Can I build a website?
Everyone’s talking about generative AI lately, especially the way it’s reshaping web design in real time. The idea is simple enough: describe what you want in plain language, and a chatbot turns it into a website. If that works well enough, business owners can skip the usual web design process and the platforms that come with it. Or so the story goes.
Can AI Build a website?
a.
Squarespace
b.
Video
So can AI really build a website? Here’s my short answer: yes, it can; and faster than you might expect. All three tools I tested turned the same brief into a working site. Whether that website is one you’d trust to run your business on for years, though, is another story entirely, and that’s where the real differences become apparent.
Quick verdict: Yes, AI can build a website! What it can’t do on its own is guarantee security, maintainability, or quality over the long haul. If you’re confident with the technical side and have expert eyes and skills to review the output, AI-led tools like Lovable and Claude will serve you well. For everyone else, my recommendation stands: build on an AI-assisted platform like Squarespace, where AI lends a hand while the platform carries the weight of security, updates, and scaling.
Let's put that theory to the test. For this small experiment, I gave each model the same prompt to level the playing field, providing as much textual and graphic context as I possibly could to increase the specificity of the outputs. Then I compared them in terms of customization, cost, security, SEO, accessibility, and scaling. By the end, you’ll know when AI is worth using, and why, so you can create a website that truly serves your business and gives you peace of mind.
This post was sponsored by Squarespace. Squarespace is the first platform where I learned to design websites, and it remains one of my most recommended platforms. I’ve also used and recommended others, and deployed custom websites when a client needed one. Every opinion here is my own.
Two kinds of AI website builders
Today’s AI website builders can be categorized into two:
AI-led code generators take a plain-language prompt, then generate the code and decide the tooling themselves. Lovable, Bolt, Replit, Base44, and Claude Code are popular examples. You end up owning a codebase, along with the hosting, maintenance, and security that come with it.
AI-assisted website platforms are established builders that have added AI features to help with parts of design and development. Squarespace, Shopify, and Webflow belong here. The platform remains the foundation, and AI helps you work within it.
For this test, I used Lovable and Claude as the AI-led examples and Squarespace as the AI-assisted example.
The brief: a fashion blog called Cherie
To test these tools the way a thoughtful business owner would, I wrote one detailed prompt and used it everywhere. Cherie is a fashion blog by Margaux, and it earns mainly through affiliate commissions on the products it recommends. The prompt names the brand and founder, lists the pages needed, and describes the look and feel, including color and font preferences.. It also makes one priority clear: because the blog needs constant updating, adding and editing posts has to be easy. Finally, it asks for a site that works on mobile, meets accessibility standards, and is set up well for SEO and discovery.
Test 1: Building Cherie with Lovable
1.2 Cherie homepage generated by Lovable.
After signing up at Lovable.dev, you can paste a prompt right away. Two settings are worth knowing before you do. First, Lovable offers plan mode and build mode; for a first ask, plan mode is the better choice because it asks clarifying questions before it starts building. Second, the attachment option in the lower left lets you add brand guidelines or visual references. I used plain text for this test, with color and font preferences written into the prompt.
Lovable’s clarifying questions covered how we’d write and publish articles, where to store email signups, and how to handle product recommendations. Since Cherie needs to be independent from Substack, I chose a built-in admin dashboard, storing signups on the site itself, with simple links and text for products.
Then came the plan: the pages to be built, an admin dashboard for publishing and editing posts, and notes on the technical stack. One note mattered most. The build would require Lovable Cloud, infrastructure Lovable sets up automatically (in partnership with software like Supabase) to handle the database and authentication. A static site with no ongoing edits might not need it. A blog that’s updated constantly does, because that’s what gives it a content management system. Details like these can feel intimidating, but I recommend reading through them and understanding what they mean for your site. I also confirmed the plan carried over our requests for responsiveness, accessibility, and SEO before clicking approve.
