Location Tracking & Metadata
Mobile Privacy & Data Protection · Mobile Advanced Security track
10 min read
In this module
Location: The Most Sensitive Data
Location is the most invasive thing your phone produces. A running record of where a person has been hands over their home, their workplace, where they pray, which clinic they visit, what protests they attend, who their friends are. Unlike most personal data, location history can't really be anonymised. Studies have shown four spatio-temporal points are usually enough to single someone out.
Location Determination Methods
Phones lean on a handful of technologies to figure out where they are. Each comes with its own accuracy, battery hit, and privacy footprint.
GPS
GPS gives the best outdoor accuracy, typically 3-5 metres. The receiver listens to multiple satellites and trigonometrically pins itself down. GPS is passive — the device only receives, never transmits, so the GPS network itself can't track anybody. Any app with location permission, however, can scoop up the resulting position.
WiFi Positioning
WiFi positioning leans on huge databases of access-point locations curated by Apple, Google, and friends. The device scans nearby networks, ships the observed BSSIDs and signal strengths to a positioning service, and gets a position back. It works indoors where GPS gives up, usually accurate to 10-30 metres.
Cell Tower Triangulation
The phone's connection to cellular base stations gives a rougher fix. Signal strength from multiple towers, plus a bit of geometry, yields a position estimate. Accuracy spans 50 metres in dense city blocks to several kilometres out in the countryside.
BLE Beacons
Bluetooth Low Energy beacons (iBeacons, Eddystone) handle indoor positioning. Beacons broadcast an identifier; the device measures signal strength from a few of them and triangulates inside the building. Retailers love them for mapping customer paths through stores. Same goes for museums and airports.
Location Permissions
Both iOS and Android have ratcheted down on location permissions year after year. Users get steadily finer-grained control.
Permission Levels
- While Using the App: The app receives location only when it is in the foreground and actively being used.
- Always: The app receives location even in the background. This is required for navigation, fitness tracking, and geofencing, but provides the most complete tracking capability.
- Approximate vs Precise (iOS 14+, Android 12+): Users can grant approximate location (roughly neighbourhood-to-town scale, a few kilometres; the exact coarsening differs between platforms) instead of precise GPS coordinates. This is sufficient for weather and news apps but insufficient for navigation.
Permission Evolution
Watching how location permissions evolved shows the platforms slowly waking up:
- Pre-2014: One permission grant and the app had location forever.
- 2014-2019: The "While Using" vs "Always" split arrived. Background access started needing a justification.
- 2020+: Approximate location landed as an option. Background requests now trigger a separate confirmation dialog, and iOS shows a literal map of where the app has been peeking.
Background location access (the "Always" permission) is the most privacy-invasive permission on mobile devices. It allows an app to record the user's location 24 hours a day, whether or not the app is actively being used. Both iOS and Android now require apps to justify why they need background location access, display periodic reminders to users that an app is using background location, and provide an easy way to downgrade to "While Using" only. Despite these controls, many apps request background location for features that could function with foreground-only access.
Geofencing
Geofencing lets apps define a geographic region and get pinged when the device crosses its boundary. The OS watches the location and only wakes the app on entry or exit. Cheaper on battery than polling location every few seconds.
How It Works
- The app registers a geofence with the OS: a centre point (latitude/longitude) and a radius.
- The OS uses a combination of cell tower, WiFi, and GPS data to monitor the device's proximity to registered geofences.
- When the device enters or exits a geofence, the OS wakes the app and delivers an event.
Privacy Implications
Geofencing has plenty of legitimate uses — location-based reminders, smart home automation. It also makes a tidy surveillance tool. An app with background location and geofencing permissions logs when the user arrives at and leaves specific places, all without continuous position recording. That keeps the activity quieter in OS-level location logs.
Photo Metadata: EXIF
Every photograph from a mobile device carries EXIF (Exchangeable Image File Format) metadata baked right into the file.
What EXIF Contains
- GPS coordinates: Precise latitude, longitude, and altitude where the photo was taken.
- Timestamp: Date and time of capture.
- Device information: Make, model, lens, focal length, exposure settings.
- Orientation: Which way the device was held.
- Software: Which app took the photo and its version.
Privacy Risk
Share photos with EXIF intact and recipients can pull the exact location for each one. Across a batch of photos, an attacker reconstructs movement patterns, sniffs out home and workplace, and maps the daily routine.
