AI website launch checklist

Check seven things before an AI-built site goes live: facts, accessibility, security, speed, AI search, ownership and legal pages. AI gets these wrong because they're invisible on the screen. Thirty checks below, with the numbers to aim for. Most take minutes with free tools.

This is step four of my guide on how to use AI to design a website.

Why does an AI-built site need its own checklist?

Because it looks finished before it is. The layout is polished, so nobody looks underneath. Three numbers show what's usually there:

Are the facts on the site true?

1. Every claim has a source. Numbers, awards, client names, years in business.
2. No invented testimonials. Delete any review you didn't receive.
3. Contact details are real. Phone, email, address, opening hours.
4. Prices and terms match your contracts, or are removed.
5. Placeholder text is gone. Search the site for "Lorem", "[", "TODO" and "Example".

Can everyone use it?

WebAIM finds the same six errors every year. Low-contrast text was on 83.9% of home pages in 2026, and missing alt text on 53.1%¹.

1. Contrast at least 4.5:1 for body text, 3:1 for large text.
2. Alt text on every meaningful image. Decorative images get empty alt.
3. Labels on every form field.
4. Links and buttons have readable names, never an empty icon.
5. Page language declared in the HTML.
6. Keyboard. Tab through every page. Focus is always visible.

Run WAVE or axe on each template, then fix and run again. If you sell to EU consumers, the European Accessibility Act has applied to e-commerce services since 28 June 2025⁴.

Is it secure?

AI writes code that works long before it writes code that's safe. In Veracode's 2025 tests, models failed to defend against cross-site scripting in 86% of relevant samples².

1. No secrets in the front end. API keys live on the server or in the platform's vault.
2. Database rules on. Row-level security or the platform's equivalent.
3. Forms validated on the server, with spam protection.
4. Logins use the platform's auth, never custom code the AI wrote.
5. Dependencies scanned. Run the platform's audit or npm audit.
6. A developer reviewed any login, payment or data flow.

Is it fast?

Google's Core Web Vitals set the targets for a good experience⁵.

1. Test in PageSpeed Insights, on mobile.
2. Images compressed and served at display size, in WebP or AVIF.
3. Unused scripts removed. If you can't name what it does, test without it.
4. Fonts limited to two families, with display swap.

Can AI search read it?

People click less when AI answers first. Pew found clicks on result links fell to 8% of visits when Google showed an AI summary, from 15% without one⁶. The site has to be the source that answer quotes.

1. Answer first. Each page opens with a plain 40 to 60 word answer.
2. Question headings that match how people ask.
3. Structured data. Organization, Service, FAQPage, BreadcrumbList, matching the visible text.
4. One name, one description, used the same way across the site and your profiles.

Context systems go further, with entity graphs and retrieval built for AI. See context systems and how to get your site cited by ChatGPT.

Do you own it?

1. Domain registered in your name, with renewal on.
2. Code exported or synced to a repository you control.
3. Backups and version history turned on.

Are the legal pages in place?

1. Privacy policy and cookie consent that match the tools the site actually loads.
2. Terms and company details, including your registry code where the law asks for it.

If this list feels long for one person, that's normal. I run these checks with teams as part of website design. You can also turn any failed check into a fix with the prompts in AI website design prompts.

Questions people ask