How much to charge for a website when the AI did the building

The published guides mostly agree on a shape. A small business website runs roughly one to eight thousand dollars from a freelancer and higher from a studio, with hourly rates somewhere between fifty and two hundred dollars depending on who is writing and where they are sitting. Those are the numbers you will find if you read the pages that rank for this question, and they have not moved much in years.

They are also numbers nobody sources. Every pricing guide in this genre states its ranges as fact, and not one of them cites a survey, a dataset, or a sample. Treat them as the going rate of an opinion, which is what they are.

The question underneath yours is different anyway. You are asking because the build got short. What used to take a week of evenings now takes an afternoon of prompting, and it feels dishonest to send the same invoice for it. So the number in the invoice suddenly feels arbitrary, and you want someone to tell you what is fair.

Here is the answer, and it is not a range. Your price is set by what the finished site is worth to the client and by what other people in your market charge. Your cost is set by your hours. AI moved the second number and left the first one exactly where it was. AI made the site cheaper to build. It did not make it cheaper to buy. Cutting your price by the hours you saved hands the entire gain to the client, permanently, in exchange for nothing.

What freelancers actually charge for a website

If you need a starting point rather than a philosophy, these are the web design pricing tiers the ranking guides publish. Each one is named, along with the date it carries, so you can weigh it yourself.

A one-page or landing site

Roughly $500 to $1,500. The Website Profit Course pricing guide, read in August 2026, puts a template-based one-to-three-page site at $500–$1,500. Self Made Web Designer cites $1,000–$5,000 as the common band for basic freelance sites.

A five- to ten-page business site

Roughly $2,000 to $8,000. Website Profit Course lists $2,000–$5,000 for a custom five-to-ten-page site. AgentSite's guide opens with "US freelancers charge $1,500 to $8,000 for a small business website in 2026 and agencies charge $8,000 to $15,000." This is the tier most solo work lands in, and it is the tier this article is really about.

Anything with a store, a booking flow, or a login

$5,000 and up, with no useful ceiling. Website Profit Course says $5,000–$10,000+. Prismic's guide says a freelance web developer "may charge $1,000 for a small website project and $10,000 or more for a larger" one. Once real functionality is involved, the number stops being about pages and starts being about risk, and the ranges stop meaning anything.

Website maintenance sits alongside all three, commonly quoted as a monthly fee of $100 to $500 for content updates, monitoring, and small changes.

Hourly rates, where people quote them at all, run from $50 to $200. Website Profit Course says $50–$150. Prismic says $100–$200 in the US and $10–$30 in India — which tells you the hourly rate is measuring geography more than skill.

Where these numbers come from — and where they do not

Four of the pages that rank for this question were read in full for this article. All four state dollar ranges with total confidence. Not one of them cites a source. Not a survey, not a marketplace dataset, not a sample of invoices.

Dates are the other thing worth checking before you trust a range. Wix Studio's pricing guide for designers is bylined 25 December 2024 and marked updated the following day — written before an agent could produce a first draft of a site. Prismic's guide carries "2026" in its title and a publication date of December 2022.

This page does not have better data, and neither does anyone else writing on the subject. There is no census of what freelancers invoice. What exists is a circle of guides quoting each other's ranges until they harden into common knowledge, and the honest thing to do is say so before quoting them anyway.

Your price and your cost are two different numbers

This sounds obvious written down, and almost every pricing conversation ignores it.

Your cost is your hours multiplied by what an hour of your life is worth, plus the share of your monthly overhead that this project has to carry. Self Made Web Designer is the only guide in the results that does this arithmetic — take your monthly business costs, divide by the projects you take in a month, and you have the floor beneath which a project loses you money. Their worked example: $5,000 a month across three projects means no project can be worth less than $1,667 to you.

Your price is what the site is worth to the person buying it, bounded by what they can find elsewhere. A restaurant owner is not buying eleven hours of your attention. They are buying a thing that makes the phone ring, and they are comparing you to the other people who could make it.

Those two numbers are connected by exactly one rule: price must exceed cost. Above that floor, they float independently. Which pricing model you pick decides how tightly you tie them together:

• An hourly rate ties them together completely. It is the one pricing model that breaks the moment the work gets faster, because it converts every efficiency gain into a pay cut and hands it to the client automatically. If you charge hourly and you have started using an agent, you have already given away the savings without deciding to.
• A fixed price or flat fee per project does not break. The client agreed to a number for an outcome. The outcome is unchanged. Your margin improved and that is your business, the same way it would be if you had simply gotten better at your job.
• Value-based pricing gets stronger, because it was never indexed to your effort in the first place.

The move most people make when the build gets short — quote fewer hours at the same rate — is the worst of the three. It keeps the pay cut and adds an argument about timesheets.

What AI actually took off the cost side

Be specific about this, because vague claims here are how people talk themselves into vague discounts.

What genuinely got shorter: the first draft of the markup and styles, the standard sections you have built two hundred times, the first pass at the copy, and the turnaround on "can we try it in green." Most web designers used to carry a personal template library precisely to shorten that work; an agent does what the template library did, without the part where every project starts by fighting a template into a shape it was not made for. The publishing step got shorter too — the part where a finished folder of files used to need an account somewhere, a build configuration, and an afternoon of reading. That gap was never really about difficulty, and it is largely gone.

Call it a real reduction in the hours it takes to build a website. It is the reason you are reading this. Where those hours actually went, stage by stage, is its own article; the short version is that the build shrank and the rest of the project did not.

