AI and websites

AI can draft the website in an evening. Choosing who builds it is still your job

The first version arrives within an hour and costs almost nothing, so the three routes to a website look identical at first. They are not: each sells something different and stops answering on a different date.

The first version is free now, and that changed the question

A tool that writes pages in an evening has not decided who turns them into a website. Until recently, the first question a small business asked was whether it could afford a site at all. Now a credible set of pages can arrive the same day somebody thinks of it, for the price of a subscription, and the question moves with it. Producing pages was never the hard part. Deciding what the pages should say, what has to work before launch, and who fixes it when a form stops sending in month four — that is the work, and no draft has done it.

So there are three routes to a website: building it yourself with AI, hiring a freelancer, working with an agency. For the first week they now cost roughly the same. After that they cost nothing like each other, and choosing between them is the real decision.

What the three routes are actually selling

Doing it yourself with AI is selling your own evenings, and they are easy to underestimate. The tool will generate a homepage, a services page and a contact form before lunch, and the result will look like a website. What it cannot supply is the second opinion: whether the navigation makes sense to a customer, whether the form actually sends mail, whether the offer reads the way you say it out loud on the phone. You become the quality control and the client at the same time. DIY sites rarely fail at the generation step; they stall in month two, when the person maintaining them runs out of free evenings.

A freelancer with AI is selling craft at speed, from one pair of hands. A good freelancer with today’s tools is faster than a small team was five years ago, and usually cheaper than an agency, because the drafting work they used to do by hand now happens in the background. The trade is concentration: one person holds the whole project, which is excellent while that person is available and a single point of failure when they are not. The question to ask is not whether they are good. It is what happens when you need them in August.

An agency is selling a process and somebody who is answerable. AI made agencies faster too, so part of the higher price is no longer drafting — it is having more than one person look at the site. Someone designs, someone checks, someone argues with you about the homepage, and when something breaks there is a rota rather than a favour. If your website touches money, bookings or customer data, that redundancy is the product, not the overhead.

What the quote does not say

Two quotes can look identical while buying completely different things. A fixed price and an open-ended estimate are different contracts, and so are “we build it” and “we build it and enter all your content”. Ask each candidate to list, one line per item, what is included: entering your real texts and photographs, testing the forms on a real phone, one round of changes after the first launch, and what a change costs in month five. Also ask who exactly does the work — the person who sold you the project or someone you have never met.

The cheapest line item is usually the one nobody has defined yet. Maintenance, changes and “support” appear in quotes as a shrug. Neither AI nor a handshake answers what happens when the price on your site is wrong six months from now; only a written answer does, and a candidate who refuses to write it down is telling you something.

Use AI to compare, not to decide

The most useful thing AI can do here is make three offers comparable. Paste in the three quotes, the list of what your site must actually do, and ask what each offer leaves out. Ask it to draft the same three awkward questions for each candidate — about availability, about who holds the accounts, about what happens after launch — so you are comparing answers instead of impressions. It is also the cheapest way to sanity-check a claim: “they said two weeks; what usually makes a job like this take four.”

It cannot read people. Whether the freelancer understands your trade, whether the person who sold the project is the person who will build it, whether the agency will put a named human on your account — that is judged in conversation. The answers that matter are the ones about limits, because every candidate has them and only some will say so.

How to choose without a spreadsheet

Let the site itself decide most of it. Five to eight pages, no payments or bookings, and somebody with two free evenings a week for six months: building it yourself is an honest choice, provided you accept what you will not be good at. One craftsperson whose previous work you have seen and a scope that fits on one page: a freelancer will serve you well. Decisions multiplying — payments, booking, a CRM, several people publishing content — and the agency’s process starts paying for itself.

Then the rule that beats all three: pick whoever will still answer in twelve months. The cheapest quote from someone who stops replying in July is the most expensive option on this list, because the alternative to a working website is not a cheaper website — it is no website, or a rebuild. Continuity is worth more than the discount, in every route.

The test to run before you sign

Give the same three questions to everyone: who exactly does the work, what is not included, and what happens the first time something breaks. Compare the specifics rather than the promises — names, weeks, written answers, and what each candidate refuses to promise. The one who answers plainly, including about their limits, is the one you can build with.

The draft is the cheap part now. The website — the decisions behind it, the checking, the twelve months after launch — is what you are actually buying, and it deserves to be bought from somebody who will be there when you call.


Related service: Website and web application development