Research — What MotionSites.ai actually is, and how it is built

Lead-authored research pass, 2026-10-04. Sources are linked inline. Everything below was verified against the live site, its shipped bundles, and third-party references — not taken from marketing copy.

1. The idea in one paragraph

motionsites.ai is not a website builder and it is not a template marketplace. It is a library of extremely precise, copy-paste AI prompts that make an LLM coding agent produce a cinematic landing page. The product is the prompt text. The tagline is literal: "Build beautiful landing pages in minutes with our ready-to-use prompts. Just copy, paste, and launch." The user pastes one prompt into Lovable / Cursor / Claude Code / Codex and gets back a finished, animated page.

The business is therefore selling the prompt corpus, not software. Everything else on the site (Academy, MCP server, Animated Backgrounds, Affiliates, custom-prompt service) exists to feed, defend or upsell that corpus.

Source: homepage, request page, MCP page.

2. Why the prompts work — the format is the moat

I read four prompts from the publicly archived corpus (xiiiabu/motionsites.ai). The format is consistent and is the real invention:

The Asme prompt, for example, does not say "make a glassy hero". It specifies the .liquid-glass class down to background: rgba(255,255,255,0.01) + background-blend-mode: luminosity, a ::before pseudo-element with padding: 1.4px, a six-stop linear-gradient(180deg, …) edge, and the -webkit-mask-composite: xor border-only mask trick. It specifies the video fade as a 500 ms requestAnimationFrame fade-in, 500 ms fade-out at 0.55 s remaining, fadingOutRef guard against re-trigger, 100 ms reset to currentTime = 0. That level of specification is why the output looks art-directed rather than generated.

Takeaway for our clone: the prompt is the product, and specificity is the value. A vague prompt library is worthless. Our upgrade is to make each prompt verifiable.

3. Site structure (verified from the shipped router, not guessed)

I extracted the complete route table from /assets/index-g57HynbI.js (43 route components, 42 paths). The public information architecture is:

Core funnel

Free traffic + lead magnets

Money

Internal /admin/{affiliates,creators,featured,grant-access,popularity,bolt-cta,manus-cta}, /upload-prompts, /home-old, /.lovable/oauth/consent.

The gallery itself

Above the grid: category tabs All · Popular · Apps · Sections · Recent. Each card is a looping preview (.gif or Mux-hosted animated.webp) plus a title and category badge, and a copy action. The homepage grid mixes hero sections, full landing pages, footers and CTA sections — so a user can buy one section or a whole page. Sections are the cheap entry point; landing pages are the expensive one. That is a deliberate pricing ladder.

4. Tech stack (from response headers, bundle and asset paths)

Product surfaces worth noting: the MCP server is a single HTTP transport endpoint — claude mcp add motionsites --scope user --transport http https://…/functions/v1/mcp — and the free tier is gated to 3 free prompts; paid plans unlock "500+". So the actual corpus is ~500 prompts, of which 65 are publicly archived.

Creator: the site links to designrocket.io, youtube.com/@ViktorOddy and x.com/viktoroddy. The Academy page's featured videos are his.

5. Business model, decoded

Free tier (3 prompts)  ──►  paid "Unlimited" subscription  ──►  done-for-you prompts
        │                              │                              $99 / $249
        │                              │
        ├── Academy videos  ───────────┘        (top of funnel, YouTube-native)
        ├── Animated Backgrounds / Gradients    (free SEO surfaces, "Copy URL")
        ├── MCP server                          (distribution into Claude/Cursor/Codex)
        ├── Affiliates                          (paid acquisition)
        └── Creator program (submit & earn)     (supply-side scaling of the corpus)

Three things are clever here and worth copying:

1. The MCP server is distribution, not a feature. Once the library is an MCP server, every AI agent in the user's editor can pull a prompt without visiting the site. The site becomes a backend. That is the strongest idea on the page.
2. Free surfaces are the SEO moat. /backgrounds and /gradients are free, copy-pasteable assets that rank and get shared. They cost almost nothing to produce.
3. Section-level SKUs. Selling one hero section separately from a full landing page lets a buyer enter at a low price point and upgrade later.

6. Cross-check

I asked an independent model (gemini-3.5-flash, Workbuddy proxy) to review the above findings against the raw evidence and flag anything unsupported. Result and any corrections are recorded in 04-crosscheck.md.

7. What we will do differently (the "upgrade on top")

The original sells prompts on vibes — a looping GIF and a claim. Nothing tells you whether the prompt still works, what it costs to run, or which model produced the preview. That is the gap our free clone attacks:

Everything we ship is static HTML/CSS/JS, so it costs $0 to host on GitHub Pages or Cloudflare Pages and cannot rot.

Sources

• motionsites.ai — homepage, /backgrounds, /mcp, /academy, /request
• github.com/xiiiabu/motionsites.ai — archived 65-prompt corpus
• Product Hunt — Motionsites
• Tencent Cloud developer article — Chinese-language walkthrough
• Scamadviser listing — domain age / trust signals
• Shipped bundles: /assets/index-g57HynbI.js, /assets/styles-DxzY-U5-.css, /assets/Index-C7hLBIV7.js