How Our Speed Test Works

The dataset at a glance

Where the test runs: Cloudflare edge servers

The test runs entirely in your web browser using Cloudflare's open-source speed test engine. Your browser downloads and uploads blocks of data to and from Cloudflare's speed test service, which sits on Cloudflare's global edge network. Cloudflare uses anycast routing, so you are automatically connected to the nearest Cloudflare data centre — for UK users that is normally one in the UK.

Because the test server is a large, well-connected network rather than a server inside your ISP, the result reflects what you would experience reaching real websites and services, not just the link between you and your provider.

Max Throughput and Quality modes

You can switch between two modes above the start button. Both measure the same things; they differ in how much data they move.

Max Throughput (the default)

Designed to get closest to your broadband package's headline speed. After 10 latency measurements it runs progressively larger transfers — downloads of 1 MB, 25 MB, 100 MB and 300 MB and uploads of 1 MB, 25 MB, 75 MB and 150 MB. Large transfers give TCP time to reach full speed, which matters on fast full-fibre lines. Only transfers lasting at least 1.2 seconds count towards the speed figure.

Quality

Mirrors Cloudflare's own connection-quality test. It takes 20 latency measurements and then runs many smaller transfers, starting at 100 kB and stepping up to 250 MB for download and 100 MB for upload. It shows how your connection behaves for everyday browsing, streaming and video calls rather than its absolute peak. Transfers lasting at least 1 second count towards the speed figure.

In both modes the engine stops stepping up to bigger transfers once a transfer takes longer than about a second, so slower connections don't download the largest files and use less data.

What we measure and how it is calculated

• Download and upload speed (Mbps). Each transfer is timed using the browser's own network timing data and converted to megabits per second. Your result is the 90th percentile of the qualifying transfers — close to the best sustained speed reached, while ignoring a single lucky spike.
• Ping (ms). The time for a tiny request to reach Cloudflare and come back while your connection is otherwise idle. We report the median of the latency measurements, with the server's own processing time taken off.
• Jitter (ms). How much your ping varies: the average difference between consecutive latency measurements. Low jitter matters for gaming and video calls.
• Loaded latency. Ping measured while a download or upload is running. A big jump compared with idle ping is a sign of bufferbloat — your router queuing too much traffic.
• Connection type (estimate). We label a result as likely wired (jitter under 2 ms), likely Wi-Fi (5 ms or more) or mixed. This is a rough guide from jitter, not a reading of your actual device settings.

Why your result may differ from your ISP's test

It is normal for two speed tests to give different numbers. The most common reasons:

• Wi-Fi. Our test measures your whole path, including Wi-Fi. Walls, distance and interference often cut speeds well below what reaches your router. Your ISP's guaranteed speeds are measured at the router. Our Wi-Fi guide covers the fixes.
• Your device and browser. Older phones, laptops and network adapters can't always keep up with gigabit connections, and browser extensions or background downloads take a share.
• Other people using the connection. Streaming, cloud backups or game downloads elsewhere in the house all share the same line.
• Different test servers. ISP speed tests usually run to a server inside the ISP's own network. We test to Cloudflare's edge, which is closer to how you reach real websites.
• Upload on very fast full fibre. On FTTP lines of around 500 Mbps and above, Cloudflare-based tests are known to under-read upload compared with ISP-branded or iperf-style tools, which are built to push a line to its absolute limit. Max Throughput mode gets closer, but may still read lower.
• Time of day. Many networks slow down in the evening peak (roughly 7pm–11pm). See why broadband is slow at night.
• VPNs. A VPN adds a detour and its own speed limit, and changes which provider we detect.

How ISP and location are detected

When the page loads, your browser asks an IP geolocation service (ipapi.co) about your public IP address. It returns the organisation that owns that block of addresses, an approximate town, the country and a postcode district. We use these to label your result with an ISP and a town.

