Most runners who finish a marathon wearing a GPS watch seem to have run more than 26.2 miles. At the 2026 London Marathon, watches for 571 runners recorded a median distance of 42,722 meters against an official course of 42,195 meters. A small slice of that gap is intentional; every certified marathon course carries a built-in safety margin of about 42 meters. The runner's own path and the watch's own error account for the rest, roughly 485 meters, and no research study has figured out how much each one contributes.
How a marathon's distance is set
A person riding a bicycle decides how long a marathon is, without any satellites involved. World Athletics, the sport's global governing body, and the Association of International Marathons and Distance Races publish the rulebook for measurment rides. This rulebook calls for a mechanical counter clamped to the bike's front wheel hub. This device originally built by Alan Jones is still called the Jones Counter. But the counter doesn't measure distance. It counts how many times the wheel turns, and on a wheel a little over two meters around, each turn works out to roughly 23.6 counts. That means a single count on the display equals about 9 centimeters of road, and a full marathon involves around half a million of these.
Before any of that, the measurer has to calibrate the bicycle, which means laying out a short, straight flat stretch of road of at least 300 meters, and measuring it precisely with a steel tape. The human measurer then rides that short stretch eight times total, four before tackling the real course and four after to work out exactly how many counts equal a meter on that particular bike and day. Only then does the actual course get ridden twice along the tightest legal line the rulebook allows. Hugging every curb within 30 centimeters, cutting across intersections diagonally whenever shorter, ignoring what any real runner would do. The handbook asks the reader to visualize it as a string being pulled taut through the course. "The actual path of any given runner is irrelevant," it states.
However, a wet bicycle tire can undo all of that work. As water evaporates off a rolling tire, it cools the rubber, and the chill alone can shift the counter's calibration by an amount equal to the entire 42-meter safety margin built into the course, while steel tape has its own temperature problem. At 10 degrees Celsius a measurer laying out a 600-meter calibration course adds 7 centimeters. At 25 degrees the measurer removes 3 centimeters. If either of course measurements are more than 0.08% apart (34 meters over a full marathon) a third ride is required.
The certified course margin
Every certified course carries a built-in cushion called the short course prevention factor. Each kilometer gets measured to read out as 1,001 meters instead of 1,000. Doug Thurston is a World Athletics and AIMS Level A measurer. He describes the effect as adding "a little more than five feet each mile," which comes to about 42 meters over the full 26.2 miles. A measurer riding the shortest possible route can never do it with perfect precision, so the extra distance guarantees the course is at least as long as advertised, never shorter. A certified marathon can legally measure a few dozen meters over 42,195. Never under.
The 42-meters figure is importnt for another reason. World Athletics also limits how far a marathon's measurement can drift from the true distance, capping it at 0.1 percent of the race, which comes to 42 meters. The short course prevention factor gets built into the course on purpose. The 0.1 percent error limit is just what a careful measurer works to stay inside.
The official 42,195 meters plus the 42-meter margin puts the course itself, measured along the tightest legal line, at about 42,237 meters minimum. The median finisher at the 2026 London Marathon recorded 42,722 meters on a GPS watch. That's according to Terra, a fitness-data company that filtered 921 uploaded activity files down to 571 confirmed marathon finishers. All of them were people who happened to own a GPS watch linked to an app in Terra's network and not a cross-section of the 56,000 people who started the race.
How a GPS watch measures distance
A GPS watch doesn't follow a string pulled along a legal route. It calculates its own position from satellite signals, and that position estimate is noisy. Buildings bounce signals around. Tree cover weakens them. And the receiver only checks its position once a second or so. A 2016 paper in the International Journal of Geographical Information Science, "Why GPS makes distances bigger than they are," found that this noise doesn't cancel out as a route gets longer. An average distance measured by GPS runs longer than the true distance between two points.
