Run an internet speed test on a phone and then on a laptop and the numbers may be far apart. Repeat the test five minutes later and they may change again. That does not automatically mean the broadband provider or testing website is manipulating the result.
A speed test measures the performance of one device, through one network path, to one group of servers at a particular moment. Change any part of that chain and the result can change.
What does a speed test measure?
Most tests report three headline figures:
- Download speed: how quickly data reaches your device.
- Upload speed: how quickly your device sends data.
- Latency or ping: how long a small data exchange takes, usually shown in milliseconds.
Some also report jitter, packet loss and latency while the connection is busy. Those extra measurements can explain why a connection with a high download number still feels poor during calls or gaming.
Why the selected server matters
The test does not measure speed to the entire internet. It transfers data between your device and a selected test server.
A nearby server with excellent connectivity may produce a higher result and lower latency than a distant or heavily loaded server. Automatic selection generally chooses a responsive option, but different testing services have different server networks and routing arrangements.
For comparison, use the same service and server. To understand real-world performance, test several reputable services and destinations.
Wi-Fi can be the bottleneck
Your internet plan may deliver hundreds of megabits per second to the router, while the device receives far less over Wi-Fi.
Wi-Fi performance depends on distance, walls, interference, channel width, frequency band, router placement and the radio capabilities of both router and device. A phone connected on a crowded 2.4 GHz channel can test much slower than a laptop using a clean 5 GHz or 6 GHz connection.
For the clearest view of the broadband line, connect a capable computer to the router with Ethernet and ensure the cable and ports support the expected speed.
Device hardware changes the result
Older phones and laptops may have slower Wi-Fi standards, limited Ethernet ports or processors that struggle with very fast encrypted transfers. Battery-saving modes and thermal limits can also affect performance.
A device with a 100 Mbps Ethernet port will not demonstrate a 500 Mbps plan, no matter how fast the incoming line is. Likewise, a cheap USB network adapter can become the limiting component.
Before blaming the connection, confirm that the test device can handle the advertised rate.
Other traffic shares the connection
A speed test competes with everything else using the line. Cloud backups, operating-system updates, security cameras, video streaming and another person's download can reduce the available capacity.
Modern tests often use multiple parallel connections to fill the link quickly. That can reveal maximum throughput, but it may not resemble the behaviour of one ordinary download.
To create a repeatable baseline, pause large transfers and test when other devices are relatively idle. Then run a second test during normal household use to understand the experience under load.
Time of day and congestion
Internet networks share capacity at several points. Evening demand can create congestion in a neighbourhood access network, at an internet provider's interconnection point or on the route to a particular service.
A single slow evening result does not prove a persistent fault. A pattern recorded at the same times over several days is more useful evidence.
Save screenshots or results with date, time, connection type and test server if you plan to contact the provider.
Why browser and app tests can disagree
A native speed-test app may use the device's networking system differently from a web test running inside a browser. Browser extensions, privacy tools, antivirus scanning, VPNs and background tabs can add overhead.
The services may also use different numbers of parallel streams, test durations and methods for discarding the slow start or unstable samples.
Neither result is automatically wrong. They answer slightly different measurement questions.
What a VPN does to speed tests
A VPN adds an encrypted tunnel and an extra routing point. The connection first travels to the VPN server and then onward to the test destination.
That can reduce throughput and increase latency, particularly if the VPN server is far away or overloaded. Testing with the VPN on shows the performance of the VPN-protected connection. Testing with it off is better for diagnosing the direct broadband line.
Both measurements can be useful; label them clearly.
Mbps is not MB/s
Internet plans and speed tests normally use megabits per second (Mbps). File-download tools often display megabytes per second (MB/s). One byte contains eight bits.
A 100 Mbps result therefore has a theoretical ceiling of about 12.5 MB/s before protocol overhead and other losses. Confusing the two units can make a normal download look eight times slower than expected.
How to run a fair speed test
Use a simple repeatable process:
- Restart only if you are specifically testing whether a restart fixes the issue.
- Use Ethernet for the broadband baseline where possible.
- Pause large downloads and disconnect a VPN.
- Choose the same server for comparison.
- Run three tests and consider the median, not only the best number.
- Repeat at different times of day.
- Separately test Wi-Fi in the rooms where devices are actually used.
The Ethernet baseline answers whether the internet line reaches the router properly. Room-by-room Wi-Fi tests answer whether the home network distributes that connection effectively.
Which number matters most?
It depends on the task. Large downloads benefit from throughput. Video calls and competitive games care about latency, jitter and packet loss. Cloud backups need upload capacity. A home with many simultaneous users needs enough shared capacity and a router that manages congestion well.
The biggest number is not always the best description of connection quality.
The bottom line
Internet speed tests differ because they measure a changing path involving the device, Wi-Fi or Ethernet link, router, provider, internet route and test server. Compare results under controlled conditions, use multiple measurements and separate broadband performance from Wi-Fi performance.
For more digital-platform context, read how AI Overviews affect search results and browse the Explainers section.
