How a Location Snapshot Becomes a Suspect
To understand why Chatrie treats a geofence request as a constitutional search, it helps to follow the Location History record before, during, and after police access. This chapter traces the historical Google system and the three requests used in the Virginia bank-robbery investigation. The goal is to see the transitions clearly: a record that begins as a technical estimate of a device’s position can become a detailed movement history and then an identified person.
What the record contains before police ask for it
Location History was not a single ping created when an officer pressed a button. In the system described by the courts, it was a time-stamped record of where a cell phone had been. Google’s service collected location information approximately every two minutes, including when the user was not actively using the phone. The historical record drew on several signals: GPS, nearby Wi-Fi networks, Bluetooth beacons, cell sites, and other location information such as IP data.
Those inputs served different purposes. A GPS signal can suggest a device’s position outdoors. Nearby Wi-Fi networks and Bluetooth beacons can help locate it in a dense environment. Cell-site information can provide another estimate when the device is communicating with a network. Google combined these signals into a location point and displayed a confidence interval around it. The point was therefore an estimate, not a guarantee that the user stood on one exact spot. The courts nonetheless described the resulting service as unusually precise—within roughly 20 meters in the Supreme Court’s account—and capable in some circumstances of estimating elevation. A phone’s recorded position could potentially indicate which floor of a building it occupied.
The cadence matters as much as the precision. A record collected every two minutes can show whether a device stayed in one place, moved through an area, paused, or followed a route. It can distinguish a phone that remained near the crime scene from one that passed by. It can also connect the crime scene to the places visited immediately before or after it. A single point answers a narrow question: where might the device have been at one moment? A sequence of points can answer a different question: what path did the device take?
The record also had a personal meaning that separated it from a simple technical log. Google presented Location History to users as a Timeline or virtual journal of past travels. A user could look back at a restaurant, a friend’s home, a vacation stop, or a route taken on a particular day. The same database could support consumer features and commercial uses while retaining a detailed account of a person’s movements. That dual role is important. A record can be useful to a service provider and still be something a user reasonably experiences as their own history.
In the historical system, Google stored the data in the cloud rather than only on the phone. The circuit opinions refer to the repository as the Sensorvault. Once Location History was enabled, the record could be associated with a user’s account and devices, while Google used the data to improve services and build models about travel patterns. The information therefore had two characteristics that matter to the later legal dispute. It was broad enough to be searched after an event, and it was structured enough to be queried for a particular place and time.
That combination creates retrospective power. Police did not have to attach a tracking device to a suspect before the robbery or decide in advance whom to follow. They could identify a location of interest after the fact and ask which devices appeared there. The database made it possible to turn the question “Who was near this place?” into a search of records created before investigators knew the answer mattered.
The difference between a location estimate and a movement record should not be overstated in either direction. The opinions describe confidence intervals because the data is not perfectly accurate. A device may be assigned a point that does not place it exactly where its user stood, and an account may not correspond neatly to the person carrying a device at every moment. At the same time, uncertainty at the point level does not erase the value of a sequence. Several points, their timestamps, their estimated accuracy, and their relation to recognizable places can produce a persuasive route even when no individual point is infallible.
This is why the system’s technical behavior cannot be separated from the constitutional question. The important issue is not merely that Google possessed a large amount of information. It is that the information was frequent, precise enough to distinguish nearby locations, persistent enough to be consulted later, and organized so that a request about an unknown person could produce a set of candidate devices.
The three requests and the choices they enable
The historical Google process turned that capability into an escalation. The police first supplied a geographic area and a period surrounding the crime. In Chatrie’s case, the proposed geofence was a circle with a 150-meter radius around the credit union. The first request covered the hour from 30 minutes before to 30 minutes after the robbery. Google returned Location History data for devices that appeared within that area and period, but the returned accounts were initially represented by numerical or pseudonymous identifiers rather than names.
The first output was limited in one sense and broad in another. It was limited because it concerned the defined place and time, and because the identifiers did not directly announce the users’ names. It was broad because the request was not directed at one known account. Every device that met the geographic and temporal description could enter the initial dataset. The output could include coordinates, timestamps, the source of the location estimate, and the confidence interval around it. Police could then examine the set for patterns that appeared relevant to the robbery.
This stage changes the investigation’s unit of analysis. Before the request, the subject is an unknown population near a crime. After the request, the subject is a collection of device histories, each marked by a pseudonym. The police may not yet know the people behind those identifiers, but they can compare the device paths with facts from the crime: when the robber arrived, where the robber approached from, how long the event lasted, or whether an accomplice may have been nearby.