One study tested this pattern directly. The researchers strapped eight sport watches to a single runner and measured distance error against a trundle wheel accurate to a centimeter. They tested it on an open running track, on a downtown street in Biel, Switzerland, and on a wooded trail under forest canopy. All eight watches were single-band models from 2018 and 2019, with no dual-frequency receiver in the test. The results, published in JMIR mHealth and uHealth in 2020, are not what most runners would guess:
Where the data disagrees. Runners commonly assume tall buildings and heavy tree cover are what make a GPS watch read a course as long. The eight-watch study found close to the opposite. On the open track, with a clear view of the sky, every one of the eight watches measured more distance than the reference wheel recorded. In the city and under the trees, where signals reflect and drop out most, seven of the eight watches measured less. Multipath reflections and brief signal loss do not add distance in these conditions; they smooth the recorded track, cutting corners a runner ran and removing distance.
USATF's 2014 position statement on GPS makes a related point. Buildings and tree cover can throw a GPS unit off in either direction, the statement says, and separately, most runners don't run the shortest possible route the course was measured along, so "the distance recorded by their GPS device will usually be longer than the certified length of the course." Several secondary sources have attributed a figure of 1 to 2 percent to that statement.
What happens once real runners are on the course
In an actual race, nobody can cleanly separate how much of a watch's overshoot comes from the device and how much comes from the runner. Runners weave through a crowd, drift wide around a corner, and start their watch in a packed corral well before the timing mat. Every large dataset that compares watch readings against a certified course length has both effects mixed in. A 2017 study of Germany's Trollinger-Marathon compared 179 GPS sport watches and 30 phone apps against a certified half-marathon distance. The watches averaged 0.6 percent long, the phone apps 1.7 percent long, a gap wide enough to matter but useless for telling device error apart from route error, since both groups ran the same road. A 2020 study of South Africa's 56-kilometer Two Oceans Marathon, covering 255 runners across eight device categories, found relative error running between 0.6 and 1.9 percent depending on the device, using the identical mixed baseline.
Terra's London data offers one way to see the runner's own contribution without needing to isolate it perfectly. Sorted by finish time, the fastest quarter of runners recorded a median of 42,547 meters. The slowest quarter recorded a median of 42,856 meters, 309 meters more, on the identical road, on the same day, wearing largely the same mix of devices.
A watch alone can't explain that gap. Nothing about a Garmin or a Coros changes based on how fast the person wearing it runs. What changes is how much weaving through a crowded field a slower runner does, how many aid stations get crossed diagonally instead of skipped past, and how much longer a watch keeps counting during a slow walk through a packed start corral. Terra's researchers found this pattern, and so did the running-data site rat.run, analyzing the 2018 London Marathon separately. Recorded distance climbs steadily from roughly 42.5 kilometers for three-hour finishers toward 43 kilometers for five-hour finishers. Two independent datasets, eight years apart, agree on both the size of the effect and its direction. That's about as solid as this kind of evidence gets outside a controlled lab.
Why nobody publishes the split
No study divides that 485 meters into a device share and a runner share, because the two kinds of research operate at different scales. Studies that measure real marathons, Terra's data and the Trollinger-Marathon and Two Oceans studies among them, compare a watch's total reading against the certified course length. That total already has both causes baked into one number, and there's no way to pull them apart after the fact. Studies that do isolate device error precisely, like the eight-watch comparison against the trundle wheel, use reference distances between 400 meters and 4.3 kilometers. Short enough to control tightly, but far too short to stretch across a 42-kilometer race, where errors partly cancel over time in ways a short test course never captures. Thus, no article can hand its readers a tiny percentage split between the two causes. It simply doesn't exist yet
The direction, at least, is well established. The course itself is built slightly long on purpose, by a documented rule, measured to a precision most runners never think about. Everything added after that, on race day, leans long too, for reasons a steel tape and a Jones Counter were never built to record. 42cal has covered where the 42,195-meter figure came from in a separate piece on why the marathon is 26.2 miles and not a rounder number.
A runner pacing off a watch that's already reading a few hundred meters long is running a true pace slightly slower than the watch's live number shows. By the finish, the official time tends to land behind whatever the watch implied for most of the race. Checking splits against a pace calculator, using the actual mile or kilometer markers on the course rather than a watch's running distance counter, keeps that gap from becoming a surprise in the closing miles.
