How to Build a Lead-Generating Website With AI

A lead-generating website is an ordinary website plus one thing done on purpose: an unbroken path from any page a stranger lands on to a message sitting in an inbox you actually read.

Most business sites have that path, technically. It runs through a contact page nobody visits, a form with eleven fields, and a notification email that goes to an address the owner stopped checking in 2023. The path exists and nothing travels down it.

This is a build guide for the path itself. It assumes you can already get a site online, either by hand or with AI, and it is about the specific parts that turn a visit into something you can reply to.

What counts as a lead#

A lead is a message from an identifiable person that you can answer. That is the whole definition, and it is stricter than it sounds.

A thousand visits is not a lead. A page view on your contact page is not a lead. A form submission you never received is not a lead, and that last one is far more common than anybody expects, which is why a whole section below is about the email.

So there are two numbers, and you want both before you change anything: how many messages arrived last month, and how many of them got a reply. If you cannot answer the second one, fix that before you build anything new. A site that generates leads into a void is an expensive way to disappoint people.

Step 1. Pick one action per page#

Before any building, decide what a visitor is supposed to do. Not three things. One, per page.

A page offering a free quote, a newsletter signup, a phone number, a brochure download and a chat bubble has not given the reader five options. It has given them a decision they did not come to make, and the common response to an unexpected decision is to leave.

The action can differ by page, and it should. A service page asks for a quote. A pricing page asks them to start. An article asks for the newsletter, because someone reading about a problem is usually not ready to buy a solution. Matching the ask to where the reader is in their thinking is most of what "conversion optimisation" means once you take the jargon off.

This is the first place AI genuinely saves you time. Writing one call to action is easy. Writing thirty, each phrased for the page it sits on and none of them identical, is the kind of task that gets half-done and then abandoned. Hand it the page and it drafts the ask; you decide whether the ask is the right one.

Step 2. Build the pages leads actually come from#

Leads do not arrive evenly across a website. They come from a small number of pages where someone with a problem meets a specific description of the solution.

In practice that means one page per service, and one page per area you serve if location matters to your trade. Not one page listing everything. Someone searching for emergency drain clearing in a specific town needs to land on a page about emergency drain clearing in that town, not on a homepage that mentions drains in a list of nine services.

Blog articles are a different job. They rarely convert on the spot, and judging them by direct enquiries is how good articles get deleted. They earn attention and they answer the questions people ask before they are ready to buy. Give them the newsletter ask, not the quote form.

The reason this step used to get skipped is that it is a lot of writing for pages that each get modest traffic. Fifteen service and area pages is a fortnight of work you keep postponing. That is precisely the constraint AI removes, and the step-by-step build guide covers the mechanics of getting a site to that shape.

Step 3. The form#

Every field is a small reason to leave. So the question for each one is not "would this be useful to know" but "would I rather have this answer than have this person at all".

For most businesses the honest minimum is three things: a name, one way to reach them, and one open box for what they need. Budget dropdowns, how-did-you-hear-about-us, preferred contact time and company size are all questions you can ask in the reply, once you are talking to a human being who has already decided to talk to you.

Then three details that cost nothing and get skipped constantly.

• Test it on a real phone. Not a resized browser window. A form that looks fine on a desktop can put the submit button under a sticky footer on a specific phone, and you will never see it because you never fill in your own form.
• Say what happens next. "We reply within one working day" removes a real hesitation. Silence about the timeline is the reason people fill in three competitors' forms instead of one.
• Confirm the submission properly. A page that goes blank, or scrolls back to the top with no message, reads as broken. Some of those people submit again, and some of them leave believing it failed.

AI writes the labels, the error messages and the confirmation text quickly and in a consistent voice, which matters more than it sounds when the alternative is one field labelled "Enquiry" and another labelled "Tell us a bit about your project." What it should not do is decide which fields exist. That is a judgement about your business, and the cost of an extra field is paid by you, in leads that never arrive.

Step 4. The email that has to arrive#

This is the step that quietly ruins otherwise good websites, and almost nobody checks it.

Your form sends a notification. That notification is an email, from your domain, and mail providers decide whether to deliver it based on three DNS records: SPF, which lists who may send as your domain, DKIM, which signs the message so it can be verified, and DMARC, which tells providers what to do when the first two disagree.

If your domain has none of them, your own lead notifications are markedly more likely to be filtered, and so is the reply you then send to the customer. That second one is the expensive half: you answered, they never saw it, and to them you are the business that ignored them.

The rules tightened recently. Google and Yahoo require SPF and DKIM from senders generally, and a DMARC policy from bulk senders, where bulk means 5,000 or more messages a day to personal inboxes. Since November 2025 non-compliant bulk mail is rejected outright rather than sent to spam. A small business is nowhere near 5,000 a day, so the bulk rules are not aimed at you, but the same records are what filtering decisions are made from at any volume.

Step 5. Consent, if you plan to market to them later#

Answering someone who wrote to you needs no special permission. Adding them to a mailing list is a different act, and in the EU and UK it needs consent that is freely given, specific, informed and unambiguous.

The practical consequence is narrow and worth knowing: a pre-ticked box is not consent. The Court of Justice of the European Union settled that in the Planet49 ruling in 2019, and it is the single most common way a small business form gets this wrong. Bundling it in is the second, so the newsletter opt-in has to be its own unticked checkbox, not a line of small print under the submit button.

Keep the enquiry and the list separate and both stay clean. The person who wanted a quote gets a quote, and the person who wanted the newsletter asked for it.

Where AI does the work, and where it cannot#

Split by whether the task needs judgement about your business or just needs doing, repeatedly, without getting bored.

The bottom row is the one that decides whether any of this was worth doing. Everything above it produces a message in an inbox. Only a person turns that message into a customer, and no amount of tooling has moved that line.

How this usually fails#

In rough order of how often it happens, from what breaks first:

1. The notification never arrives. Filtered, or sent to an address nobody opens. The site looks like it produces nothing.
2. It arrives and nobody answers. The lead books whoever replied first. This is an operations problem, not a website problem, and building more pages does not touch it.
3. The form asks too much. People start it and stop. You never learn they were there, which is what makes this one hard to notice.
4. There is no page for what people search. The path is fine, but nobody reaches the start of it.
5. The page is fast and pretty and says nothing specific. The visitor cannot tell you apart from the other tab, so they use price to decide.

Notice that the first two have nothing to do with how the site was built. If you are choosing between spending a week on new pages and spending ten minutes testing your own form, test the form.

A test you can run today#

Half an hour, no tools, and it tells you which of the five failures above is yours.

1. Fill in your own form from a phone. Use a real address you can check. Watch what the page does after you press submit.
2. Find the notification. Inbox first, then spam. If it is in spam, or absent, stop here and fix the DNS records before anything else.
3. Time the reply. Not your intended reply time. The one that actually happens when nobody knows they are being watched.
4. Count the fields. Cross out every one you could ask in the reply instead. Most forms lose four or five.
5. List what customers actually phone to ask. Check a page exists for each. The gaps are your next pages, and they are worth more than a redesign.

Steps one to three cost nothing and need no AI at all. That is deliberate. Tooling is worth adding once you know which part is broken, and it is mostly wasted before that.

If you want the wider view, of how the pieces fit together and which number to watch at each stage rather than how to build them, we set that out separately. And if you are still weighing whether an AI-built site is good enough to put your name on, we looked at that honestly here.