• The provider is the owner of your IP address, usually a legal company name such as “British Telecommunications PLC”. We map these to the brand names people recognise (BT Broadband, in that example).
• If you're on a VPN, a work network or a mobile hotspot, the owner will be that network rather than your home broadband provider.
• The town is an approximation. IP location often points to where your provider's equipment is rather than your home, so smaller places can be under-counted and big cities over-counted.

How results are anonymised

When a test finishes, the result is sent to our server, which stores it in an anonymised form:

• IP address truncated on the server before storage. For IPv4 the last number is replaced with zero (a /24 block shared by 256 addresses); for IPv6 only the first four groups (the /64 network prefix) are kept. The full address is never saved.
• Postcode district only. We keep the outward part of a UK postcode (for example “NN1”), never the full postcode. Our server strips anything more detailed before saving.
• No account, name or email is needed to run a test, and none is attached to a result. If you choose to sign up for deal alerts after a test, that is stored separately and covered by our privacy policy.

The IP address shown on the results screen is looked up by your browser for your information only; it is not part of what we save.

How spam and outliers are filtered

A public speed test attracts some fake submissions. Every result is checked on our server before it is saved:

• Range checks. Download and upload must be between 0 and 6,000 Mbps (enough for the fastest UK multi-gig packages), ping no more than 5,000 ms and jitter no more than 2,000 ms.
• Plausibility checks. Results are discarded if download plus upload exceeds 10,000 Mbps, if upload is more than double download and above 200 Mbps (not possible on UK consumer lines), or if a speed above 500 Mbps comes with a ping under 1 ms.
• Rate limiting. No more than three results a minute are accepted from the same anonymised IP block.
• Blocklists. IP blocks used in past spam attacks, offensive or fake provider names and made-up locations are rejected. Text fields are cleaned and length-limited.
• Audit log. Rejected submissions are logged so we can spot new attacks.

Our UK statistics also leave out results from outside the UK, and provider statistics leave out results where the provider couldn't be detected. The monthly reports go further and exclude data centres, cloud hosts, CDNs, VPN exits and corporate security proxies, so they reflect home broadband.

How rankings and medians are calculated

• Medians, not averages. Every speed we publish for a provider, town or month is a median — the middle result. A handful of very fast or very slow tests can drag an average a long way; a median barely moves.
• ISP pages and rankings need at least 3 tests before a provider appears. The rankings are ordered by a confidence-weighted score that pulls providers with few tests towards the overall median until they have around 10 tests, so a couple of lucky results can't put a small provider top.
• Town pages are created as soon as a town has a test, so treat towns with only a few tests as indicative.
• Monthly reports only rank providers and towns with at least 10 tests that month. Months are calendar months in UK time, and a month with fewer than 200 tests is marked as limited data. To stop a few people re-testing the same line many times from skewing the figures, each connection is counted at most once per day.
• Rankings are calculated automatically. Advertising and affiliate relationships have no effect on them.

Limitations of the data

Our data is useful, but it isn't a census of UK broadband. Please keep these limits in mind:

• Self-selected sample. People tend to run a speed test when something feels wrong, or just after upgrading. The results are not a random sample of UK homes.
• Real-world, not line speed. Results include Wi-Fi, devices and household use, so they are usually lower than the speed delivered to the router.
• Repeat tests. Each test counts separately, so a household that tests often has more weight.
• Approximate ISP and town. Both come from IP lookups and can be wrong, as described above.
• Small samples. Many smaller providers and towns have only a handful of tests. Look at the test count before drawing conclusions.
• Not comparable with Ofcom headline figures. Ofcom's Connected Nations 2025 report puts full-fibre availability at 78% of UK homes, gigabit-capable coverage at 87% and the average maximum download speed at 285 Mbps. Those describe what lines can deliver; our medians describe what people actually measured in their homes, so they are naturally lower.

Spotted something that looks wrong? Let us know — we read every message.