How to Tell If a Website Is Made by AI: 10 Checks You Can Run Yourself

Our 10 signs of an AI-built website explains what to look for. This guide is the practical companion: ten checks you can run yourself, in the order we'd run them, with exactly where to click and what each result does and doesn't prove. The first five take a minute in any browser. The rest go deeper, into the files and records that AI app builders can't easily hide.

Each check has a label for how much a positive result is worth. Strong means one hit is close to an answer on its own. Medium means it names the platform but not necessarily AI. Weak means it only counts alongside something stronger.

1. 1. Look for a badge or credit lineStrong

   Scroll to the very bottom of the page and look at the bottom corners of the screen, on desktop and on a phone. AI app builders such as Lovable, Bolt.new and Base44 show a small badge naming the platform on free-plan projects. Some site owners also credit the tool in the footer on purpose.

   What it proves: If it's there, that's close to a direct answer. If it isn't, it tells you nothing: every one of these builders lets paying customers remove the badge.

2. 2. Open the page source, not the inspectorMedium

   Press Ctrl + U (on a Mac, ⌥ ⌘ U) to see the HTML exactly as the server sent it. The inspector shows the page after JavaScript has rewritten it, which hides some clues and adds others. Search the source for generator: some builders name themselves in a meta name="generator" tag near the top.

   What it proves: A generator tag is a reliable label for the platform. But many platforms are ordinary website builders or CMSs, so first check whether the name it gives is an AI tool at all.

3. 3. Notice how empty the source isWeak

   Scroll through the source. If it's a few dozen lines with a single empty container (often <div id="root">) and a script tag, the whole page is drawn by JavaScript after it loads.

   What it proves: This is how most AI app builders ship pages, but it's also how countless hand-written React and Vue apps work. On its own it proves nothing. What it does tell you is that checks 6 to 8 are where the real evidence will be.

4. 4. Read where the files load fromStrong

   In the source, search for https:// and look at the domains that images, fonts and scripts come from. Design-first builders such as Framer, Wix and Webflow keep serving files from their own content domains even after the site moves to a custom domain, usually a domain with the builder's name in it.

   What it proves: This is hard to remove by accident, because it means moving every image and script. Ignore ordinary links in the page text: a link to a builder's homepage is a mention, not evidence.

5. 5. Check the response headersWeak

   Open the browser's developer tools, go to the Network tab, reload, and click the first request. Under Response Headers, look at server, x-powered-by and any request-ID header named after a host.

   What it proves: Headers tell you the host, and occasionally the builder. Most hosts, such as Vercel, Netlify and Cloudflare, serve hand-built and AI-built sites alike, so treat this as background, not as a verdict.

6. 6. Open the main JavaScript file and search itStrong

   In the source, find the main script. On AI app builders it's usually under an /assets/ path with a hashed name like index-a1b2c3.js. Open it in a new tab and search for the builder's name. It's one long line, so let the browser's find feature do the work.

   What it proves: Look at what surrounds each match. Words in visible copy ("we love Lovable") mean nothing. Code does: an error-reporting hook or configuration entry named after the builder, or the address of the builder's own backend. When we tested real Lovable apps, these hooks sat a few hundred kilobytes into the file on every one of them.

7. 7. Try the source mapMedium

   Add .map to the end of the script's address and load it. If you get a JSON file with a "sources" list, it shows the original file names the app was built from.

   What it proves: Folder names and packages can point to a specific builder or to an AI SDK running in the browser. Don't expect much, though: none of the real AI-built apps we checked published one.

8. 8. Look up the DNS recordsStrong

   Use any online DNS lookup, or in a terminal run dig +short A example.com, dig +short CNAME www.example.com and dig +short TXT example.com. Builders tell their customers which address to point the domain at, and some ask them to keep a verification record in place.

   What it proves: A record that only one builder uses is strong evidence, and it stays visible for as long as the site is hosted there. Many addresses are shared by thousands of unrelated sites, though. We list which ones are which in what DNS records reveal about who built a website.

9. 9. Read the words and look at the images criticallyWeak

   Look for copy that could describe any business, bracketed placeholders left in by mistake, and stock-like images with odd details such as warped text, extra fingers or impossible reflections. For an image you suspect, upload it to a Content Credentials checker such as contentcredentials.org/verify. Some image generators attach a signed record saying the image was made with AI.

   What it proves: Plenty of people write bland copy, and plenty of AI-built sites are edited carefully. Most sites also strip image credentials when they resize uploads. Use this only to support stronger evidence.

10. 10. Check the site’s historyWeak

    Look the domain up on the Wayback Machine (web.archive.org) and in a domain registration (RDAP) lookup. Note when the site first appeared and how much it changed between snapshots.

    What it proves: A brand-new domain whose very first snapshot is a finished, multi-section site fits a one-afternoon build, but so does a well-prepared launch. This is context, never proof.

How to weigh what you find

One strong result is usually enough: a badge, the builder's runtime code in the main script, a DNS record only that builder uses, or files served from a builder's own content domain. Two medium results that agree with each other are nearly as good. A pile of weak results, such as an empty page source, a popular host and generic copy, is a hunch, not a finding. Write down what you found and where, so someone else can check it.

What none of these checks can see

AI coding assistants that run on a developer's own computer, such as Cursor or Claude Code, edit an ordinary codebase that's then deployed like any other. There's no badge, no builder backend and no builder DNS record, so a site made that way looks exactly like a hand-written one. When all ten checks come back empty, the honest answer is "no evidence either way", not "made by a human".

The faster way

Our free detector runs checks 2 and 4 to 8 for you in a few seconds: it reads the page source, the response headers, up to three of the site's own JavaScript files and its DNS records, and shows the exact evidence for every point of the score. The methodology explains how each signal is weighted.