Our process
Every site and landing page goes through the stages a professional web studio follows, from a written brief to a handover with no trace of us in the files. AI models do the work, and every stage is checked by an automated test, a separate AI review against a written standard, or both. Anything that fails is fixed, or explained in writing, before you see it.
Each stage of the job, and what checks it
AI models do the work at every stage below, and nothing moves on because the model that built it says it’s right.
- Automated test. Software measures the result against a fixed pass mark, the same way on every job.
- AI review. A separate AI reviewer, started fresh, works through the written standard and gives evidence for every item. It never sees what the AI that built the work thought of it.
Brief and scope
Your answers to the scoping form become a written brief. Anything missing or unclear comes back to you as a written question, and the quote sets the scope, price, delivery date and revision rounds before work starts.
What checks it
- AI review of the finished work against this brief: every page, section and state it asked for, and nothing it didn’t.
Structure and content
The pages, their sections and their order are planned from the brief. If you don’t supply the copy, it’s written from the brief and your client’s current site, with no invented facts, reviews or figures. You check every fact about your client before anything is published.
What checks it
- AI review that every claim has a source in the brief or your client’s own material, nothing is invented, and every page has a clear next step.
Typography, grid and colour
Each page uses at most two typefaces on a set scale, with lines of about 45 to 75 characters. The grid holds its alignment at every width, spacing comes from one scale and the palette is small. We don’t use templates, stock photos of people or decorative clip art.
What checks it
- Contrast test (axe): text meets the WCAG 2.2 AA minimums of 4.5:1, or 3:1 for large text.
- AI review of the rendered screenshots at every width, against a design standard that lists the templated looks that fail on sight.
Responsive build
The pages are built to work from 320 px phones up to wide desktop screens, with touch targets of at least 44 px and nothing that works only when a mouse hovers over it.
What checks it
- Screenshot test (Playwright) of every page at 375, 768, 1280 and 1920 px in three browsers, flagging any page that scrolls sideways or shows a script error.
- AI review in a real browser of touch targets and anything that relies on hover.
Accessibility
Built to WCAG 2.2 AA: headings in order, page landmarks, labelled form fields and clear error messages, alt text that says what an image is for, nothing shown by colour alone, and motion that respects the reduced-motion setting. This is not a formal conformance audit.
What checks it
- Accessibility test (axe) on every page: zero violations, apart from any exception you agree to in writing, which the results list.
- AI review in a real browser: every page by keyboard alone with visible focus, at 200% zoom and at 320 px wide.
Speed
Images are sized to their space and saved in modern formats, fonts are trimmed to the characters needed and loaded early, unused scripts are removed, and pages stay under 500 kB where possible.
What checks it
- Speed test (Lighthouse) of every page on a simulated slow phone: performance 95 or more, main content shown in under 2 seconds, little jumping about as it loads (layout shift under 0.05) and quick to respond to a tap (blocking time under 150 ms), or a written reason for any page that falls short.
Technical SEO
Every page has a title and description, and on each page meant for search no other page shares them. Each of those pages also has one preferred address (a canonical address), so search engines don’t count copies. Tags set how a link to each page looks when shared. The site has a sitemap, robots.txt and, where it fits, structured data: machine-readable details, such as the business’s name and address, that search engines can show.
What checks it
- HTML test (html-validate): valid HTML on every page.
- Link test of the built site: no broken links or missing files.
- Search and best-practice test (Lighthouse): best practices 100 on every page, and search 100 on every page meant to be found.
Browser and screen-size testing
The screenshot test captures every page in all three browser engines: Chromium (Chrome and Edge), Firefox and WebKit (Safari), at phone, tablet and desktop widths.
What checks it
- Screenshot test (Playwright), the same run as in stage 4.
- AI review of every screenshot for clipped text, overlaps and awkward line breaks.
Copy editing
Copy is in Australian English, plain and specific. Stock marketing phrases and filler are cut, and every number has a source.
What checks it
- Spelling test (cspell) of every page’s visible text against the Australian English dictionary; common American spellings fail.
- AI review against the copy standard and its list of banned stock phrases.
Final review
A separate AI reviewer, which didn’t build the work, gets only the work, your brief, the written checklist and standards, the screenshots and the test results. It never sees the opinion of the model that built it. It marks every item pass or fail with evidence, and anything it couldn’t check is reported as not checked, never as a pass.
What checks it
- Every test above runs again after the fixes.
- AI review again on each fix until the list is clear. Anything left unfixed comes with a written reason.
Handover
You get the files and the test results. Nothing in the files names Blindstamp or the AI models used.
What checks it
- Leak scan of every delivered file, file name, image, font and their metadata for Blindstamp’s name, the name of Blindstamp’s owner, and the AI model’s name and its maker’s. One hit fails it.
Who does what
AI models do the building, the writing and the checking at every stage above. Blindstamp approves every quote and price. For a build, the test results come to you with the files.
AI builds quickly. Left to itself, it also turns out the generic, slightly wrong work people have learned to spot. So every job is held to a written standard, and a separate AI that didn’t build it does the review.
Audits and fixes
An audit follows its own standard. Every finding is reproducible, with the tool, the page, the measured value and the date, and every Lighthouse figure it quotes comes from the median run of three. The finished report gets the same separate AI review and leak scan. A fix starts by reproducing the problem. For files we change, the same tests then run again on every page the fix touches; a fix on a platform comes as instructions that say what to re-check. Either way, a separate AI reviews the result too.
How a job runs
- You send the scoping form. It asks only what the job needs, and the answers arrive as an email from you.
- You get a quote by email. It sets the scope, price, delivery date and revision rounds. Work starts once you accept it, and on jobs over A$500 once the 50% deposit is paid.
- The work is done and checked. If a question comes up while it’s under way, it comes by email.
- You get the files, and for a build, the test results. Revisions are asked for and answered in the same thread.
- You’re invoiced at sign-off. The sign-off invoice comes through Stripe, due within 7 days. The copyright we hold in the work passes to you once the job is paid in full. Material made mostly by AI may not be protected by copyright under Australian law, so we assign what rights we hold and don’t promise every part is protected.
White-label terms, in brief
- Nothing in the delivered files mentions Blindstamp or the AI models used: no credits, comments, meta tags, file names or document properties.
- We never contact your clients, during a job or after it, apart from a test message through a form on a client’s site, sent only with your OK in writing.
- Your briefs, your clients’ names and everything you send are kept confidential and used only for your jobs.
The full terms come with your first quote. Terms and privacy has a summary.
Start with one audit
The audit is the smallest job: one site, every finding with its evidence, and a fix list ranked by what matters most to the business, as a PDF under your agency’s name. It’s a fair way to judge the work before you send anything bigger. The sample audit shows what you’d get.
A$295. No GST is charged, because we’re not registered for it.
Start with an auditor write to hello@blindstamp.com.au