How Accurate Is Google Maps Timeline? Seven Studies Checked

For a decade, Google Maps Timeline was the largest accidental mobility dataset in the world. Anyone with an Android phone and Location History switched on was quietly producing a minute-by-minute diary of where they went, and researchers learned to ask participants to hand it over through Google Takeout. Epidemiologists used it to reconstruct exposure, transport planners used it in place of paper travel diaries, and forensic investigators used it in court.

Then Google changed where the data lives. In a December 2023 announcement, Marlo McGriff, director of product for Google Maps, said Timeline would be "saved right on your device," that Location History would be off by default, and that new users would get a three-month auto-delete instead of the previous 18 months. Any cloud backup would be encrypted so that no one could read it, "including Google." The web version of Timeline is gone; Google's Maps help page now routes exports through the phone itself, under Settings, Location, Location services, Timeline, "Export Timeline data."

That raises two questions for anyone who has relied on this data, in a lab or in their own life. How accurate was Google Maps Timeline in the first place? And what is still possible now that it no longer sits on Google's servers?

What the validation studies actually measured

At least seven peer-reviewed studies have checked Google Location History against something closer to ground truth: a dedicated GPS logger, a scripted itinerary, or a participant's own diary. They do not measure the same thing, which is the first thing to understand about them.

Study	Sample	Result
Ruktanonchai 2018	21 people, GPS logger, 7 days	Median gap from GPS 64 m; ~85% match on 100 m cells
Macarulla Rodriguez 2018	2 phones vs reference GPS	True spot inside Google's radius: 52% on GPS, 7% on Wi-Fi
Goodspeed 2018	21 people, GPS watch, 28 days	Fewer missing days than the watch; more accurate at home
Cools 2021	5 phones, scripted trips, 12 days	51% of 934 places and 32% of 888 trips matched
Hystad 2022	25 people, 1.83M GPS points	79.5% captured within 10 m, 94.8% within 100 m
Amram 2025	357 people, 25 with activity monitors	Walking trips 3,150 counts/min vs 489 in vehicles
Marzorati 2026	294 people, diaries, 30 days	72% mean coverage at 200 m; 12.5% of days empty

Chart: Stanford Tech Review analysis of the studies in the table. Each bar uses the metric its paper reports, so the bars are comparable in direction rather than in definition.

The finding: the closer you get to behaviour, the worse it gets

Laid side by side, the results follow a single gradient. Across five validation studies published between 2018 and 2026, Google Location History captured between 32% and 95% of what actually happened, and the figure fell every time the question moved from "where was this phone" to "what trip did this person make."

Point-level questions do well. Hystad's team, working with nearly two million GPS fixes, found that Google had a point within 100 metres of 94.8% of them. Ruktanonchai's group reported a median gap of 64 metres between Google and a dedicated logger. If you want to know which neighbourhood someone spent the afternoon in, Timeline was very good.

Event-level questions do much worse. Cools and colleagues scripted 888 trips and found that only 32% appeared in Timeline the way they actually happened. The stops Timeline missed lasted 355 seconds on average, under six minutes, against 762 seconds for the stops it caught. The Marzorati study, the largest so far at 294 people, found that one in eight person-days contained no recorded activity whatsoever, and that a home-to-work trip appeared on only 62.5% of days when the participant reported going to work.

The forensic study adds a warning about the accuracy circle Google draws on the map. When the phone had a GPS fix, the true position fell inside that circle only 52% of the time. When it was relying on Wi-Fi, the figure was 7%. The circle is a confidence estimate, not a guarantee, and in this test it was optimistic.

The honest verdict on Google Maps Timeline as a research instrument is that it is an excellent record of places and a mediocre record of journeys.

Why the on-device move matters more to researchers than to users

For an individual, the December 2023 change is mostly a privacy improvement. The data is on a device you control, the default retention is shorter, and Google says it cannot read the backup.

For researchers, it breaks the pipeline. The standard method was data donation: a participant downloads a Google Takeout archive and shares the location file with the study. A 2025 paper by Oje et al. in the International Journal of Health Geographics now includes a section on "policy changes impacting data availability," noting that users "may no longer be able to export their location history data from the cloud" and that future studies should expect much less data.

Donation was already the bottleneck before the change. In Hystad's study, 61 of 378 people contacted (16.1%) donated their history. In a 2025 Dutch panel study by Keusch and colleagues, 256 of 2,086 adults aged 50 and over (12%) completed a donation, and Android users asked for Google Location History were 14 percentage points less willing than iPhone users asked for Apple Health data. Moving the export from a website to a settings menu on each phone adds one more step to a process that already loses most of its participants.

Some study designs have not caught up. A protocol published in BMJ Open in 2026 still plans to collect "location history (timeline)" through Google Takeout, with no mention of the change.

What still works

The data has not disappeared; it has moved. A participant, or anyone curious about their own movements, can still export Timeline from the phone and inspect it. The difference is that the export is now a local file rather than a cloud archive, and that makes the choice of tool more important.

A file containing months of precise locations is exactly the kind of thing that should not be uploaded to an unknown server to be viewed. A browser-based Google Timeline visualizer that parses the export locally, without a backend, lets a person see what their history contains before deciding whether to share any of it with a study. That is a reasonable consent step in its own right: the Marzorati results suggest participants will find missing days and phantom trips, and they should know that before their data becomes someone's dataset.

For research teams, three practical adjustments follow from the evidence above:

1. Ask point questions, not trip questions. Exposure to a neighbourhood, a park, or a pollution source is where the studies show Timeline performing well. Counting trips or measuring stop durations under ten minutes is where it fails.
2. Treat Google's accuracy radius as a lower bound on error. The forensic test found the true location outside the circle in nearly half of GPS-based readings.
3. Budget for low donation rates and an extra export step. The 12% to 16% donation rates above were achieved when export happened in a browser. Instructions now need to be written per phone, and participants on the three-month default will only have three months to give.

Timeline is now a smaller, more private, and more personal record than it used to be. The research on it says the same thing it always said: trust it for where, be careful with when, and do not rely on it for how a person got from one place to the next.

Cover image: a GPS satellite on display, via Wikimedia Commons, public domain.