The homepage came out closely tied to the prompt, down to the hand-drawn fashion illustrations it generated throughout. One practical note: partway through generating, Lovable told me I’d run out of credits, and I had to upgrade to Lovable Pro to continue.
The admin panel deserves careful reading. Whoever visits the dashboard link first and signs up claims owner access, so do this yourself before sharing the link with anyone. Once signed in, the dashboard lets you write new posts and edit existing ones: title, body text, and a cover image file.
For everything else, you ask the chat. You can also annotate on the canvas, change text inline, or select a single element and ask for a revision to that element only. When I selected an image and asked for a more realistic version, Lovable changed only that image. When I asked for a folder dropdown in the navigation, it built the dropdown and flagged several security issues in the same breath. Those flags are worth addressing before you publish.
Publishing is a single button. Lovable gives you a built-in domain immediately, and you can connect a domain you already own or buy one through Lovable.
Lovable at a glance:
Plan mode is the better choice for a first ask—it asks clarifying questions before building anything.
One detailed prompt produced a coherent homepage, hand-drawn illustrations, a set of pages, and an admin dashboard within minutes.
Expect to hit a credit ceiling: I had to upgrade to Lovable Pro partway through generating.
Claim the admin dashboard before sharing the link—whoever signs up first gets owner access.
New features can surface security flags in the same breath; address them before you publish.
Test 2: Building Cherie with Claude Design and Claude Code
1.4 Cherie homepage generated in Claude’s design mode.
Claude.ai offers different modes: Chat, Cowork, Code, and Design. For building websites, I find it most efficient to start in design mode (through the design tab, or directly at claude.ai/design), especially if you already have a design system to set up first. For this test I pasted the same prompt, using the Opus 5 model, Claude’s most capable model at the time of recording.
Claude asked clarifying questions, the first being how we’d add blog posts: edit a simple post file, or connect a proper CMS (separate software requiring its own signup). I chose the simpler option, knowing we could switch later. Tech-savvy users can also point Claude at an existing GitHub codebase at this stage.
The generated site had working navigation between pages. Claude doesn’t generate images; it asked me to drop in my own or browse local files. An edit button lets you select any element and change its text, colors, and some layout properties, though you can’t drag elements or sections around. As with Lovable, you can select an element and ask for a change, and the folder dropdown request worked here too.
Managing blog posts is where the two tools diverge. When I asked how to edit posts, Claude pointed me to a posts.js file, the single file we’d chosen to edit. Workable, but not friendly for a founder publishing every week and quite intimidating for non-developers. To set up a proper CMS, we have to move from Design mode into Code mode.
Publishing also takes more steps. The share button offers a downloadable HTML file to host anywhere, or publishing as a Claude artifact limited to the Claude domain. Under “More formats and apps” there’s an option to send the project to Claude Code, which is where the flexibility lives. I started a web session, which carried the context over from design mode, and Claude Code asked its own round of questions. Its recommendation, to keep things simple, was still the single file, so we chose Next.js with a proper CMS. It flagged that this setup is heavier to host and maintain. For the signup form, it recommended embedding a Substack form or using Formspree. For hosting, it pointed to Netlify or Vercel.
About fifteen minutes later, Claude Code returned a long, developer-facing explanation of publishing with a CMS, and pointed to a README for the rest: sign up for Sanity.io, retrieve tokens, sign up for Netlify or Vercel, connect and deploy. For anyone without a development background, this is where the project turns intimidating.
1.5 The README instructions generated by Claude Code.
To be fair, a single-page site with no blog and no ongoing edits wouldn’t need nearly this much setup. But even a simple Claude-built site still needs hosting if you manage yourself. If you’re comfortable with that, the workflow I’d recommend looks like this: connect GitHub through Claude’s connectors, create a private repository, prompt Claude to push the code there, connect the repository to a project in Netlify or Vercel, then attach a custom domain from the hosting settings.
