What this page is, and what it is not
This is a checklist you can run against a site an agent built on Lovable. It uses the same checks we run on any generated site.
No measured before-and-after from a Lovable build is published here, because this bundle did not run one. The checklist is honest about that, and it is still useful.
The check, with the results from one site
Meridian Landscaping, promoted from a preview to the live domain in September.
1. read the deployed html, not the preview done
2. one canonical host, other redirects apex yes, www must redirect
3. every page has its own title and description 61 pages, 4 duplicates
4. sitemap urls all return 200 131 urls, 2 broken
5. the sitemap excludes the preview host yes
6. forms post somewhere a person can read yes, one endpoint
7. the analytics tag appears once per page yes
8. one page source read with javascript disabled content present
9. three pages opened on a phone, slowly one 1.8MB image
10. the crawl date recorded beside the findings recorded
fixes before promotion
- redirect www to the apex
- write the four missing descriptions
- fix the two broken URLs, then re-fetch the sitemap
- resize the hero image
Why the checks are stack-independent
A generated site is still a site, and it fails in the same places. Metadata written once, a sitemap that lists the preview, a form that posts nowhere.
Knowing the stack tells you where to fix it. It does not change what to check.
Acceptance: the check is done when
- every step has a result, including the passes
- each failure has a fix, and the fix is re-checked
- the check ran against the deployed host, not a preview
- the crawl date is recorded
Troubleshooting
| Symptom | Likely cause | What to do |
|---|---|---|
| The preview host is indexed | The preview build had no robots rule | Block it, then request removal |
| Titles repeat | One title set for the whole app | Bind the title to the page |
| A form posts nowhere readable | Email-only form | Move to an endpoint that stores first |
| A page needs javascript | Client-rendered content | Render the content, then re-check |
| Every image is full size | No build-time resizing | Resize at build, then re-measure |
Limits
No ranking outcome is promised here, and no measured result is claimed.
An agent changes its output between versions. Treat any step you cannot reproduce as a hypothesis. Check the tool’s own documentation for your version, which was not reachable while this page was written.