Durable test data

Each of the three fixed briefs was pasted word for word into Durable's 'Share more details' box, after its four short questions. The clock ran from pressing Start building to the moment the finished draft was on screen. The free plan cannot publish, so a live address was never reached.

The cafe draft showed an old placeholder (a carpentry site) at 57 seconds and the finished cafe copy by 81 seconds. The photographer's site was done after 70 seconds and the shop's after 60. The median is 70 seconds.

Durable asks four questions first: the kind of business, where it is, its name and a phone number. A fifth screen, 'Where did you hear about us?', can be answered or skipped. Then comes 'Share more details', where I pasted the whole brief. The brief's address, hours and prices were not asked for separately.

Each site came with a menu, a footer, a contact form, stock photos, and a floating cart and booking button. The site is 'Draft, not published' and has a private preview address.

• Times were read from my own session log, with the page checked every five seconds.

The twelve points were read in the HTML of the first drafts' preview address. The sitemap and robots.txt of a preview are not the ones a published site has, so the test is not graded.

9 of 12 points were there on the preview: an editable title and description per page, one H1 on every page, no skipped heading level, alt text on images, a canonical tag on all 21 pages, Organization structured data and social-card tags (Open Graph and Twitter) on every page.

The preview's robots.txt reads 'User-agent: * Disallow: /' and its sitemap is empty, and every page says 'noindex, nofollow'. A published site may differ, so I could not read them. Many descriptions ran long: 15 of 21 pages were over 155 characters. Every image file is named 'image' and has no width or height set.

The dashboard has a 'SEO & GEO' screen that reads 'Visibility 0 /100 Not visible' and lists four to-dos, the first being to publish the website, worth 10 points. Google Search Console and Bing (WB-S1) and AI-crawler rules (WB-S4) need a published site, so they were not run.

• Everything here is the preview address; a published site may add or change tags, a sitemap and robots.txt. The cafe was read a few minutes after I had changed its headline.

Each screen of the dashboard (Website, SEO and GEO, Customers, Money, Sell, Studio, Bookings, and the Blog, Leads and Reviews helpers) was opened as far as the free plan allowed. At each wall I recorded its words and the price to pass it.

Blocked: publishing and a custom domain (the plan page says 'Use your own domain or get one free'; Launch only), selling ('Upgrade to start selling. Unlock checkout and start receiving orders'), bookings ('Setup bookings', with 'Publish your website' beside it), the blog writer ('Out of credits', then 'You've reached your blog generation limit. Upgrade to continue generating blog posts.'), the lead and Google review helpers (both 'Disabled' and 'Out of credits'), and the Studio ('You're out of credits. Buy a credit pack or upgrade your plan to keep going.').

The Sell screen lists the four products I never gave, as 'Listed' with prices, and reads 'You're almost there. Publish your website so customers can find your shop.' The Customers screen is empty ('You don't have any customers, yet') and has Import contacts.

Form messages, a test order, tax, shipping and a mailbox could not be tried: the site is not live and checkout is locked. The Money screen offers 'Apply for an account' for a business bank account, and I did not apply.

• No domain was connected, no order was placed and no mailbox was made.

I looked for an export in the editor and in the settings, searched the help centre for 'export my website', and read the account settings and the plan page's guarantee.

I found no way to export the site's code, pages or images. The help centre's search for 'export my website' returned an article on exporting customers as a CSV file and one on viewing the sitemap, and nothing on the site itself.

The plan page says: 'If Durable isn't right for you, message us within 30 days of your first subscription payment for a full refund, including annual plans.' Its header says 'Try it risk free for 30 days'. I did not click past it, because the next step takes a card.

The free account's only way out is 'Delete account', under Profile: 'This account will be permanently deleted. You will no longer be able to access your account or any businesses associated with this account. This action is irreversible and cannot be undone.' There is no cancel path on a free account (WB-O1), and I did not press delete.

Restore an earlier version (WB-R1): not tested. A theme or colour change shows an 'Undo' note, and the AI panel says 'Use Revert to here on a previous message in this panel to undo while the session stays open'. I found no list of saved versions.

• Wording read on 28 October 2026. I did not read the terms of service in full.

I emailed Durable support one message with four questions on 28 October 2026 (the thread is logged in the evidence folder), opened the in-app help panel and searched the help centre five times. I did not search my inbox for replies.