Claude at a glance:
Design mode is the most efficient starting point, and Claude asks its own round of clarifying questions.
The generated site had working navigation, but no images—you supply your own.
Editing a selected element is easy enough; managing blog posts through a posts.js file is workable but intimidating for non-developers.
A proper CMS means moving into Code mode and wiring up Sanity, tokens, and Netlify or Vercel—the point where non-technical founders may feel out of their depth.
Test 3: Building Cherie with Squarespace
1.6 Cherie homepage generated through Squarespace’s Blueprint AI.
Squarespace’s onboarding is now AI-powered, though there’s no box to paste a prompt into. You answer questions one at a time based on your brand needs. For Cherie, I selected fashion and style as the category, and chose publishing a blog plus a contact form as the functionality.
After signing up, you land in Blueprint AI, Squarespace’s guided AI building tool. Each brand personality option shows an example of its vibe; I chose “sophisticated.” Blueprint AI then generated assets from the answers so far, created About and Contact pages automatically (I added a Services page with a checkbox), and suggested a color palette and font pairings to match. Then I clicked finish.
1.7 Choosing a brand personality in Blueprint AI.
Everything generated at this stage is a starting point. From the dashboard, Website and Pages lets you add pages and sections. In the editor, you can drag elements, add blocks like video, images, and forms under View Layouts, explore layout suggestions, and add pre-made sections or build one from scratch. Here’s a step-by-step editing tutorial for Squarespace ->
For Cherie, the essential piece is the blog, so I added a blog collection under the main navigation and chose a layout. Squarespace doesn’t currently have an AI chat assistant that makes edits for you, but the framework for editing and publishing content is already built in. Opening a post and hitting edit shows how easily you can add video, galleries, text, and accordions. There’s nothing to connect. The CMS is Squarespace. Publishing means choosing a paid plan under the billing menu.
Squarespace at a glance:
There’s no prompt box; Blueprint AI walks you through questions on category, functionality, and brand personality instead.
Everything generated is a starting point, refined in an editor built for direct manipulation.
The CMS is Squarespace itself—there’s nothing to connect and no separate software to maintain.
Publishing simply means choosing a paid plan; hosting is part of the package
What happens after launch: 8 comparisons
Building the site is the easy part. What a business owner lives with is everything that comes after. Here’s how the three builds compare.
1. First-draft direction
This is where AI-led tools shine.
AI-led tools (Lovable and Claude):
One detailed prompt gave Lovable enough to produce a coherent homepage, hand-drawn style illustrations, a set of pages, and an admin dashboard within minutes; Claude produced a full site with working navigation from the same prompt.
Both asked clarifying questions before building.
For exploring ideas, testing a direction, or showing a client what a concept could look like, this kind of first draft is valuable.
Treat it as a starting point and review it, since the same prompt can produce a different result on the next run.
AI-assisted tools (Squarespace):
Blueprint AI takes a guided route. It doesn’t read a prompt; it asks about category, functionality, and brand personality instead.
It generates simple pages, a palette, and font pairings to match—more predictable and less exploratory.
You land directly in an editor built for the next step.
2. Ease of customization
AI-led tools (Lovable and Claude):
Most changes go through the chat. You can also annotate the canvas, change text inline, or select a single element and ask for a revision to that element only, which worked well when I asked for a more realistic version of one image.
Claude’s design mode adds an edit button for text, colors, and some layout properties on a selected element, plus the same select-and-ask approach for everything else.
Any change the interface doesn’t expose means writing a prompt, consuming credits, and waiting for a regeneration.
Because the output is non-deterministic, a small request can occasionally touch more than you intended or lead to incoherent elements.
AI-assisted tools (Squarespace):
The editor is built for direct manipulation: drag elements, add blocks, swap images, attach links, edit alt text, add animations, and freely add, delete, or reorder elements in a section like testimonials.
