I built a website with AI. What should I do next?

Why does an AI-generated website still need work?

A page loading successfully does not prove that the entire product works. A form may display “Message sent” without saving or delivering anything. Products may come from sample data, and an administration link may point to a mock screen.

Whether you started with ChatGPT, Claude, Lovable, Bolt or v0, ask the same question: can visitors complete the tasks they came for? Assess the running website and its code together. A screenshot or an HTML export cannot establish whether server-side features work.

• Write down the three most important user tasks, such as requesting a quote, browsing a product or signing up.
• Try each task as a new visitor and confirm that the expected record or message actually exists.
• Record problems with the page address, device and steps to reproduce them.

How do you fix an AI website that breaks on mobile?

Fixed-width cards, oversized headings and viewport-height spacing can look acceptable on a desktop while failing on a phone. Review the layout, content priorities and how components behave at narrow widths together.

Define consistent typography, colours, spacing and button styles so pages generated in different sessions feel like one product. Navigation and essential actions should work with touch and a keyboard, without depending on hover.

• Check headings and cards at phone widths of 320–430 px, on tablets and on wide screens.
• Fix horizontal overflow, clipped text, headings underneath the navigation and excessive top spacing.
• Review form labels, error messages, colour contrast and visible keyboard focus.

What does refactoring AI-generated code involve?

Refactoring improves the structure of code while preserving its expected behaviour. Examples include replacing five copies of a button with a shared component or separating data operations from interface code. The aim is to make future changes reliable.

Document the current flows and add tests for critical behaviour first. Then address unnecessary dependencies, unused files, duplicated styles and temporary workarounds in small steps. Check the build and key journeys after each change. Working code does not need a rewrite simply because AI produced it.

• Confirm that the project builds reproducibly from a clean installation.
• Separate sample data from real data sources and development settings from production settings.
• Document setup, configuration and deployment so another developer can follow them.

Are AI-generated websites secure?

Being generated with AI does not, by itself, make a site secure or insecure. Review authentication, permissions, data handling and dependencies. HTTPS matters, but it does not stop one customer accessing another customer’s records.

Private API keys must not be shipped in browser code. Hiding a button is not authorization: sensitive operations need permission checks on the server. Validate input on the server and review upload restrictions and request limits. Combine automated scanning with human review and workflow testing.

• Use test accounts with different roles to verify that each can access only permitted records.
• Revoke and replace exposed secrets; deleting them from the current source code is not enough.
• Review vulnerable dependencies and session handling without using real customer data for testing.

OWASP: Secure Coding with AI ↗

Why is my AI-built website slow?

Large images, autoplay videos, animations and unnecessary JavaScript can increase the initial loading cost. Measure what causes the delay, then adjust image sizes, loading priorities, fonts and server response. The main image and an image at the bottom of the page do not necessarily need to load together.

The good Core Web Vitals thresholds are LCP of at most 2.5 seconds, INP of at most 200 milliseconds and CLS of at most 0.1, assessed at the 75th percentile of visits. Use PageSpeed Insights to investigate issues and examine real-user data when available. A high laboratory score alone does not guarantee speed on every device or a top search position.

• Test forms and content pages as well as the homepage on a slow mobile connection.
• Reserve space for images to reduce layout shifts during loading.
• Compare measurements before and after changes under equivalent conditions.

web.dev: Web Vitals ↗

Can an AI-generated website rank on Google?

Yes, an AI-built website can appear in Google Search. Google’s guidance on AI content focuses on quality and usefulness rather than the tool used. That is not a promise of indexing or rankings. Write accurate service descriptions that answer visitors’ questions; avoid producing near-identical location and service pages.

If your site is missing, start with URL Inspection in Search Console. Look for an accidental noindex directive, crawl restrictions, an incorrect canonical URL, missing content or access errors. Review page titles, descriptions, internal links and the sitemap. JavaScript is not automatically a problem; verify that search engines can access the important content.

• Keep staging and production indexing settings separate.
• Match canonical and hreflang URLs to the actual language versions.
• Use structured data only for accurate content visible on the page.

Google Search Central: AI-generated content guidance ↗

How do you prepare the website for production?

Moving a local website to a server is only part of launch preparation. Verify the domain, HTTPS, environment variables, database, email delivery and any payment integration in the production environment. Check the platform’s code export, hosting and data migration terms before planning a move.

A missing page should return an appropriate 404 or 410 response. Showing “Page not found” with a successful 200 response can result in a soft 404. Redirect important old addresses to relevant replacements, rather than sending every invalid address to the homepage.

Prepare a rollback plan alongside the launch plan. Test that backups can be restored, monitor error logs and decide how failed forms or payments will be detected. Assign responsibility for maintenance and security updates after release.

• Test both successful and failed critical journeys in staging.
• Identify where personal data is stored, who can access it and how unnecessary records are removed.
• Include post-launch checks, backups and rollback instructions in the handover.

Google: HTTP status codes and crawling ↗

Should you improve or rebuild the site, and what affects cost?

Make that decision after an assessment. If the design and structure are usable, improving the existing project can make sense. If the data model, permissions or infrastructure cannot support the required workflows, rebuilding particular modules may be more sustainable.

Page count alone does not determine cost. Membership, payments, administration screens, data migration, integrations, existing defects and testing requirements affect the work. An initial review should establish findings, priorities and acceptance criteria before a fixed price or timeline is proposed.

A practical priority order might be: fix access to another customer’s records, repair a form that loses messages, then address mobile overflow and speed. The handover should include changes made, measurement results and maintenance notes.

• Bring the site URL, a problem list and the user journeys you need.
• Prepare information about the source code and platform; do not send passwords or private keys in an initial message.
• Clarify exclusions, deployment responsibilities and maintenance arrangements in the proposal.

Frequently asked questions