No reply had arrived when I wrote this, so support is not graded. The help panel offers a help centre, an email address (support@durable.com), a live chat that says 'We typically reply in under 3 minutes' and a switch to give support access to your account. I did not use the chat.

Five help-centre searches (WB-H1): 3 of 5 questions were answered. 'Connect a domain I bought elsewhere' found 'Connect an Existing Custom Domain to Your Website'; 'form messages not arriving' found 'Receiving Website Contact Form Submissions', which explains where messages go; 'add custom code' found two articles. 'Change my credit card' found articles about team cards and images, and 'export my website' found a customer CSV export only.

The help centre's article on custom code and embed sections contradicts the AI helper's answer that raw HTML cannot be added; the article may apply to a paid plan.

• One email, sent once.

Every page of the three first drafts was loaded at 360, 768 and 1440 pixels wide on the preview address. The checker counted sideways scrolling, tap targets under 24 pixels, text under WCAG AA contrast (axe-core), placeholders, broken images and overlapping text, once per type per page. I then ran axe-core on every page at 1440 and 390 pixels, walked each page with the Tab key and compared the three drafts. The drafts were checked before I edited anything.

36 faults on 21 pages: 16 on the cafe (tap targets under 24 pixels and overlapping text on all 8 pages), 4 on the photographer (pale text on 3 of 5 pages and small tap targets on 1) and 16 on the shop (tap targets and overlapping text on all 8 pages).

The overlap is real. On a phone the round 'Built with Durable' badge sits over body text at the bottom-left of every page, and a cart and a Book a Table button float over the page. No sideways scrolling and no broken image was found at any width.

Accessibility (WB-D3): no critical problem on any page. The photographer's pages have serious colour-contrast problems on 3 of 5 pages (146 elements at desktop width, mostly white text on a photo); the cafe and shop have none. The cafe and shop product pages have no main heading (moderate). Walking the pages with the Tab key reached 157 of 216 focusable items on the cafe, 107 of 116 on the photographer and 159 of 215 on the shop; a focus ring was missing on 22, 23 and 22 of them at desktop width.

Look-alike check (WB-A2): the three sites share a layout (a big headline, a photo, three feature cards, customer quotes, a questions block and a footer) but differ in colours, photos and fonts (the photographer's site is set in Lora). They look like one family with a theme each.

Switching template after the content is in (WB-D1): possible and quick. Themes shows six options in about 28 seconds, each already holding my edited text. I applied the dark green one ('Professional Structure applied'): layout, colours and fonts changed, my headline and button text stayed. 'Confirm this theme' or 'Undo theme' ends it, and the top bar can show a phone and a tablet view.

• The pages were checked on the preview address, not a published site, and the checker flags the floating badge and buttons that a paid plan may remove. The tap-target and contrast counts follow the same rules as for Wix, Lovable and Hostinger.

Each brief has 20 facts. Every page of each first draft was read and each fact marked right, missing or wrong. Anything the AI added that the brief did not give was listed as invented: customer quotes, staff, products and claims. Generic marketing wording was not counted.

54 of 60 facts were right: 17 of 20 on the cafe, 18 of 20 on the photographer and 19 of 20 on the shop. 4 were wrong and 2 were missing.

Wrong: the cafe said 'open Tuesday to Sunday' where the brief gave Monday to Friday and Saturday to Sunday, and its home page priced the loaf at $9.50 where the brief said $9.00; the photographer's booking page headline said '90 Bluebonnet Court' (the address further down was right); the shop's home page priced the Speckled Mug at $38 where the brief said $34. Missing: the cafe's Saturday and Sunday closing time, and the photographer's $1 per mile beyond 100 miles.

Made up: 16 named customer quotes on the cafe (for example 'Alex Thompson, Regular since 2019'), 8 on the shop (four signed 'Person 1' to 'Person 4' and four from 'Collector in Raleigh' and other cities), a 'Marta Ruiz, Associate Photographer' and a 'Jordan Lee, Studio Assistant', 'Catering for 40, done 100 times over', farm partners, and four products I never listed (two gift cards, a home latte kit and a ripple bowl set).

The brief said 'don't add reviews, awards or staff we haven't mentioned' on all three sites. Durable added reviews and staff anyway. The photographer's site had no made-up reviews.