Blueprint AI’s palette and font pairings are suggestions, and all of it can be changed without writing a prompt.
For everyday adjustments, that directness might be better than a chat assistant, though the user needs to be familiar with how to navigate the platform.
3. Extent of customization
AI-led tools (Lovable and Claude):
The widest reach: almost anything can be requested through the chat—new fields in the admin dashboard, a navigation dropdown, a payment flow, a different illustration style.
The limits show up in the editing interface. Claude’s design mode, for instance, lets you change text, colors, and some layout properties on a selected element, but moving elements or sections around means asking for it.
Every custom feature you add is one more piece of code to secure, audit, and maintain.
AI-assisted tools (Squarespace):
Works within a defined framework of pages, sections, and blocks, with plenty of room to move: choose from layout suggestions, add pre-made sections or build one from scratch, and give each blog post its own arrangement of images, video, galleries, and columns.
For projects that need to go further, Squarespace also accepts custom CSS and JavaScript.Learn more how you can extend Squarespace features with code ->
4. Maintainability
For Cherie, maintainability comes down to one question: how easy is it to publish and update blog posts week after week?
AI-led tools (Lovable and Claude):
Lovable’s admin dashboard handles basic posts, but formatting options are limited, and adding galleries, videos, or images between blocks of text isn’t flexible.
You can prompt Lovable to add more element types, but that takes time, and their position stays standardized across every post in the collection.
Any edits outside the blog go through the chat, unless you specifically ask for those controls to be added to the dashboard too.
Claude-built sites are more involved again: either a single posts.js file to edit by hand, or a separate CMS such as Sanity, where the same kinds of limitations show up.
Beyond content, there’s the site itself to keep running. AI-generated sites depend on packages that need updates and can become vulnerable over time, and someone has to own that responsibility.
AI-assisted tools (Squarespace):
The built-in CMS lets you add any block (images, video, even columns) inside any post, and every post can have a different layout.
There’s no separate software to connect or keep updated, and platform updates are pushed to every site automatically.
For ongoing maintenance and updates, I find it a more delightful interface to work in.
5. Cost
AI-led tools (Lovable and Claude):
Expect to subscribe to the first paid plan just to finish the build, and to a higher tier for more complex functionality. Claude is similar: free credits run out quickly and may not get you to the finish line.
Because Cherie’s CMS lives on Lovable Cloud, it needs the Pro plan for cloud backend access, and each plan carries a set allowance for cloud usage. Cloud usage is metered runtime, and the allowance is finite.
High traffic, elevated form submissions, or heavy asset uploads (images, video) can push past the allowance and trigger overage billing.
Multiple sites can live under one plan, but they draw from the same pool of credits.
AI-assisted tools (Squarespace):
There’s a free trial, and publishing requires a paid plan. As of this writing, Basic is $19 per month and Core is $29 per month; I recommend Core for most of my clients.
Every plan includes unlimited bandwidth, with only video hosting and storage capped by tier, so more images, pages, or traffic won’t lead to surprise bills.
6. Security
This is the most important comparison. Lovable has integrated security checks that catch common issues, and that’s the nature of large language models: they spot patterns well, and they also make mistakes. Lovable’s own documentation says it can’t guarantee complete security.
AI-led tools (Lovable and Claude):
Revisiting the Cherie build, the flagged issues came with a button to fix them all. It’s still worth understanding where the risks come from.
Features that trigger Lovable Cloud create databases—in our case, ones storing contact form responses and blog post drafts. Keeping that data out of public view depends on correctly configured row-level security (RLS) policies.
For simple projects, AI can usually configure them, but it can get them wrong, which is why Lovable’s Security Best Practices checklist emphasizes configuring and testing them.
Security is also ongoing. As packages behind the site age, vulnerabilities can appear, which is why Lovable recommends regular security reviews.