And what it did not take off

Everything on this list still takes exactly as long as it did in 2023.

• The brief. Working out what the business actually sells, and to whom, before anyone writes a heading. A business owner who could answer that precisely would not need you.
• The parts of the job that were never the build. Copy that sells rather than describes, the SEO groundwork, image optimization, the accessibility pass. An agent helps with all of it and finishes none of it.
• Getting content out of the client. This is the one that eats calendars. Nothing about an agent shortens the four weeks it takes someone to send you their photos and their opening hours, and an unbounded revision cycle can turn a profitable quote into a loss on its own.
• The decisions. What goes above the fold, what gets cut, when to tell a client their idea will not work. An agent will happily build the wrong thing beautifully.
• Checking it on real devices. Not the preview. The actual page, on a phone, on the connection your client's customers have.
• The handoff. The domain, the access, who edits it in March.
• Responsibility. If it breaks, they call you. That has not been automated and is not close to being automated.

If you look at where a client project actually spends its calendar time, the build was rarely the largest block. It was just the block that felt like work.

The hours you saved are not free hours

Two things quietly eat the savings, and only one of them is obvious.

The obvious one: reviewing what the agent produced. Reading it, catching the confidently wrong bit, fixing the thing it invented. That is real time and it scales with how much you let it write unsupervised.

The less obvious one is that people are bad at measuring their own speedup. In July 2025, METR ran a randomized controlled trial on experienced open-source developers. Sixteen developers, 246 real issues on repositories they had contributed to for years. Before starting, they expected AI to speed them up by 24%. Afterwards, they believed it had sped them up by 20%. Measured, they took 19% longer.

Run the margin math on your own numbers

The pricing calculators in this genre ask for your experience level and hand you a range. That is a horoscope. This takes ten minutes with your last invoice and tells you something true about your own business.

1. Step 1Start with what you chargedTake the price of your most recently completed project. One real number, not an average.
2. Step 2Count every hour it really tookInclude the brief, the calls, the chasing, the revisions, the handoff. Not just the hours that felt like production.
3. Step 3DividePrice divided by hours is your actual rate on that job. It is usually well below the rate you tell people you charge.
4. Step 4Subtract your overhead shareTake your monthly business costs, divide by projects per month, and subtract that from the price before you divide again. This is your real margin.
5. Step 5Now ask the only question that mattersRe-run steps two and three using the hours that project would take you today. Then ask how many more of those projects you can actually sell this month.

Step five is where the exercise earns its keep. If a project used to take you 40 hours and now takes 16, your effective rate went up two and a half times — but only on the projects you actually book. The saved hours convert to money only if there is demand waiting to fill them. Otherwise you have bought yourself free afternoons, which are lovely, and which no client pays for.

Almost every argument for cutting your price assumes you will make it up in volume. Check whether that volume exists before you assume it. For most solo practices the binding constraint was never production capacity. It was finding the next client.

If you pass the savings on, you cannot take them back

A discount is not an event. It is a new number in a client's head, and in the heads of everyone they refer you to. Quote $1,200 for something you used to quote $4,000 for, and $4,000 is gone — not just from that project, but from that relationship and its referrals. Prices ratchet down easily and up with enormous friction. Ask anyone who has tried to raise their rates on an existing client how that conversation goes.

There is a collective version of this too. Market rate is not a law of nature; it is the average of what everyone in your market is willing to accept. If enough people index their prices to their build time, the band moves down for all of them, and the tools that made the work faster end up transferring the entire surplus to buyers. That is a defensible thing to care about even if you are only looking at your own quarter.

Where cutting the price is genuinely right: breaking into a new niche where you need two portfolio pieces more than you need the money. A deliberately cheap productized package — one page, fixed scope, one revision — used as the front door to a longer relationship and a way to meet new clients. A client whose budget is real and whose work you want. Those are decisions. Discounting because the software got faster is not a decision; it is a reflex, and the client did not even ask.

What the site still costs to keep alive

The build is a one-time number. The site is not. Whatever you decide about price, do not let these fall off the invoice.

The domain name has an annual cost that somebody keeps paying, and the site needs an address with HTTPS on it, so pointing the client's own domain at the site is a step someone has to own. Content updates after handoff are the item people forget most often; changing one line on a live site is quick, but "quick" and "free" are not the same word. Someone has to hold the access. And if you promised the site would rank, the SEO work does not stop on launch day either.

The usual advice is to bundle website maintenance into a single monthly fee rather than itemizing, and the usual advice is right, because one number is easier to sell than six.

One honest limit while you are scoping: a static site does not, by itself, receive a form submission or process an order. That needs a separate service sitting behind the page, and it is worth knowing which side of that line the project falls on before you quote it. It is also worth deciding early whether the client wants a site or an application. The economics in this article only hold for the first one.

Where Birta fits

Birta is the publishing layer after the agent. The finished page goes live from the same conversation that produced it, at a real HTTPS address, and stays editable from there: change one file, publish again, roll back to a previous version if the change was wrong. Images and fonts travel with the site. You can point the client's own domain at it, see the project's traffic, and connect your agent through the browser rather than by hand.

Where it stops, plainly: it publishes sites, not applications. Video and PDF files are not served, so those still need somewhere else to live. A connected agent can declare a contact form and get its submissions in the project's panel, so that line item is no longer separate work to price. Taking payments is still not something Birta does; that stays the client's problem to solve with a separate service, and you should price it accordingly rather than assume the host covers it.

None of this changes what you should charge. It changes how many of the hours in step two of the worksheet are still there.

Frequently asked questions