Stripping EXIF on Sharing
- iOS: Since iOS 13, the share sheet includes a "Location" toggle that strips GPS data before sharing. However, this is not enabled by default for all sharing methods.
- Android: Google Photos strips EXIF location when sharing links, but sending the original file (via email or messaging) preserves metadata.
- Social media: Most platforms (Instagram, Facebook, Twitter/X) strip EXIF data on upload. However, the platform itself retains and processes the location data.
Here are three photos sent as original files, so their EXIF survived. Click a GPS tag to see where its bytes sit in the APP1 segment, then line up the capture times and positions and decide what kind of place the spot is near. Finish by stripping each photo and watch what's left to read.
Read the location out of a photo
3 small JPEGs, each with an APP1 segment holding EXIF in a TIFF structure, as a phone camera writes it. Illustrative data, generated in code: the places sit in a public city park and are made up, and so is the phone.
Park, lakeside path. File size 497 bytes.
- SOI 2
- APP1 356
- DQT 69
- SOF0 13
- DHT 22
- DHT 22
- SOS 10
- Scan data 1
- EOI 2
Valid JPEG structure, SOI to EOI. Numbers are segment sizes in bytes.
FF E1, length, "Exif\0\0"GPS pointer in IFD0GPS IFD and its valuesSelected tag
Addresses are file offsets. Offsets inside the TIFF count from the MM mark at 0x00C, so add 12 to find them here.
| IFD | Tag | Type × count | Value |
|---|---|---|---|
| IFD0 @8 | ASCII × 13values at 98 | "Examplephone" | |
| ASCII × 5values at 112 | "EP-7" | ||
| SHORT × 1value inside entry | 1 | ||
| ASCII × 11values at 118 | "Camera 4.2" | ||
| ASCII × 20values at 130 | "2026:03:03 07:12:40" | ||
| LONG × 1value inside entry | offset 150 | ||
| LONG × 1value inside entry | offset 200 | ||
| Exif @150 | UNDEFINED × 4value inside entry | "0232" | |
| ASCII × 20values at 180 | "2026:03:03 07:12:40" | ||
| GPS @200 | BYTE × 4value inside entry | 2, 3, 0, 0 | |
| ASCII × 2value inside entry | "N" | ||
| RATIONAL × 3values at 290 | 51/1, 30/1, 3103/100 | ||
| ASCII × 2value inside entry | "W" | ||
| RATIONAL × 3values at 314 | 0/1, 9/1, 5357/100 | ||
| BYTE × 1value inside entry | 0 | ||
| RATIONAL × 1values at 338 | 3150/100 |
Seconds are stored to a hundredth, about 30 cm of latitude. Paste 51.508619, -0.164881 into any map and it lands on the spot.
| Photo | Taken | Decimal position | From centre |
|---|---|---|---|
| 1 | Tue 07:12 (early morning) | 51.50862, -0.16488 | 7 m |
| 2 | Wed 21:46 (late evening) | 51.50891, -0.16422 | 49 m |
| 3 | Fri 07:05 (early morning) | 51.50845, -0.16531 | 42 m |
3 photos, all within 50 m of 51.50866, -0.16480. Times use DateTimeOriginal, the capture time; IFD0 DateTime changes when an app rewrites the file.
Which kind of place is this spot most likely close to?
Stripping only cleans your copy. A platform that strips EXIF on upload still received the original, location and all.
Network Metadata
App permissions aren't the only path to location data. The device's network activity leaks plenty by itself.
Cell Tower Logs
Cellular carriers log which cell tower each device was attached to, and when. Those logs paint a continuous picture of the device's approximate location, no app permission required. Carriers keep the data for billing, capacity planning, and lawful intercept. In many jurisdictions, a court order is enough for law enforcement to pull cell tower location records.
WiFi Probe Requests
With WiFi turned on, devices send out probe requests hunting for nearby networks. Historically, those probes carried the device's real MAC address and a list of previously connected SSIDs. Anyone with a receiver pointed at the right frequencies could:
- Track the device's presence at a location over time (using the MAC address).
- Infer locations the user has visited (from the SSID list - e.g., "Hilton_WiFi", "Stanford_Guest").
