AI-built websites usually look finished before their public signals are ready. The right launch checklist starts with what crawlers, link preview systems, AI assistants, and browsers can fetch without credentials.

AI readiness signals

Publish llms.txt when you have official docs, tools, policies, or product pages that assistants should inspect first. Keep it short and link only canonical production URLs. If you serve markdown variants, return them only when Accept: text/markdown is explicitly preferred and include Vary: Accept.

Technical SEO and sharing

Every public route needs a unique title, meta description, canonical URL, and Open Graph preview. The canonical URL in HTML should match the sitemap and any llms.txt reference. Do not let staging, preview, or localhost hosts leak into production metadata.

Security and transport

Serve HTTPS, redirect HTTP to HTTPS, add HSTS once subdomains are ready, and ship a Content-Security-Policy header. At minimum, define default-src, base-uri, object-src, frame-ancestors, and a deliberate script-src strategy.

Performance evidence

Check the landing page on a real public URL. Core Web Vitals and Lighthouse diagnostics are not the whole launch story, but large layout shifts, slow LCP elements, and render-blocking assets are visible quality problems.

Verify with a scan

Run the website readiness checker before launch. It gives a public report with the exact evidence rows, priority fixes, and copyable remediation prompts you can hand back to your coding agent.