1.8 Security issues flagged by Lovable after adding a navigation dropdown.
AI-assisted tools (Squarespace):
Works from a standardized framework and set of security protocols. Updates are pushed to every Squarespace site automatically, so owners never need to track underlying packages and code.
One concrete example: when I inspected the code of a Lovable-generated contact form, it had no spam filtering and no submission limits. Squarespace forms include guardrails like Google reCAPTCHA by default.
It’s a good reminder that an AI security check won’t consistently catch every gap.
7. SEO and accessibility
We asked both AI-led tools for a site that was accessible and SEO-optimized, then audited the results with the browser inspector and dedicated tools. Both Claude and Lovable passed most criteria, including semantic markup. But gaps remained:
A keyboard trap when opening one folder, found by testing keyboard navigation with the Tab key, the way someone who can’t use a mouse experiences a site.
Incorrect ARIA labels on one section. ARIA attributes matter a great deal for screen reader users.
No responsive image sizing from either Lovable or Claude.
No structured data in either page source.
Squarespace implements responsive image sizing automatically, and every page ships with structured data and an auto-generated sitemap.
What’s worth noting is: AI tools are non-deterministic. The same prompt can produce different code from one run to the next, and quality may differ depending on the model used. The meta description on this build was generated correctly. There’s no guarantee it would be tomorrow. That’s the case for a human in the loop on any AI-generated code. In contrast, managed platforms like Squarespace stay consistent because the fields and configurations already exist.
8. Scalability
Say Margaux wants to grow Cherie beyond blogging and affiliate links into memberships or paid digital downloads.
AI-led tools (Lovable and Claude):
With Claude Code, the stack gets considerably more complex. Ask Claude what’s involved and it will walk you through setting up Stripe, generating tokens, and wiring them into the code. Those Stripe tokens are secrets and must never appear in front-end code, so this calls for caution.
Lovable Cloud simplifies this somewhat. Switching to plan mode produced a plan with a new page and new databases for customer and payment information.
It looks straightforward on the surface, but security issues can come up along the way, and anything touching customer and payment data needs to be understood before it’s fixed with a button.
AI-assisted tools (Squarespace):
Selling is built in. The store can sell physical goods, services, or digital downloads—exactly what Cherie would need—with the download delivered automatically after purchase.
The Selling menu handles payments, taxes, and shipping, and customers get a standard account to track orders or manage a membership.
The configuration is standardized, and the security is handled by the platform.
The comparison at a glance
Aspect #listtotable
AI-led tools (Lovable, Claude)
AI-assisted platforms (Squarespace)
First-draft direction
Strong. Full pages built-out with content suggestions. Useful for exploring ideas, though it should be treated as a starting point.
Less detailed. Limited to simple sections, page structures, and style direction. Less exploratory and more predictable.
Ease of customization
Edits go through the chat or by selecting an element and prompting. Takes time and consumes AI credits.
Direct editing in a visual editor: drag, swap, reorder, and restyle without writing a prompt. But the user needs familiarity with the interface.
Extent of customization
Nearly unlimited possibilities. Any feature or element can be requested, and each addition adds maintenance and security work.
Wide within the platform’s framework of pages, sections, and blocks, with custom CSS and JavaScript available to extend features.
Maintainability
CMS is customizable but takes time or a separate app to set-up. Blog dashboards have limited formatting.
CMS built in. Any block can be added to any post, and platform updates are pushed automatically.
Cost
Paid plan needed to finish the build, plus metered cloud usage with overage billing.
Flat monthly plan ($19 Basic, $29 Core as of this writing) with unlimited bandwidth.
Security
Automated checks catch common issues but can’t guarantee coverage. RLS policies, packages, and reviews are your responsibility, on an ongoing basis.
Standardized protocols with updates pushed to every site automatically. Forms include reCAPTCHA by default.
SEO and accessibility
Output may vary between runs. Semantic markup mostly present, but gaps found: keyboard trap, incorrect ARIA labels, no responsive images, no structured data.