The second request expanded what could be learned about selected devices. In the Virginia warrant, officers narrowed the first list from 19 users to nine. Google then provided additional, still-anonymized location data for those users over a two-hour period. Unlike the first request, this movement information was not confined to the original geofence. It could show where a device traveled before and after the bank visit, including locations outside the circle.
That expansion is not a minor technical refinement. The first request asks whether a device appeared in a defined place during a defined time. The second can show how the device entered, left, and continued moving. A route may help police eliminate a passing phone or focus on a device whose path seems consistent with the known facts. But the same route can reveal a home, workplace, school, hospital, place of worship, or other distinctive destination. The geographic boundary that narrowed the first request no longer performs the same limiting function.
The second stage also creates a feedback loop. Police decide which identifiers deserve more information by inspecting the first output. The reasons for that decision may include repeated presence inside the geofence, a path that appears to match the crime, or the possibility that a user was a witness or accomplice. Whatever the reason, the decision is made after the initial data has been returned. The later request is therefore shaped by the provider’s earlier response and by the officers’ interpretation of it.
The third request attaches identity to the narrowed record. After reviewing the broader movement data, officers reduced the group from nine to three and asked Google for identifying information. Google supplied names and contact information associated with the selected accounts. At that point, the investigation no longer concerned anonymous or pseudonymous devices near a bank. It concerned named people whose location histories had been used to generate investigative leads.
The numbers in Chatrie’s case—19, then nine, then three—are not a standard set of limits. They show the logic of the workflow. A geofence request can begin with a group defined only by place and time, then become a movement search for a chosen subset, and finally become an identity request for a still smaller subset. The reduction in the number of users does not mean that the earlier steps were irrelevant. Each step supplies the information used to decide who will be examined more closely at the next one.
The institutional roles are distributed across that sequence. Police define the crime-related location and time, apply for legal authority, review the first output, and choose which devices to pursue. A magistrate reviews the initial application and issues the warrant. Google searches its stored records, produces the requested material, and historically used its own review process to respond to the request. The legal dispute partly concerns whether that private technical work should be treated as separate from the government’s search or as an action performed under government compulsion.
The roles also show why the word “anonymized” cannot finish the analysis. An identifier can hide a name at one moment while still exposing a device’s movements. A movement pattern can become distinctive when it begins at a particular home and ends at a particular workplace. Public records or other available information may then make it possible to connect the pseudonym to a person. And even when re-identification does not occur, the government may possess a record of the device’s route that can be used to decide whether to request the person’s identity.
The process can generate a useful lead precisely because it preserves movement context. In Chatrie, the final data showed that one user entered the geofenced area shortly before the robbery and then headed toward a residential area. In Smith, the Fifth Circuit described a related investigation in which the first-stage data eventually led officers to two named defendants and to additional phone, account, and historical location inquiries. The geofence did not by itself prove the entire case; it opened a path for further investigation.
That path has an error surface. The courts note that Location History points are estimates and that false positives can sweep innocent bystanders into an investigation simply because their devices were near a crime. A person may be present for an innocent reason, a device may be carried by someone other than the account holder, or a location estimate may overlap the boundary of the search area. The Fifth Circuit also records that some requests take a long time to process and many do not lead to arrests. These observations do not establish a universal failure rate. They establish why each returned device should remain a lead whose significance depends on context, not an automatic finding of guilt.
Scale changes the practical meaning of that uncertainty. When an officer follows one known suspect, an error affects the interpretation of that suspect’s record. When a system first produces a list of everyone whose device may have been nearby, the initial error can distribute suspicion across people who had no connection to the crime. Narrowing the list may reduce the number of people under scrutiny, but it does not explain by itself why the first group was eligible for later scrutiny or who authorized the escalation.
The historical workflow therefore has a clear lifecycle. Location History is generated continuously in the background, stored and organized in a repository, queried after a crime, returned first as a place-time set, expanded into movement context, and finally linked to account identity. Information is not merely added at each transition; its meaning changes. A pseudonymous point becomes a route. A route becomes a reason to select a user. A selected user becomes a named investigative subject.
Once those transitions are visible, the legal question becomes more precise. Police did not merely receive a neutral answer to a question about a public street. They compelled access to a database that recorded users’ movements and used the returned material to decide whose history and identity to examine next. The next issue is whether that government access is a Fourth Amendment search. Chatrie answers yes for historical Location History, but the answer becomes understandable only when the technology is seen as a movement record rather than a single snapshot.