You Can Build Your Own Website With AI. Read the Fine Print First.
AI app builders like Lovable make it possible for a small business owner to build their own site in an afternoon. Here's what you take on when you do, and when it's still the right call.

If you run a small business, you've probably seen the ads by now. Describe the site you want, and a tool like Lovable builds it while you watch. A couple of years ago that would have sounded like a gimmick. Today it mostly works.
I build with AI every day, and it's a big part of how one developer can cover work that used to take a small team. So I'm not going to tell you these tools are a trap. You can build a lot on your own now. What I want to cover is what you're signing up for when you do, because some of it isn't obvious until something goes wrong.
When doing it yourself is the right call
I'll start with the part that doesn't help me sell anything.
If you need a simple site that says who you are, what you do, and how to reach you, an AI builder is a reasonable choice. Same for an event landing page, or a mockup you want to show a developer later so you can point at it and say "like this, but real." As long as the site has no logins or payments, and the most sensitive thing anyone types into it is a phone number, most of what follows matters a lot less.
The picture changes once the site starts holding customer data, taking payments, or running part of how your business works.
Lovable's terms put security on you
Lovable is upfront about this, which I respect. Its security documentation says, "You are responsible for ensuring that your app meets the security requirements appropriate for its use case." Lovable does run automated security scans before you publish, and the same page says those scans can't guarantee complete security.
The terms of service go further. You're solely responsible for reviewing and validating what the AI produces, and its output shouldn't be relied on without independent review and testing.
Almost every AI coding tool has language like this, and I think it's a fair position for a tool to take. The tool writes the code, but it has no idea what your data is worth or what your customers would think if it leaked. If you build it yourself, though, the "independent review and testing" is your job. Asking the same AI that wrote the code whether the code is secure doesn't count for much.
This has already gone wrong at scale. In 2025 a researcher scanned 1,645 apps from Lovable's public showcase and found about 170 that let anyone read, and in some cases change, data in the app's database without logging in. The database permissions that should have locked that data down were missing or broken. It was filed as CVE-2025-48757, and Lovable disputed it on the grounds that each customer is responsible for protecting their own app's data. So the owners of those apps were on the hook, and I'd guess most of them had no idea.
I wrote more about how this kind of problem hides behind a working demo in The Demo Always Works. That's the Problem.
The AI only knows what you tell it
An AI builder does a good job with what you ask for. The trouble is what you don't know to ask for.
Most owners know to ask about SEO by now. Fewer know that the emails your contact form sends can land in spam, or never arrive at all, unless your domain is set up to vouch for them. Nothing on the site looks broken when that happens. The leads just stop, and you assume it's a slow month.
Payments have more of these. Payment providers often send the same "this invoice was paid" message more than once, and software that doesn't expect the repeat can record a payment twice or email the customer two receipts. When I added online card payments through Square for one client, handling those repeats was written into the plan before any code, along with checking that each message really came from Square and not from someone pretending to be it.
Then there's the question nobody asks until it's too late: if the database gets wiped tomorrow, how far back can you restore it, and has anyone ever tried?
You can ask an AI about any of these, and it will give you a decent answer. You have to know the question exists first. Twenty years of building software is mostly a long list of those questions, and when I build with AI, they go into the conversation from the start.
Every fix can break something else
The first version usually goes well. The trouble starts around the tenth change. You ask the AI to fix how invoices print, and it fixes them, and somewhere along the way it also changes something on the quote page that you won't open again until a customer calls about it.
I've run into this myself. It's the most common way AI-written code goes wrong for me, and it's why every project I build has a thorough set of automated tests: small programs that check the important parts still work every time anything changes. Many of them are end-to-end tests, which click through the site the way a customer would, requesting a quote or opening an invoice. Once something works, the tests keep it working as the project grows, and a change that breaks it doesn't go live.
If you're building on your own, the test is usually you, clicking around after each change and hoping you looked at the right page.
Looking at how the business runs
When Chad at Nelson Drone Solutions came to me, he wanted a website where customers could request a quote. That's what I built first. But once we talked about how a drone spraying business actually gets from a phone call to a paid invoice, a quote form was clearly only one piece of it.
Here's what the site does now. Customers can outline their own field on a satellite map when they request a quote, and the acreage and the quote come out of that outline using Chad's own rates. The same field boundary carries through to the work order and the invoice, and it can be exported as a file for the drone or saved to the customer's own Google Drive. Service agreements get signed online, customers can pay their invoices by card, and every quote, work order, and invoice is in one dashboard instead of an email thread and a spreadsheet.
He didn't ask for any of that up front, and he wouldn't have known to type it into a prompt. He was thinking about a website. I was thinking about his week.
Someone who already knows your system
When something breaks on a site you built yourself, you're the one who has to notice, work out why, and fix it, usually at the end of a day you already spent running your business. A form stops sending email without telling anyone. A browser update makes a page look wrong on half your customers' phones. Some owners enjoy that kind of thing. Most of the ones I talk to would rather call someone.
The bigger difference is what happens when nothing is broken. Chad's site went live in April. Since then it has picked up the online service agreements, field coordinates printed on the paperwork, and customer details that fill in on their own when he starts a new form. None of that was in the original plan. It came from him using the system every day and telling me what slowed him down, and from me already knowing the code well enough to add it without starting over.
What it costs
A monthly AI builder subscription is cheaper than hiring me. That's true, and for the simple sites I described earlier, it may be the better deal.
The math changes when the site holds anything you'd be in trouble for losing, or when it could be doing more of your actual work. A data leak costs something, and so do leads lost to a spam folder and evenings spent chasing a bug. Those costs just don't come with an invoice.
If you're weighing it
If you've already built something with an AI tool and aren't sure it's safe to put customer data into it, or you're about to start and want to know which parts you can handle yourself, I'm happy to take a look. Sometimes the honest answer is that what you've got is fine.
— Nathan
Questions people ask
Can I build my own business website with an AI tool?
For a simple site that says who you are, what you do and how to reach you, an AI builder is a reasonable choice. The picture changes once the site holds customer data, takes payments, or runs part of how your business works.
Who is responsible for security on a site built with an AI builder?
You are. Lovable's security documentation and terms of service say the person building the app is responsible for its security and for reviewing and testing what the AI produces. Almost every AI coding tool has language like this.
What goes wrong with AI-built sites over time?
The first version usually goes well. The trouble starts around the tenth change, when a fix in one place quietly changes something in another. Automated tests catch that. Without them, the test is you, clicking around after each change.
Related services
Written by Nathan Wallman. How I use AI on this blog.