GPS navigation apps calculate your arrival time by combining three things: your current speed, the best route through the road network, and predictions about what traffic will look like on every segment of road between you and your destination. The result is an estimate that updates continuously as you drive, pulling in fresh data every few seconds.
What seems like a simple countdown on your screen is actually the product of satellite physics, graph theory, machine learning, and millions of other drivers feeding data back into the system. Here’s how each piece works.
How Your Phone Knows Your Speed
Everything starts with GPS satellites. Your phone receives signals from multiple satellites orbiting Earth, and by measuring tiny differences in how long each signal takes to arrive, it pinpoints your location. But location alone doesn’t give you speed. For that, your receiver uses something called the Doppler shift: as you move toward or away from a satellite, the frequency of its signal changes slightly, the same way a siren’s pitch changes as an ambulance passes you. By measuring that frequency change, your phone calculates your instantaneous velocity with high precision. According to Penn State’s GNSS program, this velocity measurement has been one of the primary applications of GPS since the technology’s earliest days.
Your phone typically updates this speed reading once per second. The navigation app takes that real-time speed and compares it against the expected speed for the road segment you’re on, which feeds into the broader arrival time calculation.
Finding the Fastest Route
Before the app can estimate when you’ll arrive, it needs to figure out which roads you should take. Road networks are treated as graphs: intersections are points (nodes), and the roads connecting them are lines (edges) with associated costs like distance, speed limit, or current travel time.
Two classic algorithms do the heavy lifting. Dijkstra’s algorithm finds the shortest path by systematically exploring every possible route outward from your starting point until it reaches the destination. It guarantees the optimal answer but checks a lot of unnecessary roads along the way. A* search improves on this by adding a heuristic, essentially a rough estimate of how far each intersection is from the destination as the crow flies. This lets it focus on roads that are heading in the right general direction, dramatically reducing computation time while still finding the best route.
In practice, modern navigation apps use heavily optimized versions of these algorithms. A road network for an entire country can contain tens of millions of segments, so apps preprocess the network into hierarchical layers. Long highway segments get priority in cross-country routes, while local streets only get evaluated near your origin and destination. This is why route calculation feels nearly instant even for a drive spanning several states.
Predicting Traffic Along Your Route
The route alone isn’t enough. A 30-mile stretch of highway might take 25 minutes at 2 a.m. and 90 minutes during rush hour. To predict what traffic will look like on each segment of your route at the time you’ll actually be driving on it, apps layer in two types of traffic data.
The first is historical. Navigation providers have years of aggregated speed data for virtually every road, broken down by time of day, day of week, and season. They know that a particular interchange slows to a crawl on weekday mornings between 7:45 and 8:30, or that a beach road backs up on summer Fridays. This historical baseline gives the app a starting prediction for every segment.
The second layer is real-time. Apps like Google Maps and Waze collect speed and location data from millions of active users. When someone has the app open (even in the background), their phone reports its speed and position back to the servers. Waze, for instance, draws from a community of millions of users whose phones continuously share this information. If a highway segment normally flows at 60 mph but current users on that road are crawling at 15 mph, the app knows there’s a problem and adjusts your ETA accordingly.
Google Maps takes this a step further with machine learning. The company deployed a graph neural network system that models the road network’s structure and predicts how traffic conditions will evolve over the course of your trip. This matters because your ETA depends not just on current conditions but on what traffic will look like 20 or 40 minutes from now when you reach a given stretch of highway. The model captures complex patterns: an accident on one road causes backups on parallel routes five minutes later, or a stadium event creates a wave of congestion that moves outward through surrounding streets.
Accounting for Stops, Turns, and Intersections
Raw driving speed on open road is only part of the picture. A significant chunk of urban travel time is spent not moving at all, waiting at traffic signals, decelerating into turns, or sitting in queue behind other vehicles.
Navigation apps build these delays into their models. Traffic signals introduce what engineers call “control delay,” which includes the time spent slowing down as you approach a red light, waiting in the queue, and accelerating back to speed when it turns green. Federal Highway Administration data puts the typical lost time at about 4 seconds per signal phase just from startup delays and clearance gaps, and that’s before counting the actual time spent waiting for a green light. On a route with 30 signalized intersections, these small delays compound into minutes.
Left turns across traffic, stop signs, highway on-ramps, school zones: each type of maneuver has a characteristic delay profile that the app estimates based on historical travel times through that specific intersection or segment. The app doesn’t need to know your traffic light’s exact cycle. It knows from aggregated data that cars passing through a particular intersection at 5:15 p.m. on a Tuesday typically take 45 seconds to get through, and it uses that average.
Why Your ETA Keeps Changing
The arrival time on your screen isn’t calculated once and left alone. It recalculates continuously, and the inputs shift constantly. You hit an unexpected red light, and the ETA ticks up by 30 seconds. Traffic clears on the highway ahead, and it drops by two minutes. A crash is reported three miles ahead, and the app recalculates both the time and potentially the entire route.
Rerouting decisions follow a cost-benefit logic. The app compares your current route’s projected time against alternatives and suggests a change when the savings cross a threshold. In practice, this threshold can be quite small. Waze users have noted being rerouted off highways and through side streets to save as little as 60 seconds, a source of frequent frustration. Some drivers have asked for the ability to set a minimum time savings (say, 10 minutes) before the app suggests an alternate route, but most apps don’t offer that level of control.
The continuous recalculation also explains why your ETA often becomes more accurate as you get closer to your destination. Early in a long trip, the app is predicting traffic conditions 30 or 60 minutes into the future, which is inherently uncertain. As you get closer, more of the remaining route reflects real-time conditions rather than forecasts.
How Accurate the Estimates Really Are
For most trips, modern navigation apps land within a few minutes of the actual arrival time. Short drives in familiar urban areas tend to be remarkably precise, often within a minute, because the app has dense real-time data and fewer unknowns. Longer trips, rural routes with sparse user data, and drives during rapidly changing conditions (a sudden storm, a major accident) produce larger errors.
No single app is consistently the most accurate. Google Maps and Apple Maps use different routing algorithms, different traffic models, and different data sources, which means they sometimes produce noticeably different ETAs for the same trip. On any given day, one may outperform the other depending on the region and conditions. The quality of the underlying data matters most: an app with more active users on a particular stretch of road will generally give a better estimate for that road.
The biggest source of error isn’t the technology itself but unpredictable events. A traffic signal stuck on red, a train crossing, a last-minute lane closure: these are things no algorithm can anticipate until users on the ground start reporting slower speeds. Once that data flows in, the ETA corrects, but there’s always a lag of a few minutes between a new disruption and the system’s awareness of it.