How we test: one brief, five dimensions, no exceptions

How we test AI website builders comes down to a single discipline: every tool answers the same question, and we grade the answer the same way. No bespoke briefs that flatter one product's strengths, no scoring criteria invented after the fact. This page documents the whole method so you can check our work.

The standard brief

Every builder is asked for the same thing: a five-page website for a fictional local coffee roaster. Home, menu, about, wholesale, contact. We chose it because it's the most ordinary brief imaginable — and ordinary is where these tools actually get used.

Each page stresses something different. The home page tests whether the tool can produce a hero, a value proposition, and a brand voice that isn't beige. The menu tests structured, repeating content — items, descriptions, prices — which template-driven generators frequently fake with a photo grid. The about page tests long-form copy and imagery. The wholesale page is the trap we're proudest of: it demands a form and a business-to-business pitch, and weaker flows quietly replace it with a generic 'Services' section. The contact page tests hours, a location, and the concrete details generative tools are most tempted to invent.

We evaluate against the brief at the level of documented, verifiable behavior: what the tool's generation flow accepts as input, what page structures it can produce, what its plans and published limits allow, and where the result can actually be published. When a claim would require a measurement we didn't make — load times, uptime, performance scores for generated sites — we don't make the claim.

The five scoring dimensions

Each builder's rating out of 10 is built from five dimensions, weighted equally. A tool can't buy its way up one column to hide a disaster in another — a 9 in design with a 4 in price honesty lands exactly where you'd expect.

Prompt fidelity

Did we get what we asked for? Five pages means five pages; a wholesale page means a wholesale page. Tools that honor the structure of a brief score well here; tools that route every input into the same template do not.

Design quality

Would a working designer wince? We assess typography, layout, spacing, and how far the output sits from the obvious template — judged from the builders' own generated results and documented design systems, not from marketing screenshots.

Editability

Every AI draft needs human edits — so how painful are they? We weigh what the editor lets you change, whether AI-generated sections stay editable after generation, and how much the tool fights you when you want something it didn't suggest.

Publishing & hosting

Can you actually ship? We check what each plan publishes to (subdomain versus custom domain), what branding rides along, bandwidth and visitor caps, and whether the free tier produces a live site or a hostage.

Price honesty

The dimension the industry earned. We compare the advertised price with the real cost: intro rates that require multi-year prepays, renewal jumps, credit systems with expiry rules, and the gap between monthly and annual billing. A builder can be cheap and still score badly here if finding the true price takes forensic work.

How often we retest

This category moves fast — plans get renamed, credit systems replace message counts, and entire products get deprecated mid-cycle. So every price and plan detail on this site carries a verification date, and we re-verify pricing against vendors' live pages before each major update to the best AI website builders ranking — most recently in August 2026. When a builder ships a meaningful change between passes, we revisit its report rather than waiting for the calendar.

Anything we couldn't verify is labeled that way in the copy — you'll see phrasing like 'around $20/mo as of our last check' wherever a vendor doesn't publish a number plainly. We'd rather look uncertain than be wrong.

Why every report has caveats

A review with no cons is an advertisement. Every tool on our bench has flaws — page caps, renewal cliffs, expiring credits, deprecated predecessors — and we print them next to the strengths, because the caveat is usually the fact that decides the purchase. If a report here ever reads like a brochure, write in and tell us; that's a defect.

Where the money comes from

Some outbound links on this site are affiliate links, always routed through /go/ so you can spot them. If you buy through one, the vendor may pay us a commission at no extra cost to you. The firewall is simple and absolute: scores and rankings are set by the method above, and no commission, partnership, or vendor conversation can move them — several tools we link to pay us nothing at all, and they're ranked by the same rules. The affiliate disclosure spells out the details, including which builders we currently have relationships with.