Your AI-built website looks finished. That doesn’t mean it’s ready.

We’ve already said the part most web agencies avoided saying for too long: Can ChatGPT actually build a website? Yes — and that was never the hard part.

The code isn’t the impressive part anymore. Give a capable AI tool a decent prompt and it can produce a modern homepage, responsive sections, service pages, a contact form and styling that looks respectable enough to launch.

That creates a different problem. The website can look finished long before it’s actually ready.

A clean homepage tells you almost nothing about whether the old pages were accounted for, the forms reach the right person, the claims are true, the mobile menu works, the analytics survived, or Google can still find the pages that used to rank.

AI is extremely good at producing visible progress. It’s less reliable at noticing work that was never included in the prompt.

Before you replace your current website with an AI-built version, inspect the parts that are easiest to miss and most expensive to discover after launch.

Start with the old website, not the new homepage

Most people review a redesign by opening the new homepage and asking whether it looks better. That’s the wrong comparison.

The real comparison is between everything the current website does and everything the rebuilt website still does.

Your old site may be ugly and disorganized, but it probably contains more than the pages visible in the main navigation. There may be service pages, location pages, blog posts, PDFs, quote forms, thank-you pages, scheduling links, advertising landing pages, tracking scripts and URLs that quietly receive search traffic.

An AI builder will happily create the pages named in the prompt. It won’t necessarily stop and investigate the rest of the site unless someone gives it the crawl data, the analytics, or a direct instruction to do so.

That’s how a redesign quietly deletes parts of the business.

Before anyone debates colors, animations or photography, get an inventory of the current site and make a decision about every meaningful URL. It should be kept, combined, replaced, redirected or deliberately removed.

“Wasn’t included” isn’t a migration strategy.

Submit every form yourself

Don’t accept a screenshot proving that the form displays correctly. Submit it.

Use a real email address and phone number. Complete the process from the same kind of device a customer would use. Then follow the lead all the way through the system.

Did the visitor see a useful confirmation? Did the notification reach the right person? Did the contact enter the CRM? Was the source recorded? Did the correct automation run? Did the conversion event fire? Did the sales team receive enough information to act?

AI can create the visible portion of a form in minutes. That doesn’t mean the form is connected to anything.

The same problem appears with scheduling buttons. The button opens a calendar, so everyone assumes the job is finished. Nobody checks whether it’s the right calendar, whether availability is accurate, whether the confirmation email makes sense, or whether the appointment reaches the correct team member.

The question isn’t whether the button works. The question is whether the business process works after someone clicks it.

A form can also be wired correctly and still lose leads — a scoring-based bot check that nobody acts on, or a notification sent from the visitor’s own address. Contact form not sending leads? Check these two things covers both.

Read every claim as if an angry customer will challenge it

AI fills gaps. That’s useful during brainstorming and dangerous when publishing factual statements about a business.

A vague note about fast service becomes “same-day service.” A regional company becomes “serving customers nationwide.” An experienced team becomes “more than 20 years of experience.” An occasional service becomes a core offer available to every customer.

Those claims won’t look suspicious inside polished website copy. They’ll sound exactly like the claims on every competing website.

Someone who knows the business needs to verify the service areas, experience, certifications, guarantees, response times, financing options, team credentials, statistics, testimonials and results.

Don’t ask the AI that wrote a claim to verify whether the claim is true. It’s working from the same incomplete context that created the problem.

Ask where the information came from. If there’s no source, no confirmation and no accountable person behind it, rewrite it or remove it.

Check whether every page has a job

AI-generated websites frequently contain all the expected sections and still have no clear conversion path.

The homepage introduces the company. The service page explains the service. The about page describes the team. The contact page displays a form. Everything exists. Nothing builds toward a decision.

Open each important page and ask one question.

What should a qualified visitor do after reading this?

The answer can’t be “understand the business better.” Understanding should lead somewhere.

A service page might need the visitor to request a quote. A case study might send them toward a related service. A pricing page may need them to book a consultation. A blog post should help them diagnose a problem and decide what to do next.

Once the action is clear, inspect the page backward. Does the headline confirm that the visitor is in the right place? Does the content answer the questions that prevent the action? Is there believable proof? Does the call to action match how ready the visitor is? Does the form request only the information the business actually needs?

A redesign that wins compliments but produces the same number of qualified inquiries failed. We don’t care whether AI or a human designed it. The standard is the same.

Test mobile where the layout actually breaks

“Responsive” often means someone dragged a desktop browser window narrower, watched the columns stack and moved on. That isn’t a mobile review.

Use actual phones and test more than one screen width. Open the menu, follow the navigation, tap the main buttons, submit the forms, rotate the device and increase the text size.

Don’t test only the cleanest page with the shortest headline. Open the pages with long service names, validation errors, alerts, real customer reviews and actual content.

