The AI builds a website in ninety seconds. It does not build your website.

At what they do, yes. An AI website builder takes a few sentences about your business and returns a complete, coherent, reasonably attractive multi-page site in under two minutes. The layout is sensible, the fonts pair well, the colors do not clash, and the pages are in an order that makes sense. Ten years ago that was a week of work. Judged as a first draft, it is a genuinely good tool. The confusion starts when the first draft gets mistaken for a finished product. What the generator produces is a template arrangement filled with plausible copy: a hero line about your commitment to quality, a three-column services block with generic service names, an about paragraph that could belong to any business in your industry, and stock photography of people who do not work for you. Everything that makes a website win a customer instead of merely existing — your specific offer, your pricing posture, your service area, your proof, your photos, your forms wired to the place you actually read leads — is the part the AI cannot generate because it does not know it. The builder gets you to eighty percent of a website that looks done and zero percent of a website that is specifically yours.

Most AI builder plans that remove the platform's branding and connect your own domain land between $16 and $49 a month, or roughly $190 to $590 a year, and the number climbs once you add e-commerce, a booking app, extra storage, or the premium tier that unlocks the better templates. Add a domain at $15 to $25 a year and a paid stock-photo or app subscription and $300 to $800 a year is a realistic all-in figure for a working small business site. Over five years that is roughly $1,500 to $4,000, and none of it buys you anything you keep. A custom build is $1,200 and up as a one-time cost, and then either self-hosting at $6 to $18 a month or ArdinGate managed hosting at $60, $100, or $180 a month depending on tier. A $1,200 single-page custom site on Core hosting at $60 a month totals $4,800 over five years. So on pure cash the AI builder wins that comparison and it is not close. The honest framing is not that custom is cheaper — it usually is not at this size — but that you are buying two different things. One is a rental that stops existing when you stop paying. The other is an asset you own, on hosting you can change, that a different developer can pick up and keep working on.

Not for being AI-generated. Google's stated position is that it rewards helpful, original content regardless of how it was produced, and its spam policies target scaled content abuse — mass-producing pages primarily to manipulate rankings — rather than the use of a generative tool. A site whose copy came out of an AI builder is not penalized by rule. The practical risk is different and more mundane: the copy an AI builder generates about a plumbing company in Ohio is close to identical to the copy it generates about a plumbing company in Oregon, because both were produced from the same model, the same prompt shape, and the same template. It is not duplicate content in the technical sense, but it is undifferentiated content, and undifferentiated content has nothing to rank for that a hundred other pages are not also competing on. You do not get demoted. You just never surface, because there is no reason for you to. The fix is not avoiding AI. The fix is having something specific to say — your actual service area, your actual pricing posture, your actual process, the objection your customers actually raise on the phone — and putting that on the page instead of the generated version of it.

Almost never in a form you can use. Platform export features vary, but the pattern holds across all of them: you can usually get your content out — text, images, product lists, a customer or subscriber list — and you cannot get the site out. There is no button that hands you a folder of files you can upload to a different host and have your website appear. The design lives in the platform's rendering engine, the pages are assembled at request time from the platform's database, and the interactive parts are the platform's apps running on the platform's infrastructure. Framer and Webflow are partial exceptions with code-export options on higher tiers, but the exported result is generally a static snapshot rather than a maintainable codebase, and the moment you export it you have lost the editor that was the reason you chose the platform. Plan accordingly: treat an AI-built site as a rental with a good view, not as equity. Moving off it later means a rebuild, and the thing worth protecting on the way out is your domain, your content, your web addresses, and your analytics history — not the site itself.

The generation takes ninety seconds. Getting to a site you would actually put on a business card takes most people between eight and thirty hours spread over one to three weeks. That gap is the whole story, and every demo skips it. The generated draft arrives with placeholder services that are not your services, an about section describing a company that is not yours, stock photos of a team that does not exist, a contact form pointed at the email address you signed up with, business hours that are wrong, no service-area pages, and no proof of any kind. Replacing all of that is not hard work, but it is real work, and it is the work nobody budgets for because the demo made it look finished. Add the domain connection, the email deliverability setup so form submissions do not land in spam, the legal pages, the analytics install, and the mobile pass where the generated layout breaks on a phone, and a realistic first launch is a couple of weekends. A custom single-page build runs one to two weeks and a multi-page build three to six weeks — but those timelines include the copy, the structure, and the launch checklist rather than handing them to you as homework.

