An AI site is a draft until you verify ownership and the form
AI can assemble a tidy website in an evening. That is dangerous: you can end up with six pages of the same phrases, invented benefits, and zero understanding of how the company actually helps.
The mistake I would remove first: Publishing made-up case studies and reviews.
What to prepare and what result to expect
- Outcome: You know what can stay in the AI site, what to rewrite, and what to reject before publication.
- List the required pages, URLs, fact sources, domain/hosting access, form, analytics, and export method.
- Keep source data and access rights separate from the output so you can verify what the AI site produced before publication.
- Check the source HTML, mobile view, forms, canonical, sitemap, speed, domain ownership, and whether you can export the content.
Run checks before publication
- Fact-check the content
Verify the name, services, prices, address, phone, timelines, photos, and case studies against real data. Remove invented reviews and promises.
Проверьте: Every commercial claim is confirmed by the business owner.
Если не сработало: Replace disputed copy with a client question or a draft note.
- Verify technical ownership
Domain, DNS, hosting, repository, analytics, and the form must be under company access—not only on a contractor or builder account.
Проверьте: Another specialist can get access without the layout author’s permission.
Если не сработало: Secure ownership first, then enable a paid plan.
- Check the SEO basics
Open the source and find title, description, H1, canonical, robots.txt, sitemap, and the service copy. Make sure URLs can change without breaking the structure.
Проверьте: The page is available without JS and has one clear intent.
Если не сработало: Request an export or choose a stack with readable HTML.
- Test the lead path
From a phone, submit the form and check the notification, CRM, UTM, and post-submit message. Open the site in a private window and verify the domain.
Проверьте: The test lead reached the owner and includes the source page.
Если не сработало: Keep the form manual and do not turn on ads.
I give the model facts first, not “make it pretty”
The brief should include the service, audience, region, real timeline, price or an honest “we don’t publish it,” constraints, and proof. In the prompt, ban inventing clients, numbers, certificates, timelines, and guarantees; return missing facts as `needs_clarification`.
Publish one service page first, not a hundred variants for word swaps. After publication, open Search Console → `URL Inspection` and check canonical, noindex, and sitemap. On staging use `noindex,nofollow`, and remove it before release.
You write a service page only from these facts.
Do not invent numbers, clients, timelines, or guarantees.
If a fact is missing, return needs_clarification.
Return: title, description, H1, structure, 5 questions, and one CTA.I do not score by pages generated
I check manual fact-checking, a unique example, search impressions, query-to-offer match, and organic leads. In GA4 I fire `generate_lead` only after a successful form submit, verify it in Realtime, and only then mark a key event. Without facts, cases, and a clear offer, AI only masks emptiness.
I fact-check before publication
Build a checklist: service name, timelines, price, geography, constraints, case studies, contacts, and legal wording. Every item needs an owner or a proof link. Return AI copy without a source as `needs_clarification`.
For the first 28 days measure impressions, queries, organic sessions, `generate_lead`, and lead quality—not blocks created. If the page gets traffic for the wrong intent, rewrite the promise and structure instead of piling on more keywords.
When an AI builder helps—and when it gets in the way
| Criterion | Question | Good sign |
|---|---|---|
| Input | What exactly goes into the AI site before publication? | List the required pages, URLs, fact sources, domain/hosting access, form, analytics, and export method. |
| Action | What is the system allowed to do on its own? | Only pre-listed actions, without access to the entire account |
| Verification | How do you know the result is acceptable? | Check the source HTML, mobile view, forms, canonical, sitemap, speed, domain ownership, and whether you can export the content. |
| Failure | Where does an unclear case go? | Do not publish the page if you cannot fix facts, get domain access, or verify where leads go. |
What should change after setup
You know what can stay in the AI site, what to rewrite, and what to reject before publication.
What a pretty generator does not check for you
Publishing made-up case studies and reviews.
Handing domain and form access to a contractor’s personal account.
Treating a pretty preview as a working HTML site.
Not checking export and site edits after launch.
When a template or custom build is the better choice
You need a specialist if the AI site must integrate with a CRM, support many locales, include a catalog, or run complex custom logic.
What to check on a staging domain
Who is this approach to an AI site before publication for?
An AI builder speeds up the draft, but it does not check whether the services are true, whether the form works, or whether you can change the site in six months. Judge ownership and quality, not the preview image.
Where do I start if everything is still done manually?
List the required pages, URLs, fact sources, domain/hosting access, form, analytics, and export method.
How do I check that the setup will not cause harm?
Check the source HTML, mobile view, forms, canonical, sitemap, speed, domain ownership, and whether you can export the content.
What if the result is unclear?
Do not publish the page if you cannot fix facts, get domain access, or verify where leads go.
