How accurate is GPS speed?

How GPS speed is actually calculated

Your phone doesn’t have a speed sensor. It estimates speed one of two ways:

1. Doppler shift: GPS chipsets can measure the shift in frequency of satellite signals caused by your motion, and derive speed directly from that. This is what populates the coords.speed value in a browser’s Geolocation API on devices that support it (most Android phones). It’s generally more stable than position-based calculation, especially at low speeds.
2. Position-over-time: when direct speed isn’t available — iOS Safari never provides it — speed has to be calculated from how far you moved between two GPS fixes, divided by the time between them. This is what this speedometer falls back to automatically.

Typical accuracy you should expect

Consumer-grade GPS position accuracy is typically 3–10 meters under open sky, reported to you as the accuracy value alongside every position fix. Speed accuracy derived from Doppler shift is usually better than position accuracy alone — often within 0.5–1 km/h at steady highway speeds. Position-based (fallback) speed calculation is noisier at low speeds, because a few meters of GPS noise is a much bigger fraction of the (small) distance you covered in one update interval than it is at higher speeds.

In practice: expect GPS speed to be close and useful, but don’t treat it as a legally precise instrument — it isn’t calibrated or certified the way a vehicle speedometer is.

What makes GPS speed less accurate

• Urban canyons: tall buildings reflect signals (multipath), adding noise.
• Tree cover and tunnels: block or degrade satellite visibility entirely.
• Cold start: right after opening the page, the first few fixes are the least accurate.
• Low speed: at walking pace, GPS noise is a larger fraction of your actual movement.
• Device quality: older or budget GPS chipsets are simply noisier than newer ones.

How this speedometer handles it

We discard any GPS fix with a reported accuracy worse than 35 meters rather than show it to you, and we smooth the last few accepted readings together to reduce jitter. This trades a small amount of responsiveness for a number that’s meaningfully less noisy than raw, unfiltered GPS output.