Evidence guide

Internet speed, without invented universal ratings

A speed result is useful only when its scope and comparison are clear. This guide uses provider-published requirements and the site's documented browser-to-edge method.

Sources reviewed July 10, 2026.

What the measurements mean

WiFiSpeedTest does not label failed HTTP requests as end-to-end packet loss. It also does not claim to measure radio signal strength or the full capacity of an ISP plan. The exact sample counts, formulas, and trimming rules are in the measurement method.

Published service requirements

Published use caseProvider figureHow to use it
Netflix 720p3 Mbps or higherCompare with the measured download result while accounting for other traffic on the same connection.
Netflix 1080p5 Mbps or higherNetflix's recommendation applies to one stream under its documented conditions.
Netflix 4K15 Mbps or higherA household with simultaneous traffic needs capacity beyond a single stream's figure.
Zoom web app, 1080p group video3.8 Mbps down / 3.0 Mbps upBoth directions matter; Zoom lists other figures for other modes.
Microsoft Teams meeting video, best performance4 Mbps down / 4 Mbps upMicrosoft publishes lower minimum and recommended figures for several scenarios.
FCC fixed-broadband policy benchmark100 Mbps down / 20 Mbps upThis March 2024 benchmark is not an average, plan guarantee, or application requirement.

How to make a comparison defensible

  1. Record the device, browser, connection type, test endpoint, and time.
  2. Pause avoidable downloads, updates, cloud backups, and other local traffic.
  3. Run several tests under the same conditions and compare their median rather than selecting one extreme.
  4. Repeat after changing one variable, such as moving closer to the access point or using Ethernet.
  5. For an ISP-plan dispute, also use the ISP's prescribed test method and keep its terms and result records.

There is no truthful universal percentage that every subscriber should receive from every advertised plan. Access technology, service terms, local network, test route, device, protocol, and congestion all affect the observation.

Primary sources

Run the browser-to-edge testRead the measurement guide