Why internet speed tests disagree in Canada
A 400 Mbps result on one test and 66 Mbps on another does not automatically mean your connection changed. A test measures a path, a server and a method at a particular moment.
Start with the server
Your browser is only one end of a speed test. The other end is a measurement server or content network. A slow server, a distant route, congestion between networks, or a limit on the testing endpoint can lower the reported number even when your access line is capable of more. Testing services may select different nearby servers and therefore travel across different networks.
AI Speed measures traffic to this site's hosted endpoint. It does not measure the physical maximum of your fibre, cable or wireless connection. If another test gives a much higher result repeatedly, treat the lower result as a possible path or server limitation. It should prompt investigation, not a claim that your provider has reduced your plan speed.
Small requests can understate fast connections
A small download spends a meaningful share of its time starting the request and waiting for a response. On a fast connection, browser scheduling, server work and round-trip delay can distort the estimate. A larger sustained transfer reduces that fixed overhead. Parallel transfers can make better use of a high-capacity path, although they also increase data usage and may produce a higher number than a single application flow.
AI Speed warms the connection, takes eleven small round-trip samples for median ping and the median change between samples for jitter, then reads streamed data from four parallel downloads and measures combined bytes over elapsed time. Upload uses parallel incompressible test data and counts only bytes confirmed by the endpoint. The test stops at a time or data cap below 100 MB. On a very fast link the capped sample may last under 1.5 seconds; repeat that result and compare another endpoint. The reported ping is unloaded HTTP latency to AI Speed, not game-server ping, loaded latency or packet loss.
When fast Mbps still feels slow: check latency under load
A connection can deliver strong download and upload numbers yet become sluggish during a backup, large download or game update. The missing measurement may be latency under load: delay measured while the connection is busy. The CRTC describes this as measuring latency while a speed test heavily uses the broadband connection. Cloudflare likewise distinguishes idle latency, measured before a transfer, from loaded latency measured during download or upload traffic.
AI Speed currently reports unloaded HTTP latency only. It does not run a loaded-latency or packet-loss test, so its ping result cannot rule out delay that appears during heavy traffic. First run AI Speed with other activity paused. Then use a test that explicitly reports idle and loaded latency. Repeat on Ethernet and Wi-Fi without changing anything else. If loaded latency rises on both while idle latency stays steady, repeat with backups and uploads paused and check whether your router offers queue-management controls. Treat the pattern as diagnostic evidence, not proof that a particular device or provider is at fault.
Record the test service, server or endpoint, connection type, time and whether another device was uploading. Avoid publishing screenshots that expose an IP address or account details. Use the home internet test worksheet for a shareable record or the connection diagnostic for symptom-by-symptom follow-up.
Compare results under the same conditions
Run each service several times on the same device within a short period. Pause large downloads, video calls and cloud backups. Test once over Ethernet if you can, then test Wi-Fi where you normally work. A wired result close to your plan with a slower Wi-Fi result points toward local radio conditions or equipment. Distance, walls, interference and the Wi-Fi standard can all matter.
Record download, upload and latency separately. A fast download number does not guarantee low delay for video calls or games. If you compare providers, use repeatable conditions and a representative sample at different hours. One result from one household is not a provincial or ISP average.
A repeatable comparison you can do at home
- Write down your plan’s advertised download and upload speeds and whether they are stated as “up to” speeds.
- Use the same computer and browser. Connect by Ethernet, disable a VPN temporarily if safe, and pause backups and downloads.
- Run three AI Speed tests with a short pause between them. Record the middle download and upload values rather than selecting the fastest.
- Repeat with another service immediately afterwards. Record its server name or location if provided.
- Repeat over Wi-Fi in your normal room, then repeat the wired comparison in the evening. Change one condition at a time.
The example below is illustrative, not a measurement of a Canadian provider.
Worked example: which result should you investigate?
Suppose a 500 Mbps plan produces wired download results of 420, 460 and 440 Mbps on one service. The median is 440 Mbps. Wi-Fi readings of 90, 110 and 100 Mbps have a median of 100 Mbps. That difference suggests investigating the local Wi-Fi path first; it does not prove the ISP is at fault. If AI Speed alone reports 70 Mbps while another service repeatedly reports 440 Mbps on the same wired computer, investigate the endpoint or route before replacing your router.
| Observed pattern | Possible explanation | Next check |
|---|---|---|
| Wired is faster than Wi-Fi on both services | Local wireless conditions or device limits | Move near the router and repeat on the same device |
| Only one testing service is slow | Endpoint, route or measurement method | Compare another server; retain both results |
| Both services are slower in the evening over Ethernet | Household traffic or network congestion | Pause other traffic and repeat at another time |
| Idle latency is steady but loaded latency rises | Queueing while the connection is busy | Repeat wired with other traffic paused; compare both directions |
When to contact your provider
If repeated wired tests across more than one service remain below your expectations, send your provider the times, plan speeds, device, connection type and measurements. Ask whether the line, modem provisioning or an outage can explain the pattern. A single low reading or HTTP latency score is insufficient evidence of a fault. The slow-internet troubleshooting guide can help separate local Wi-Fi, equipment and access-line checks first.
What does 50/10 mean?
The CRTC's universal service objective sets a target of at least 50 Mbps download and 10 Mbps upload access for Canadian homes and businesses. It is a policy target, not a promise that every test to every server will report precisely those values. Your retail plan, the technology serving your address and current conditions still matter.