The worst launch-day problems are rarely spectacular. They are a typo in a headline, a pricing page that links to a 404, a form that does not reach the CRM, a hero image with no alt text, or a page title that still says “Home copy”.
Each one is small. Multiplied across a two-hundred-page site, they add up to a first impression you did not intend.
This is exactly the kind of work where AI tools have become genuinely useful.
Why QA gets squeezed
QA happens at the end, when deadlines are tight and everyone has been staring at the same pages for weeks. People stop seeing mistakes they have read twenty times. Checking every page by hand for every issue is slow, boring and error-prone.
Machines do not get bored. That is the opening for AI.
What AI does well in QA
Copy and consistency
- Spelling, grammar and tone across every page.
- Inconsistent terminology: “sign up” on one page, “register” on another.
- Mismatched facts, such as a different phone number or price in two places.
- Placeholder text that slipped through.
Links and metadata
- Broken internal and external links.
- Missing or duplicated page titles and meta descriptions.
- Missing structured data or social sharing images.
Accessibility basics
WebAIM’s 2025 analysis found that six issue types, led by low-contrast text and missing alt text, account for about 96% of detected accessibility errors on home pages (WebAIM). These are exactly the patterns automated and AI-assisted tools are good at spotting across a whole site.
Summarising audits
Automated tools produce long reports. AI can turn them into a prioritised list: what is broken, where, how serious it is and which template causes it.
What still needs people
- Keyboard and screen reader testing. Automated tools cannot tell you whether an experience makes sense.
- Real devices. Especially mid-range phones and less common browsers.
- Forms and integrations end to end. Submit the form, check the CRM, check the email.
- Judgment. Is this claim accurate? Is this the right image? Does this page do its job?
Nielsen Norman Group’s research on AI productivity found the best results came from humans and AI working together, with AI producing drafts and people editing and checking (NN/g). QA is no different.
A pre-launch checklist
Content
- All copy proofread; no placeholder text
- Facts, prices and contact details consistent everywhere
- Legal pages and cookie notices in place
Links and SEO
- No broken links; redirects in place for changed URLs
- Unique titles and descriptions on every page
- Structured data valid; XML sitemap generated
- Search engine visibility set correctly for launch
Forms and tracking
- Every form submits and reaches the CRM
- Confirmation messages and emails work
- Analytics and consent tools working together
Accessibility
- Automated scan passed or issues triaged
- Keyboard navigation and visible focus checked
- Alt text and form labels present
Performance and devices
- Key templates tested on mobile and desktop
- Core Web Vitals within target on key pages
- Tested in the main browsers your audience uses
Make QA continuous
Google’s DORA research highlights working in small batches and strong version control as foundations for turning AI speed into reliable delivery (DORA 2025). The same principle applies to websites: smaller, regular releases with a quick QA pass are far safer than an annual big-bang relaunch.
The takeaway
AI will not make your QA team redundant. It will make them faster and more thorough, by taking over the repetitive checks so people can focus on the things that need judgment.
Pre-launch QA is part of every website project we deliver. For keeping a site healthy afterwards, see our zero-downtime migration checklist.
For the bigger picture on vibe coding, and whether AI will replace web developers, read why developers still matter in the age of AI.
