You paid for a custom website. What you may have received is a prompt, an export, and an invoice. AI site builders like Lovable, Bolt, v0 and Framer AI can produce a working marketing site in an afternoon — which is genuinely useful, but it becomes a problem when an agency bills bespoke-development rates for it without telling you.
This guide covers how to check what your site was actually built with, which signals are reliable, and what to do if the answer isn't what you were promised.
Why This Matters (And When It Doesn't)
Using AI tooling isn't inherently dishonest. Plenty of good agencies use AI builders to produce a first draft, then rewrite the output, restructure the markup and hand over something solid. The result can be perfectly fine work.
The problem is narrower than "they used AI." It's worth investigating when:
You're paying custom rates for template output
A five-figure invoice for what amounts to an unmodified AI export is a pricing problem, not a technology problem. The question isn't which tool they used — it's whether the price matches the labour.
You were told it was hand-built
If "fully custom, built from scratch" appeared in the proposal and the site ships with an untouched builder signature, that's a misrepresentation worth raising regardless of output quality.
You can't get the source
Some AI platforms keep you locked into their hosting and editor. If you were expecting a repository you own and can hand to any developer, and instead you have a login to someone else's dashboard, that affects what the asset is actually worth to you.
The Fastest Check: Look at the Metadata
Most builders leave a signature in the page source, and many agencies never remove it. Open your site, view source, and search for a generator meta tag:
<meta name="generator" content="...">
Wix, Squarespace, Webflow, WordPress and Durable all commonly populate this. It's the single highest-signal thing you can check in ten seconds, and it requires no technical background. If it names a builder, you have your answer immediately.
Our AI Website Detector automates this and the rest of the checks below — paste your URL and it returns a probability score plus the tech-stack evidence behind it.
Asset URLs and Hosting Fingerprints
The second-best signal is where your images, fonts and scripts are actually loading from. In your browser's developer tools, open the Network tab and reload the page.
Platform-hosted assets
Framer sites typically serve assets from Framer's own domains. Wix and Squarespace do the same. If the bulk of your images and scripts come from a platform CDN rather than your own domain, the site is running on that platform — whatever the agency called it.
Deployment headers
Response headers often name the host. Vercel, Netlify and similar platforms identify themselves, which is a strong hint the project came out of a modern AI-assisted stack rather than traditional hosting.
Framework defaults left in place
Default favicons, placeholder social preview images, and untouched robots.txt or manifest.json files are all signs that nobody did a finishing pass.
Reading the Markup Itself
You don't need to be a developer to spot the patterns here.
Utility-class soup
AI builders lean heavily on Tailwind CSS. If every element carries long strings of utility classes and there's almost no custom stylesheet, that's characteristic of generated output. It isn't proof — many hand-built sites use Tailwind deliberately — but combined with other signals it's meaningful.
Component library defaults
Tools built on shadcn/ui and similar libraries produce recognisable, repeated component structures. Buttons, cards and dialogs across unrelated pages that share identical structure suggest generated scaffolding.
Leftover placeholder content
Lorem ipsum in a hidden section, an unedited "Your Company" string in the footer, stock testimonials with no names, or three identical gradient hero blocks are all tells that the export was shipped without a careful review.
Signals That Are Easy to Misread
A few things get treated as evidence but aren't:
Modern design alone means nothing
Gradient backgrounds, large rounded corners and generous whitespace are current design conventions, not proof of AI generation. Good designers produce work that looks like this on purpose.
A fast build isn't suspicious by itself
An experienced developer working from an established component system can move very quickly. Speed is not evidence of shortcuts.
Detection scores are probabilistic
Any detector — ours included — returns a likelihood based on observable signals, not a verdict. A high score means the site carries multiple builder fingerprints. It doesn't establish intent, and it can't tell you how much human work went in afterwards. Treat it as one input alongside your contract and your own review.
What to Do With the Answer
If the evidence points to an unmodified AI build and that contradicts what you were sold, handle it as a commercial conversation rather than an accusation.
Document what you find
Screenshot the generator tag, the asset domains and any placeholder content. Note the date. Concrete evidence keeps the discussion factual.
Re-read the deliverables clause
Check what was actually promised. Terms like "custom design" and "bespoke build" are often loosely defined. What matters most is what the contract commits to on source-code ownership, hosting and handover.
Ask direct, answerable questions
"What was this built with?", "Do I own the repository?", "Can another developer maintain this without a subscription to your platform?" These are reasonable questions and a straight answer to each tells you what you need to know.
Establish what you can actually do with it
The practical question is portability. If you can export the code, host it anywhere and hire anyone to work on it, you have a real asset. If you can't, that limitation should have been disclosed before you signed.
Checking Content as Well as Code
The site's copy is worth reviewing separately from its markup. If you commissioned original writing, generated text is the same category of question as generated code. Our AI Content Detector analyses text for AI-generation patterns, and the AI Essay Detector is tuned for longer-form writing.
For a broader technical view of the site, the HTTP Headers tool shows the server and platform signatures discussed above, and Website Speed Test reveals whether unoptimised generated assets are affecting performance.
The Bottom Line
AI builders are legitimate tools, and a site built with help from one can be perfectly good. What you're entitled to is an accurate description of what you bought, source code you own, and a price that reflects the work performed.
Checking takes about a minute. Run your URL through the AI Website Detector, look at the generator tag and asset domains yourself, and you'll know where you stand well enough to ask the right questions.
Frequently Asked Questions
Can a detector prove my agency used AI?
No. It identifies technical fingerprints associated with AI builders and returns a probability. Strong signals like an unmodified generator tag are close to conclusive about the platform, but no tool can measure how much human work followed.
Is it wrong for an agency to use AI tools?
Not by itself. It becomes a problem when the work is billed as fully custom, when the build was misrepresented, or when platform lock-in wasn't disclosed.
What if the site works fine?
Then the technology matters less. The questions worth pursuing are ownership, portability and price — not the tool. A site that works, that you own and can maintain, is a good outcome regardless of how it was produced.
Can an agency hide that AI was used?
Yes. Removing the generator tag, self-hosting assets and rewriting the markup will defeat most detection. In practice this takes real effort — which is itself the custom work you were paying for.
Does this apply to WordPress or Webflow sites?
The same principles apply to any platform. Those builders aren't AI tools, but the underlying questions — what was built versus configured, what you own, and what you were charged — are identical.