Consistently configured: Responsive image sizing, structured data, and sitemap generated automatically on every page.
Scalability
Adding payments or memberships means bootstrapping tools, new databases, Stripe tokens, and more code to secure and maintain.
Store, digital downloads, memberships, and customer accounts are built in and configured from the Selling menu.
Best for
Technically confident owners and teams with expert review available.
Businesses without a dedicated tech team who want predictable costs and long-term maintainability.
When to choose an AI-led tool, and when to choose a platform
I truly believe there’s room for both. They suit different people and different projects.
Choose an AI-led tool like Lovable or Claude if:
You’re confident with the technical side of building websites and understand the risks of hosting and maintaining a site yourself.
You need advanced or complex functionality built quickly, and you have expert review available, at the initial build and on an ongoing basis, since those features open up risk.
You want meaningful first-draft direction for exploring ideas. Treat it as a starting point.
Choose an AI-assisted platform like Squarespace if:
You don’t have a dedicated tech team and want established guardrails, consistent configuration, and a strong foundation, with the platform handling security and updates.
Your site needs standard functionality like blogs and stores. You can still use AI through built-in features for SEO and onboarding, orextend the functionality with code.
You value long-term maintainability, scalability, and predictable costs.
So, can AI really build a website?
Absolutely. However, whether that website serves the business over the long term depends entirely on how the AI is used. On its own, AI can’t guarantee security, maintainability, or quality. It needs a reliable platform as its foundation and expert review along the way.
For business owners who aren’t especially comfortable with tech, and who don’t fully understand the risks of hosting and maintaining a site themselves, I still think a website platform is the right call. That platform can absolutely have an AI layer to assist. The platform should be the foundation.
Frequently asked questions
Can AI really build a website?
Yes. AI-led tools like Lovable and Claude can turn a detailed prompt into a working site within minutes, and AI-assisted platforms like Squarespace use AI to guide the setup and assist in aspects such as SEO and AEO. Whether that website can safely run your business for years is another matter—on its own, AI can’t guarantee security, maintainability, or quality. It needs a reliable platform as its foundation and expert review along the way.
What’s the difference between AI-led tools and AI-assisted platforms?
AI-led code generators (Lovable, Bolt, Replit, Base44, Claude Code) take a plain-language prompt and generate a codebase you own along with the hosting, maintenance, and security that come with it. AI-assisted platforms (Squarespace, Shopify, Webflow et al) are established builders with AI features layered on top; the platform remains the foundation, and AI helps you work within it.
Are AI-generated websites secure?
Not by default. Automated security checks catch common issues, but they can’t guarantee complete coverage—Lovable’s own documentation says as much. When I inspected an AI-generated contact form, it had no spam filtering and no submission limits, while Squarespace forms include guardrails like Google reCAPTCHA out of the box. If you build with an AI-led tool, plan for expert security reviews at launch and on an ongoing basis.
How much does an AI-built website cost?
With AI-led tools, expect to subscribe to a paid plan just to finish the build, plus metered cloud usage that can trigger overage billing as traffic, form submissions, or asset uploads grow. AI-assisted platforms charge a flat rate—on Squarespace, $19 per month for Basic or $29 per month for Core as of this writing, with unlimited bandwidth on every plan.
Are AI-built websites good for SEO and accessibility?
Partly. In my audits, both Lovable and Claude passed most criteria, including semantic markup, but gaps remained: a keyboard trap, incorrect ARIA labels, no responsive image sizing, no sitemap, and no structured data. And because AI is non-deterministic, the next run could produce different results. Managed platforms configure these consistently on every page.
Should I build my business website with AI or a website platform?
If you’re confident with the technical side, understand the risks of hosting and maintaining a site yourself, and have expert review available, AI-led tools can accelerate complex builds. For everyone else—especially businesses without a dedicated tech team—an AI-assisted platform is the safer long-term home, with predictable costs and security handled for you.