Modern iOS and Android use MAC address randomisation: probe requests carry a random MAC instead of the real hardware address. Trackability drops, though it doesn't disappear. Implementation bugs and timing correlations have proven good enough to relink randomised addresses back to one device.
De-anonymisation Research
Academic work has shown again and again that location data refuses to anonymise.
The Four-Point Rule
A 2013 study by de Montjoye and colleagues crunched 15 months of mobility data for 1.5 million people. The headline: four spatio-temporal points (location plus timestamp) uniquely identify 95% of individuals. Even at cell-tower spatial resolution and hourly temporal resolution, location traces are wildly unique because human routines are habitual and idiosyncratic.
You can rerun a small version of that attack below on a made-up town of 1,000 synthetic people, so its numbers are its own and not the study's. Reveal one point at a time and watch the crowd that fits thin out, then blur the data with bigger cells or longer time windows and check how many extra points it takes to get back to one.
Unique in the crowd
Pick a person, then reveal where they were at random hours of the week. Each point rules out everyone who wasn't in the same place at the same time.
Location detail
Time detail
homes of people who still fit (bigger dot, more people) person 1's home revealed point
No points revealed. All 1,000 people fit.
- Everyone1,000
Across 300 random people at this setting
Share left with exactly one match (themselves) after k random points.
- 1 point4%
- 2 points55%
- 3 points74%
- 4 points85%
- 5 points91%
- 6 points95%
Synthetic data: 1,000 made-up people on a made-up 16×16 grid of antenna cells, one location per hour for a week, built from home and work routines, outings and noise. The percentages here come from this toy town, not from any real dataset. For comparison, de Montjoye et al., “Unique in the Crowd: The privacy bounds of human mobility”, Scientific Reports 3:1376 (2013), found that four random points identified 95% of 1.5 million people in real phone records at hourly, antenna-level resolution.
Movement Pattern Analysis
Better de-anonymisation skips the specific addresses and works on the patterns:
- Home/work inference: Where the phone sleeps most nights is probably home. Where it sits during working hours is probably the workplace. Those two locations alone narrow the search dramatically.
- Routine detection: Commute paths, regular stops (gym, school, place of worship), and frequent co-location with specific other devices form a behavioural fingerprint that's stable over months.
Significant Location History
iOS Significant Locations
iOS keeps a local database called Significant Locations — frequently visited places, with arrival and departure times and visit counts. The data sits on-device, encrypted with the user's passcode, and (per Apple) isn't sent to Apple. It powers predictive routing, location-based suggestions, and Photos Memories.
You can poke at it in Settings > Privacy > Location Services > System Services > Significant Locations. Forensic tools pull the same data straight from device backups, exposing a fairly damning map of where someone spends their time.
Google Location History
Google's Location History (now woven into Timeline in Google Maps) records location continuously when enabled. Apple keeps its equivalent on-device. Google historically stored it on its own servers. Server-side storage is what made "geofence warrants" possible: court orders forcing Google to list every device present at a given location during a given time window. In December 2023 Google announced that Timeline data would move onto users' devices, with optional encrypted cloud backups, a change widely read as taking Google out of the geofence-warrant business because it would no longer hold the data.
The courts caught up at roughly the same time. In United States v. Smith (2024), the Fifth Circuit called geofence warrants unconstitutional general warrants, while the Fourth Circuit's en banc court in Chatrie split on whether one was even a search. The Supreme Court took Chatrie and ruled in June 2026, 6-3, that obtaining someone's location history this way is a Fourth Amendment search that needs a warrant, even when the user opted into Google's Location History. It sent the question of whether that particular warrant was valid back to the lower court, so geofence warrants aren't banned, but they now sit squarely under the warrant requirement.
Mobile location data is uniquely privacy-sensitive because it reveals intimate details of a person's life and cannot be meaningfully anonymised. Location is determined through GPS, WiFi positioning, cell triangulation, and BLE beacons. Network metadata (cell tower logs, WiFi probe requests) generates location data independently of any app permission. EXIF photo metadata embeds precise GPS coordinates that survive whenever the original file is shared (email, messaging, file transfer), even though most social platforms strip them from what other users see. Research has shown that four spatio-temporal points can uniquely identify 95% of individuals, making location data de-anonymisation trivially achievable with even coarsely granular data.