• Claims count when they state something about the business the brief did not give. Words such as 'worth the trip' were treated as style.

The twelve fixed edits were done on the cafe draft in Durable's editor and checked on the preview address. Edits the toolbar could not do were tried with the AI helper (the 'AI Edit' button on each section), which uses AI credits. Five written instructions (WB-A5), the template switch (WB-D1) and the credit meter (WB-A6) were run as added checks.

11 of 12 edits worked on the cafe: move a section (1 click, about 3 seconds), site font (4 clicks, about 25 seconds), site colours (6 clicks, about 40 seconds), a FAQ page with a menu link (about 60 seconds to make the page; the menu link needed 8 clicks and its label stayed 'Link'), a footer line (AI, 22 seconds), a form field (AI, 28 seconds), a button link (AI, 17 seconds), a replaced image (3 clicks, about 4 seconds), a duplicated section (AI, but it copied the intro block, not the section I pointed at), a phone-only change (AI, 25 seconds; the line is hidden at 390 pixels and shown at 1440) and a rewritten headline (AI, 18 seconds).

With no credits left, an AI edit opened 'You need more credits. This costs more than you have left. Upgrade or top up to keep going.', with 'Move up to Launch, 1,000 credits every month' and 'Credit packs from $5, They never expire'.

Embedded HTML failed. I asked for it twice. The AI answered: 'Raw HTML cannot be injected here: the design system has no HTML/embed primitive, so the section is left unchanged.' The plan page lists 'custom code' under Launch.

AI credits (WB-A6): the free account showed '70 of 100 credits used' after the three builds, six AI edits, a new page and the theme options, and '100 of 100 credits used' after three more AI edits, with 'Free credits reset tomorrow'. Median time per edit was about 20 seconds with the AI helper and about 25 seconds with the toolbar (WB-E1).

Free placement (WB-E2): not offered; sections move up and down only. A blank page cannot be started, because the AI always builds from a description.

• The AI helper's edits cost credits and I could not see the cost of one edit; the meter only shows the total. The crop tool (Adjust) was not run.

Lighthouse 13.5 with mobile settings ran three times on the home page and one inner page of each site, on the preview address (18 runs), and three times on the desktop setting (18 runs), with the computer otherwise idle. The median score of the 18 mobile runs is the grade. The pages are the first drafts: the cafe home and menu, the photographer home and pricing, and the shop home and shop pages.

The median of the 18 mobile runs was 48. By page: photographer home 56, photographer pricing 52, shop home 49, cafe menu 47, shop page 46, cafe home 45. The desktop median was 88.

Phones waited long for the main picture: Largest Contentful Paint medians ran from 6.3 to 9.6 seconds. Pages weigh 1.7 to 2.1 MB, and Total Blocking Time ran from 486 to 2143 milliseconds. Layout shift was small (0.009 to 0.022).

Lighthouse also scored accessibility 96 to 100, best practices 77 (100 on some runs) and SEO 69, because the preview is marked noindex.

• Measured on the free preview address, which answers with a noindex header. A published site may be served differently, and the speed from three regions (WB-P1) was not run. A first set of runs, made while I used the machine for other tests, scored a little lower; the idle set is the one graded.

Every plan was read on Durable's pricing page and on the in-app plan screen that opens from Publish. The figure graded is the yearly cost of the cheapest plan that connects a custom domain and holds all the pages in my briefs.

The free plan is $0: a preview, 'Add up to 10 customers' and 'Basic help center support'. The pricing page says '300 credits to try Durable's AI', but the app showed 100 credits that reset the next day.

Launch is $22 a month on yearly billing and $25 a month on monthly billing. Grow is $41 or $49. Launch has a domain (or one free, up to $20), 1000 AI credits a month and bookings; Grow adds 5000 credits, a $40 domain credit, a call with an expert, unlimited team members and translation.

In the app a '40% off your first year' offer with a four-hour countdown showed Launch at $15 a month, but the line under it read '$255 billed yearly' (15 times 12 is 180). Grow showed $29 a month and '$492 billed yearly'. I did not find which price is charged; I did not pass the plan page.

I graded 255 dollars, the billed yearly figure for Launch; 264 dollars is 12 months at $22. Renewal prices and an itemised bill (WB-M2) need a purchase.

• I paid nothing. The price shown beside '$255 billed yearly' did not add up, so the graded figure is the billed one.