Watch for the header covering the page title, forms disappearing behind the keyboard, buttons falling outside the screen, menus that won’t scroll, and sections stacking in an order that no longer makes sense. Check whether phone numbers can be tapped, whether popups can be closed, and whether animation makes the page feel slower than it looked in the design preview.

AI is good at producing the expected layout. Customers, browsers and real content have a habit of producing unexpected conditions. Test those.

Compare the old URLs with the new URLs

This is the part that can make a beautiful redesign expensive.

When an existing page disappears or changes address, the rebuild needs an explicit decision about where that old URL should go.

An AI agent can generate a redirect file. It can’t independently determine whether each old service page, article or location page should lead to a specific replacement, remain live, return a real not-found response, or be consolidated with another page. It may take the easiest route and send everything to the homepage.

Google warns against exactly that: “Don’t redirect many old URLs to one irrelevant single URL destination, such as the home page.”

The same documentation says that once you have the listing of old URLs, you decide where each one should redirect to, and that a page which has permanently moved should use a permanent server-side redirect such as a 301 or 308. We cover the full migration in Changing your business name? Somebody needs to own the redirects.

For this launch review, you don’t need to write the redirect rules yourself. You do need to see the map.

• 1. Every current URL
• 2. Its status after launch
• 3. Its new destination, when applicable
• 4. The reason for that decision

If nobody can produce that list, the migration wasn’t planned.

Make sure the staging settings don’t reach production

Development websites are often intentionally hidden from search engines while they’re being built. That’s correct. Launching without removing those restrictions isn’t.

Google’s migration guidance puts it plainly: “Don’t forget to remove any noindex or robots.txt blocks that were only needed for the migration.” Each new URL should carry a self-referencing canonical tag rather than one still pointing at the staging domain, and the new sitemap should be submitted in Search Console so Google learns about the new URLs.

Also confirm that password protection, crawler-blocking firewall rules and development-only environment settings are gone. Analytics, Tag Manager and Search Console verification should be connected to the production site rather than left behind on the previous build.

This isn’t glamorous work, which is exactly why it gets missed. The homepage can look perfect while the entire website asks Google not to index it.

Check the pages nobody wanted to design

The homepage gets reviewed by everyone. The thank-you page gets reviewed by nobody. That makes the unglamorous pages more dangerous than the homepage.

Trigger a form error. Visit a URL that doesn’t exist. Complete a booking and open the confirmation email. Use the site search when there are no results. Review the privacy policy, the cookie controls, the password-reset flow, the PDF downloads and the social-sharing preview.

These are the places where placeholder copy, stale company information, broken branding and unfinished workflows hide. They’re also the pages customers see immediately after something unexpected happens.

Confirm that the business can leave

AI makes it easier to reach launch day before anyone asks who controls the domain, the source-code repository and the analytics accounts.

That subject deserves more room than one section here. Use The accounts your business should own, not your agency to check account access, then read Do you actually own your website? for the separate questions covering the platform, the content, the code and the copyright.

The simple version: if your provider controls everything and you can’t move the site without their permission, you don’t own a business asset. You have a rental with unclear terms.

Ask for evidence, not reassurance

“Everything has been tested” isn’t a test report. “SEO is handled” isn’t a migration plan. “The form works” doesn’t prove that the lead reaches your sales process.

Before launch, ask the builder to show you:

• The old-to-new URL map
• A crawl of the finished website
• Successful form submissions and where each one landed
• Mobile test results
• Analytics and conversion events firing
• Search-indexing settings
• Performance results for the important page types
• Known limitations
• The accounts and assets transferred to the business
• The monitoring plan after launch

This isn’t needless bureaucracy. AI accelerated the production. Some of the time it saved should be spent verifying the result.

The same tension returns every month after launch, when the person reporting on the work is also the person doing it. Five things to check yourself before the next agency report lands covers what to verify for yourself instead of taking on trust.

AI isn’t the problem. Unreviewed work is.

We use AI in the websites, applications and internal tools we build. We’re not going to tell business owners that anything touched by AI is automatically cheap, unsafe or fake. That would be dishonest.

The problem is treating fast output as verified output.

A capable builder using AI should be able to produce more, test more and find more problems before launch. AI should raise the amount of scrutiny a project can afford. It shouldn’t become the reason scrutiny disappears.

The standard is simple. A website isn’t ready because it looks complete. It’s ready when the pages, claims, links, forms, tracking, migration, ownership and lead flow have been checked against what the business actually needs.

One caveat, since we’re the ones offering to look. We’ve told readers to be skeptical of the cold pitch in Are those free website audit emails worth reading?, and that skepticism is correct — a generic email claiming it found mysterious problems on your site is evidence of a script, not of a review. This is the opposite arrangement: you bring us the site, and we inspect the actual changes.