# Where AI helps in a website project > AI can help organise a brief, explore copy, build prototypes and check implementation. The useful starting point is a clear audience and a real business problem. Keep source facts separate from suggestions, review the working website and measure the result after launch. By Dom England. Published 2026-11-04. Canonical page: https://doms-lab.com/insights/using-ai-in-website-design A website project involves a lot of small decisions: what to say first, which pages belong together, how an interaction behaves and what happens when something fails. AI can help make those decisions tangible sooner. It can also produce a convincing page before the underlying offer has been thought through. I’d use AI throughout the process with a clear distinction between what the business knows and what we are exploring. A draft can be disposable. A claim about a client, a price or a service commitment needs a reliable source and a person responsible for approving it. ## Start with the business’s actual material Gather the current service descriptions, customer questions, brand assets and practical constraints. Mark what is confirmed, what is out of date and what still needs an answer. That gives the project a shared reference and makes gaps visible before they turn into polished copy. Ask AI to organise that material around the visitor’s questions. It can suggest page structures or point out inconsistencies, but the business should verify the facts. I’d keep a small content ledger for claims that matter: the claim, its source, its owner and the place it appears on the website. ## Explore options against a clear brief A useful creative prompt describes the audience, the decision they need to make and the character of the business. Include the information the page must communicate. Asking for several approaches to that same problem makes the alternatives easier to compare than asking for an impressive website in general. For example, a specialist service could be introduced through a customer problem, a process explanation or a short project story. Review which approach helps a new visitor understand the offer. Then develop the strongest direction into real content and layout, with enough detail to test on a small screen. ### A useful exploration prompt Using the supplied service notes, propose two page structures for a first-time customer. Show where each answers scope, evidence and the next step. Flag missing facts instead of filling them in. ## Prototype the uncertain part AI-assisted development can help turn a proposed interaction into something reviewable. Start with the part that is hardest to judge on paper: a product comparison, an enquiry journey or a scroll sequence explaining a complex service. Agree what the prototype should help you learn. The working version needs human review beyond whether it renders. Try keyboard navigation, narrow screens, long content, errors and reduced motion. Read the code and its dependencies. A prototype becomes a production candidate when its behaviour and limits have been checked for the actual use, rather than when the screenshot looks finished. ## Keep project evidence authentic Generated artwork can establish an atmosphere or explain an imagined concept. A case-study image has a different responsibility: it should help someone understand the work you actually did. Label a concept as a concept and use genuine product captures when showing a working interface. I’d apply the same rule to words. Avoid adding a customer quote, certification or performance figure just because a template has space for one. A concise explanation of the real contribution can be much stronger. It also makes the page easier to maintain when the product or the business changes. ## Build useful pages that can be discovered Google’s AI-search guidance says the established SEO foundations remain relevant, including crawlable content and clear technical structure. It also says there is no special schema required for its generative search features, and llms.txt is not a Google ranking shortcut. Structured data should accurately describe the visible page. AI can help prepare titles, summaries and a consistent content model, but each page still needs a purpose for a reader. Google’s spam policies address scaled content created mainly to manipulate rankings. A sensible editorial test is whether the page answers a real question with specific, useful information and sources for factual claims. Sources: [Google Search: optimising for generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide); [Google Search spam policies: scaled content abuse](https://developers.google.com/search/docs/essentials/spam-policies#scaled-content) ## Include the work after launch Before publishing, agree who owns the content, where enquiries go and how the website’s success will be measured. Keep a record of known limitations and give the business a way to request changes. An AI-assisted build still needs someone responsible for the finished experience. I’d start with one clear journey and improve it against evidence. Use AI to make more possibilities reviewable, then make deliberate choices about what earns a place in the site. The practical benefit is a shorter path from an uncertain idea to something the business can inspect, use and maintain. - Approve business facts and page copy before launch. - Check the live form and its real receiving destination. - Review metadata, canonical URLs and crawler settings. - Test the production build on a phone and with the keyboard. - Agree an owner and a first review point after launch. ## Sources & further reading - [Google Search: optimising for generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) - [Google Search spam policies: scaled content abuse](https://developers.google.com/search/docs/essentials/spam-policies#scaled-content) [Discuss your project](https://doms-lab.com/#contact)