Launch checklist
AI Website Launch Checklist: What to Test Before Publishing
By Qlinnt · Updated
Before publishing, check that visitors can complete the main task, search engines can access the intended pages, and private configuration stays private. Then test the deployed version on mobile and verify each important fix. A successful build alone does not cover these checks.

Start with the task the visitor came to complete
Choose one real journey: request a quote, create an account or buy a report. Walk through it from the public URL in a fresh browser session. Test an empty field, an invalid email and an unavailable service. A button that looks finished can still lead to a placeholder or a silent error.
For forms on your own site, use test data you control. If payments are not available, say so before the visitor reaches a checkout. Ask a second person to try the flow without your instructions; their confusion is useful evidence.
- Complete one meaningful journey on the deployed site using controlled test data.
- Check empty fields, invalid input and a service failure; make every error visible.
- Keep payment actions unavailable while payments are paused.
Check which pages should appear in search
Give each important public page a distinct title, one clear main heading and a canonical URL on your production domain. Check the deployed HTML and response headers for accidental noindex directives. A preview configuration can survive a production deployment.
Include only the canonical, public pages you want indexed in the sitemap. Keep private reports, authentication links and test pages out. Submit the sitemap in Google Search Console, then inspect an important URL. A discovered URL is not necessarily an indexed URL.
- Inspect the canonical URL, title and accidental noindex in HTML and headers.
- List only intended canonical public pages in the sitemap.
- Keep scan reports and recovery links excluded from search; verify indexing in GSC.
Use a real mobile journey, not just a smaller screenshot
At a narrow screen width, read the headline, open navigation, fill out the form and reach the result. Look for clipped labels, horizontal scrolling and buttons hidden by the keyboard. Use the keyboard on desktop too: every action needs a reachable control and a visible focus indicator.
Use PageSpeed Insights to investigate loading, layout shifts and interaction delays. Large images, third-party scripts and fonts are common places to investigate. Treat a lab score as a diagnostic signal; it is not a measurement of every visitor’s experience.
- At 390 px, reach and use the primary action without horizontal page scrolling.
- Use Tab, Enter and visible focus; label controls and announce errors.
- Investigate slow resources and layout shifts; distinguish lab scores from field data.
Inspect public configuration without sharing secrets
Confirm that server credentials remain on the server and are absent from public JavaScript and deployment artifacts. A public Supabase anonymous key is not equivalent to a service-role key; database access policies still need their own review. Do not assume every key-like string is a confirmed leak.
Check only sites you own or are authorized to test. If you discover an exposed credential, restrict the exposure and rotate or revoke it with the provider. Removing it from a page does not invalidate copies that may already exist. Never paste the secret into an AI prompt or a public issue.
- Review public JavaScript and configuration for genuine private credentials.
- Do not treat an anonymous Supabase key as proof of a secret leak or an RLS failure.
- Review private database policies and authenticated behavior separately.
Verify the deployed fix and document the remaining limits
Keep a short record of the affected URL, the observed behavior and the expected result. Change one related group of issues, deploy, and repeat the same steps. For a canonical error, inspect the new HTML; for a broken form, submit controlled test data again.
Qlinnt can inspect a public URL and up to four priority pages on the same origin. It cannot test a logged-in journey or inspect your source code. Use its findings to prioritize work, then combine them with manual testing and Search Console. “Not verified” means a check could not establish a result, not that the site passed.
- Record the exact page, observed problem and expected correction.
- Deploy the change and request a fresh measurement.
- Require a completed pass of the same check on the same page; a missing result is not a fix.
Is a website made with AI automatically bad for SEO?
No. The relevant questions are whether its pages are accessible to search engines and useful to visitors. Review the deployed site and its content rather than judging it only by the tool that generated it.
Does a clean automated scan mean my site is ready?
It means the checks that ran did not find the issues they look for. You still need to test real journeys, private areas and business-specific behavior manually.
Sources and further reading
Check your own public website
Start with a free scan. Verify your email to read every finding explained. Evidence and fix prompts are not included in the free diagnosis.
Scan my website