RFC: We need a standard for "AI SEO". Here is my proposal.

#webdev #ai #discuss #opensource

TL;DR: I built an open-source scoring engine to test how "friendly" websites are to AI Agents. The goal isn't to kill modern UX, but to enable "Dual-Mode" access.

The Problem

We build rich SPAs (Single Page Apps) for humans. We use Tailwind, Framer Motion, and heavy hydration. This is great for UX (User Experience).

But for AI Agents (RAG, LLMs, Search Bots), this "Visual Layer" is expensive noise.

Cost: Reading a hydrated DOM costs 5x more tokens than necessary.
Hallucination: Visual proximity (CSS) does not equal logical proximity (DOM).
Invisibility: Data locked behind client-side JS is often missed entirely.

The Proposal: OpenAIX Protocol

I drafted a standard called OpenAIX. It’s not about going back to raw HTML. It’s about "Visual/Data Separation".

I wrote a Python scorer to audit websites. Here is the actual logic I am testing. Does this align with how you build apps?

1. The "Dual-Mode" Check

Can your site serve two masters?

Metric: Does the site offer a "Fast Lane" for AI?
My Logic: I check for /llms.txt (Markdown summary) or JSON-LD metadata.
The Point: You can keep your heavy React UI, AS LONG AS you provide structured metadata for the Agent.
- React only = Score C
- React + JSON-LD = Score A

2. Signal-to-Noise Ratio (SNR)

Metric: (Semantic Content) / (Total Code Volume).
My Logic: If you use 50 utility classes to style a button, that's fine for UX. But is the button accessible via a clear API endpoint or <a> tag?
The Penalty: Heavy DOM nesting without semantic tags (<article>, <nav>) gets penalized.

3. Hallucination Risk

Metric: Structural integrity of key data (Price, Date, Specs).
My Logic: If "Price" and "$99" are separated by 50 <div>s, the hallucination risk increases.
- The Fix: Wrap them in a semantic container or use Schema.org.

Roast My Logic

I open-sourced the protocol and the scorer here:

👉 [https://github.com/OpenAIX-organization/openaix-core]

My Question to You:
If "AI Search" (like ChatGPT Search) becomes the main traffic source, would you bother adding a json-ld layer or an llms.txt file to your app?

Drop your URL below, and I'll run the OpenAIX Audit to see if your site is "Dual-Mode Ready". 👇