Overwhelmingly the second, with some real generation on top. Under the marketing language, the typical AI builder maps your answers to a template family, selects a color palette and a font pairing from a curated set, picks section blocks appropriate to your industry, orders them into a conventional page sequence, and generates copy to fill them. Every one of those steps is a constrained choice from a finite menu, which is exactly why the output is coherent — it cannot produce a broken layout, because it was never free to invent one. That constraint is a feature when the alternative is a blank canvas and it is a ceiling when you need to look like nobody else. Two businesses in the same industry who answer the setup questions similarly get sites that are recognizably siblings, because they are. Custom development starts from your positioning and builds the layout to serve it, which is slower, costs more, and is the only path to a site whose structure is an argument for your business rather than a generic frame around it.

Reasonably, and this is a genuine point in their favor. The major platforms handle TLS certificates, server patching, DDoS mitigation, and infrastructure security for you, and they do it at a scale and standard most small businesses would not achieve on their own. You will not get broken into because you forgot to update something, because there is nothing for you to update. Two caveats are worth naming. First, your exposure moves rather than disappearing: your site's availability, your customer data, and your uptime now depend on the platform's decisions, its outages, its billing system, and its account-security practices. A locked or suspended account takes your website offline just as effectively as a hack, and appealing it means a support queue rather than a phone call. Second, third-party apps installed from the platform's marketplace are a different risk category from the platform itself, and the review standard varies. A hand-coded site with no admin panel, no plugin marketplace, and no login page has a smaller attack surface than either option — but it does require that somebody keeps the server patched, which is exactly what managed hosting is for.

Yes, and for a genuinely new business it is often the sensible sequence. Launch fast, cheap, and reversibly; find out whether the business works; rebuild properly when the website starts carrying real revenue. Three things make that migration cheap later, and all three cost nothing to do now. Register the domain yourself, at a registrar you control, rather than letting the platform hold it — this is the single most important one, and it is the one most often gotten wrong. Keep your web addresses simple and stable, because every address you change later needs a permanent forward set up before the switch goes live or you throw away the ranking those pages earned. And keep your own copy of everything: your text in a document, your photos at full resolution, your customer list exported on a schedule. Do those three and moving off is a rebuild with a clean starting point. Skip them and it is a rebuild plus an excavation.

No, and the reason is structural rather than a matter of tuning. A hosted builder ships its own rendering engine, editor scaffolding, analytics, and app framework to every visitor on every page, because the page is assembled by that machinery rather than written out in advance. That baseline weight is there whether your page has ten words or a thousand, and you cannot remove it, because it is the product. In practice that means a generated site commonly takes two to four seconds to show its main headline or hero image on a mid-range phone on a cellular connection, and lands somewhere between a marginal pass and a fail on Google's page-experience checks without dedicated attention. A hand-coded page ships only what the page needs and typically renders its main content in under a second on the same phone. Speed is a confirmed ranking input and, more to the point, it is a conversion input: the visitor who leaves during the second and a half of blank screen never sees anything you wrote. On a brochure site the difference is a nuisance. On a site that is the primary source of leads it is money.

It goes away. Depending on the platform you get a grace period, then the site is unpublished and visitors see a placeholder or a dead address, and after a further window the content is deleted. Your domain may or may not survive that depending on whether you registered it yourself or let the platform do it during setup — which is exactly why registering it yourself matters so much. This is not a criticism of the platforms; it is what a subscription is, and it is disclosed. It is simply the difference between the two models that most buyers do not price in at the start. A custom site you own does not have this failure mode. If you stopped paying ArdinGate tomorrow, the files are yours, they run on any standard web server, and the site keeps serving as long as somebody pays a hosting bill somewhere. The content ages, but nothing switches off. Over a ten-year business life that distinction is usually more consequential than the monthly rate that drove the original decision.

When the site's job is to exist rather than to sell. Specifically: you are pre-revenue and testing whether the business is real; the website is a formality because your work genuinely arrives through referrals, a marketplace, or social; the site has a known short life, like an event page or a seasonal campaign; the total budget is a few hundred dollars and stretching to a custom build would come out of something that matters more; or you need to be live this week for a reason you did not choose. In all of those, paying $1,200 or more and waiting two weeks for a hand-coded site is the wrong allocation, and anyone who tells you otherwise is selling. The signal that the calculus has flipped is boringly consistent: the website starts producing leads, and you start noticing what it costs you when it does not. That is when speed, positioning, custom lead handling, and ownership stop being abstractions and start being the difference between a site that earns and a site that is merely present.