. Average belt speed across a stride varied from about 98% to 103% of the console's target speed. The dip was sharper during the braking phase right after footstrike, before push-off: 7.3% below target on the low-cost treadmill and 7.8% on the high-cost one, growing to 16.6% on the older of the two force-instrumented lab treadmills and 4.2% on the newer one. A higher body mass produced a bigger braking-phase dip, a statistically significant interaction, though the size of that effect depended on which treadmill and which speed. The authors' own limitations note that seven participants gave the study low statistical power to detect every effect, and that four treadmills, however deliberately varied, are not a representative sample of the entire treadmill market.
Apple's own support page on calibrating an Apple Watch instructs users to walk or run for about 20 minutes in "a flat, open outdoor area that offers good GPS reception and clear skies," stating this can improve the watch's distance, pace and calorie figures, including in situations where GPS is limited. Apple's page also says calibration should be repeated at each pace a runner trains at. Garmin's own GPS accuracy FAQ describes the same open-sky baseline in harder numbers: under clear sky conditions, it states, "Garmin devices typically record your position within about 3 meters (10 feet) of your actual location," before listing tall buildings, dense tree cover and canyons or mountainous terrain as the conditions that degrade it. A separate Garmin FAQ on indoor distance tracking recommends calibrating through a dedicated Treadmill activity of at least 1 mile (older models) or 1.5 miles (newer models), or a GPS-tracked outdoor run or walk of at least a mile, with the arm allowed to swing naturally, without holding the handrails.
How to apply this to your own tracking
A single run's distance reflects that route and that device on that day, one data point to read alongside the pattern in the table above. A route through a city block or under tree cover should read on the low side, per the combined obstructed-area figure above. The same route run on an open stadium track or a clear road should read a little high. Either direction is the satellite signal having a harder or easier time reaching the device that day. For a treadmill, the console's own displayed speed is the steadier number for pace, since a watch reading a treadmill run through its accelerometer is estimating stride and cadence from motion, a step removed from measuring the belt directly, and Willwacher et al.'s figures above show the belt itself is not perfectly steady either. Garmin's calibration steps above, an outdoor GPS mile or a dedicated Treadmill activity of 1 to 1.5 miles with natural arm swing, are the direct way to narrow the gap between a device's indoor estimate and its outdoor GPS number over time.
Compare this week's distance and pace numbers to your own last week, not to somebody else's route through a different neighborhood; in Runked the rank starts everyone at Bronze and only moves on your own runs. See how the running rank works.
Common mistakes
- Assuming a treadmill's console number is always right and the watch is wrong. Willwacher et al.'s braking-phase dips, 7.3% to 16.6% below target depending on the treadmill, show belt speed itself moves around during a run, a separate issue from the watch's own estimate.
- Expecting a park run under trees to match a stadium track run to the meter. Gilgen-Ammann et al.'s two categories point in opposite directions: down under obstruction, up on an open track.
- Holding the handrails during a treadmill calibration run. Garmin's indoor-tracking FAQ specifically calls for the arm to swing naturally during calibration, and says not to hold the handrails.
- Calibrating once and assuming it covers every pace. Apple's own support page says calibration should be repeated at each speed a runner trains at; a single session at one pace does not cover the rest.
What to do next
The Garmin and Apple Health sync guide covers exactly which sources a run's numbers can come from. And if the real question is whether to add a dedicated GPS watch on top of a phone, running app vs. GPS watch covers that choice directly.
Find out where you stand
Runked is free on the App Store. Connect your watch or track 3 runs, and meet your Runk.
Download on App StoreFrequently asked questions
Why does my running app show a different distance than the treadmill's own display?
Because a treadmill run does not use GPS at all. Garmin's own indoor-tracking support page addresses this with calibration steps: an outdoor GPS run or a Treadmill activity of a set distance, with the arm allowed to swing naturally. Separately, the belt itself is not perfectly steady: Willwacher and colleagues' 2021 study in Scientific Reports found belt speed running about 98% to 103% of the console's own target speed across a stride, with a bigger dip, 7.3% to 16.6% below target depending on the treadmill, during the braking phase right after footstrike. Two different sources of error, on two different devices, showing up as one mismatch on screen.
Is GPS more accurate outdoors in the open than under trees?
In the open, per one study. Gilgen-Ammann and colleagues' 2020 instrument validation study in JMIR mHealth and uHealth found mean absolute percentage error (MAPE) of just 0.9% to 4.1% on an open running track, all of it overestimating distance slightly per the study's Bland-Altman analysis, against a combined -3.5% to -8.9% MAPE for forest and urban areas together, underestimating distance in seven of the eight watches tested. The study grouped forest and urban results into one combined figure and did not report them separately, so it cannot say whether trees or buildings alone are worse; it only shows both are worse than open sky.
How can I get a more accurate GPS reading?
Apple's own support page recommends calibrating an Apple Watch with about 20 minutes of outdoor walking or running in a flat, open area with clear skies, at whatever pace is your normal training pace. Garmin's GPS accuracy FAQ adds a related step for any device: performing a GPS soak, waiting a few moments after the GPS-ready indicator appears before starting the activity, so the device can lock onto more satellites first. Neither company claims either step removes the error described above; both describe it only as improving results.
Does Runked correct for any of this GPS or treadmill error?
No. Runked records outdoor runs with your iPhone's GPS and syncs runs already saved to Garmin Connect or Apple Health; whatever distance and pace that source reported feeds everything downstream, including PRO's race-time predictions. It does not apply a correction of its own on top of a device's or app's tracking.