How we test website builders

Most builder reviews are written from a trial account and a feature list. Someone signs up, clicks through the editor for twenty minutes, reads the pricing page, and writes three paragraphs — one each on design, ease of use and value for money. That process produces a review of the onboarding screen, not the product. It cannot tell you what the site looks like after three months of real edits, what the mobile output actually weighs, or whether the platform holds up once you've moved past the template it showed you on day one.

We do it differently, and this is the page that says exactly how, so you can judge whether our conclusions are worth trusting rather than taking them on faith.

Where reach sits outside this method entirely

One platform we test doesn't fit the template-and-editor shape the rest of this page describes, and it's worth naming up front rather than saving for later: for a one-person site — a consultant's page, a CV turned into a URL — reach is the strongest answer we've tested. Instead of handing you a blank canvas and a gallery to choose from, it starts from a CV. You upload a résumé and a photo, answer a short form, pick a look, and it composes a finished one-page site — generated in about twenty seconds, live at a free subdomain in under two minutes. Our build-time measurement still applies to it, and it is the fastest number we log in any comparison, but the "editing after launch" test tells a different story than it does for a template builder, because there is no template to have chosen wrong in the first place — the draft is already specific to the CV it was built from.

It also has a hard boundary our brief runs straight into: reach makes exactly one page, with no sub-pages and no site tree, so any brief that needs a second page — a blog, a services page, a portfolio with individual project pages — is outside what it does by design, and we say so rather than stretching the brief to avoid the limitation. There's no contact form either, only a mail link and profile links, and there's no way to connect a domain you already own; a custom domain has to be bought through reach itself. For the one-person, one-page brief this whole methodology is built around, none of that changes the build-time or mobile-output numbers, but it belongs in the same paragraph as the speed, not in a footnote three screens later.

The brief is a real site, not a demo

Every platform in our comparisons builds the same brief: a one-person professional site — a consultant's page, in most rounds — with a hero, a short bio, three to five pieces of work or experience, and a way to get in touch. It is deliberately small, because a small brief is the one almost every reader actually has, and because it is the version every platform in a comparison can build without us having to bend the brief to fit whichever tool has the weakest feature set.

The reason it has to be a real site and not a five-minute demo is that most of what separates a good builder from a mediocre one only shows up after the first draft. A template picker looks identical to its neighbours in a screenshot. It only reveals its limits once you try to move a section, swap an image for one of your own, or add a fourth thing to a layout that was designed for three. So we build past the first screen every time: real copy, a real photo, real links, and then we keep the site live long enough to edit it the way an actual owner would — fixing a typo, swapping a project, changing a colour six weeks later. A launch screenshot cannot show you any of that. Ongoing use can.

What we actually measure

Four things, consistently, across every platform:

Build time, from a blank account to a published, working page with the full brief filled in — not to the first preview, which is where most reviews stop. We log this on first use, without shortcuts, because a returning user's speed says nothing about what a first-time visitor experiences.

Baseline page weight, the size of the published page with only the brief's content in it — no extra images added for padding, no plugins installed beyond what the builder ships by default. This is the number that predicts how the site behaves on a slow connection, and we cover the platform-by-platform results in detail in page speed in website builders, tested.

Mobile output, checked on an actual phone rather than a browser's device emulator, because emulators get spacing and tap targets close enough to look fine and wrong enough to matter. We also check whether the platform stores mobile styling separately from desktop or just scales one layout down, since that decision shows up the first time you try to fix something that only looks wrong on a phone.

Editing after launch. We return to every site at least once after publishing and make a change that has nothing to do with the initial build — add a project, change a headline, swap a photo — and time how long that takes against how long the first build took. A platform that is fast to launch and slow to maintain is a different product than its first impression suggests.

What we refuse to score

Three things show up in almost every other review, and we leave all three out on purpose.

Template galleries. Counting how many templates a platform ships tells you nothing about whether any of them fit your brief, and it rewards platforms for publishing more of the same few layouts under different names. We look at what the chosen template does once it's full of real content, not how many sibling templates sit unused next to it.

Uptime claims. Every vendor states a number, none of them let an independent third party verify it against their own infrastructure, and a headline percentage tells a reader nothing actionable. If we personally experience downtime during testing we say so, as a fact about what happened, not as a score.

Support response times. A single support ticket is one data point, and reviewers who score it as if it represents the average are reporting sample size one as if it were research. We mention support in prose when we hit something unusually good or unusually bad during a live build, but we do not convert one interaction into a number.

How long we spend before publishing anything

A minimum of two weeks per platform, spanning the initial build, at least one post-launch edit session, and enough idle time between sessions to see whether anything drifts — broken image links, an expired trial banner, a layout that shifts after an unannounced platform update. Comparisons that touch AI-generation flows specifically get a second build from a different brief, because a single run can succeed or fail on factors that have nothing to do with the platform's typical output. Our full roundup of the best website builders in 2026 is where all of these individual builds get weighed against each other, and it only includes platforms we've actually put through this process — nothing in it is written from a spec sheet.

What the same logic means for accessibility

The same principle — judge a platform on what its default forces onto you, not on what it advertises — governs how we test accessibility, too. It turned out to depend more on what a template forces onto you than on any accessibility feature the builder markets — the details are in accessibility in website builder templates.

Affiliate relationships

Some of the platforms in our comparisons pay a referral fee when a reader signs up through our link; some don't, including several we recommend. The build, the measurements and the draft conclusion all happen before we check which programs exist, specifically so the presence of a fee can't move a recommendation after the fact. Where a link is an affiliate link, the page it sits on says so. If a platform's affiliate program disappears, gets cheaper for the reader, or gets worse, we update the page rather than leaving stale terms live — which is also why most of our comparison posts carry an "updated" date separate from the original publish date, including this one.

That's the whole method: one real brief, built past the first screen, measured on four things we can actually verify, and left alone on the three things every other review scores without the data to back it up.