How we measure, and what the numbers can't tell you
The honest version
The limitation that matters most
Everyone who appears in our data chose to run a speed test. That is a self selected sample, and it is not representative of Australia. People tend to test when something feels wrong, which pulls results down, and people on fast connections enjoy testing, which pulls them up. We have no way to correct for either.
So treat every figure here as "what the people who tested got", never as "what Australians get". We publish the sample size next to every number so you can judge how much weight it carries. Where a sample is too thin to be fair, we publish nothing at all rather than publishing it with a caveat.
The test itself
Measured against our own server in Sydney, not a shared global network, so the number reflects your line rather than the distance to someone else's data centre. Four parallel connections, because a single stream materially underestimates fast links.
We measure download, upload, latency, jitter, packet loss, and bufferbloat. That last one is the delay your connection adds when it is already busy, and it is the difference between a video call that holds up while someone else streams and one that falls apart. Most speed tests skip it entirely.
Speeds shown for plans are typical evening speeds, not the maximum printed on the plan. The evening figure is what matters day to day. All prices include GST.
How we work out who your provider is
From your connection's network identity: the ISP name, the organisation, the network operator and the autonomous system number, checked together rather than one at a time. Checking them together matters more than it sounds. Belong runs on Telstra's network and Exetel runs on Superloop's, so a lazy check files those customers under the wrong brand entirely.
Where a brand rides another company's network we name the brand you actually buy from. Where a network entity exists only as the plumbing behind a retail brand, we name the brand. Nobody has ever bought a plan from Microplex.
We can identify providers we do not review or earn anything from. Being able to name your provider and having a commercial relationship with them are separate things, and conflating them would quietly bias the data.
What we exclude, and why
Of 282 results in the last 90 days, 151 are eligible for ranking. The rest are stored but kept off every board:
| Reason | Results |
|---|---|
| OBBR Pulse monitoring runs (one household can generate 48 a day) | 75 |
| Hosting, CDN, corporate or overseas connections | 53 |
| VPN or proxy detected | 2 |
| Measured speed impossible for the stated plan | 1 |
We use medians, never averages, so one unusually fast connection cannot carry a whole group. Results older than 90 days stop counting.
What you tell us, we cannot verify
Your plan tier and how your connection reaches the house are both self reported. We pre select a guess to make it one tap, but we never record a guess as though it were an answer. Where someone tells us a plan tier that their measured speed makes impossible, we keep the result and keep it off the boards.
Money, stated plainly
We earn a commission when someone signs up to a plan through us, and several of the providers that appear in this data are companies we earn from. That is worth knowing when you read a delivery figure.
It is also why the provider section works the way it does, and it is worth being specific about that.
Why there is no provider league table
The provider section is a list of companies that deliver what they sell. It is not a ranking, and there is no bottom of it.
A provider appears once at least 25 of its customers have run the test and told us their plan tier, and once the median of those results is at least 85% of what the plan is sold as. A provider that does not clear both is simply not listed. We publish nothing about it, positive or negative.
Three reasons, in order of how much they matter.
Small samples produce wrong answers about real companies. Before this was settled we looked at what the numbers would have said on the data we had. One well regarded national provider computed at 58% of tier on a sample of five. That is not a finding, it is noise, and publishing it would have been unfair to a company that did nothing wrong.
A numbered table always names a worst. Even with a good sample, a league table's real output is its last row. A median in the seventies can mean congestion in one suburb, or one household with a bad modem, as easily as it can mean anything about a provider's network. We are not willing to publish that as a judgement.
We earn commission from many of the companies that will appear here. That is exactly why the standard is set before the data arrives rather than after, and why the rules are written down on this page where anyone can hold us to them.
Delivery is capped at 100% in what we display. Measuring above your plan's tier means the line burst above what you bought, not that you received more than you paid for, and showing 112% would read as a broken number rather than a good one. The uncapped figure is kept in our own data.
Write a review of your broadband company to go in our draw to win a $100 Amazon Gift Card!
Read the competition details and terms.
Add a review