weid

English · Español

A convention for marking human contribution inside AI-assisted text. Three CSS classes, one attribute, no JavaScript.

_{Cover of the essay that serves as the real-world case (in Spanish) — the first text published with this convention. Image generated with Nano Banana Pro, which is exactly what a caption should say.}

weid- — the Proto-Indo-European root meaning to see. From it come vision, idea and evidence: literally, what is plainly seen. The whole family of this project in a single word: making visible which ideas are human.

---

Why

The debate about text and AI got stuck on a binary question: did a person write this, or did a machine? For almost everything written today, that question no longer has an answer, because it is not the question.

An essay can be drafted almost entirely by a model and be, at the same time, wholly original: the idea is someone's, the lived experience is someone's, the turn in the argument was born in a conversation between the two. The disclosure standards that exist cannot say that. They mark the whole document — "this involved AI" — and there they stop.

weid marks the other thing: what is human, and in what way it is.

The underlying bet: thinking is still a human activity. What changed is that it is now augmented. An honest disclosure ought to be able to show that instead of hiding it, and ought to be able to show it sentence by sentence.

The base rule

In a document declared as AI-assisted, unmarked text was written by the AI.

The marks point at the human. The opposite of what you would expect, and deliberately so: in an assisted text the human part is the scarce thing, and marking the scarce thing is cheaper — and more useful — than marking everything else.

How to use it

<link rel="stylesheet" href="weid.css">

<article data-ai-disclosure="ai-assisted" data-ai-model="claude-opus-5">
  <p><span class="weid-experience">Yesterday I saw a poster for a band with
  three sold-out nights.</span> Which is to say, thousands of people have
  already paid to be together seven months from now.</p>
</article>

Three classes on <span>:

Class	Marks	Rendered as
weid-verbatim	the person's own words	solid underline
weid-idea	their idea, written up jointly	dashed underline
weid-experience	their lived experience, anecdote or personal fact	dotted underline

And a fourth piece for the other side, the model's contribution:

<p><span class="weid-note">The counter-argument here was the model's.</span>
Variety is not the side effect of the system that works: it is the mechanism
by which it works.</p>

The note is ordinary text inside the paragraph. With CSS it moves to the margin; without CSS it is still there. Disclosure that falls apart when the stylesheet fails to load is not disclosure.

Whose voice?

A model in its default voice and a model running a skill you built from your own texts are not the same thing: there the voice is yours, reconstructed, even though the words are its. An optional attribute says so:

<article data-ai-disclosure="ai-assisted"
         data-weid-voice="author"
         data-weid-voice-source="tgdoppelganger">

model (the default, omit it) · author (your voice, from your texts) · third-party (someone else's voice — and there, naming them is mandatory, because imitating without saying so is precisely what has to be declared).

One more thing is missing, and it is mandatory: the legend. Without it the marks are decoration. It is in example.html, ready to copy.

Installing it on your blog

Two things, and neither is a plugin: load the CSS once and make the markup survive your editor.

Platform	How you load weid.css
Hand-written HTML	<link rel="stylesheet" href="weid.css">
WordPress	Appearance → Customize → Additional CSS, and paste the file. Or wp_enqueue_style() in the child theme.
Ghost	Settings → Code injection → Site header, inside a <style>.
Hugo / Jekyll / 11ty	Copy the file into assets/ or static/ and link it from the base layout.
Astro	import "../styles/weid.css" from the .astro file, not with @import inside another CSS file (see below).
Substack, Medium, LinkedIn	Not possible: they do not let you add CSS. The markup still survives as text, but it is invisible.

The Astro trap (and anything else with Vite)

If you put @import "./weid.css" inside your global.css, Vite does not watch the imported file: it caches the processed CSS and keeps serving the old version until you restart the server. It looks as if your new rules did not exist. Import it from the .astro component and the problem goes away.

The document-level attribute

data-ai-disclosure goes on the element that wraps the work, and that is usually painted by the theme's template, not by the editor. If you can't or don't want to touch the template, wrap the content inside the post itself:

<div data-ai-disclosure="ai-assisted" data-weid-voice="author">
  … the text …
</div>

It works the same: the CSS does not depend on the attribute. The attribute is for machines; the classes are for eyes.

WordPress: do you need a plugin?

No. weid's markup survives WordPress's sanitizer as is:

• KSES allows class and data-* as global attributes on the elements it already accepts, and <span> is one of them. So the marks and the disclosure attributes get through even if the author lacks the unfiltered_html capability.
• This is exactly why the convention uses <span class="…"> and not custom tags: <t-sentence> KSES does eat.

The only awkward part is writing the marks: in Gutenberg you have to use the code editor (Options → Code editor) or a custom HTML block, because the format bar has no button for this.

A plugin, then, would only save clicks: a format button in the toolbar and a way to inject the attribute into the <article> without touching the theme. It is not in v0 and it won't be until someone asks for it — the CSS already does the work.

What's here

File	What it is
SPEC.md	The convention. It is the product. Spanish version in SPEC.es.md.
weid.css	Reference implementation. One file, no JavaScript.
example.html	Open it with a double click. It is the demo and the visual test. Spanish version in example.es.html.
test.html	The tests. Open it with a double click, same as the demo. No dependencies.
test.mjs	The same tests plus print and the real demos, in a headless browser. Needs Playwright — see below.
PROMPT.md	The portable prompt: works with any LLM, copy and paste. Spanish version in PROMPT.es.md.
skills/weid/	The same instruction as a skill, for runtimes that support them.
skills/weid-es/	The same skill in Spanish. Install one or the other.

No package.json, no build, no CI. You copy the CSS and it works: Astro, WordPress, Ghost, Hugo, Jekyll, a hand-written .html.

test.html is the everyday test: double click, read the green. It renders fixtures against weid.css in srcdoc iframes — which inherit the page's origin, so it works over file:// — and resizes them across the 1100px breakpoint. It cannot emulate the print medium, and over file:// it cannot read example.html itself, so it checks the stylesheet rather than the demos.

test.mjs covers those two gaps and is the one thing with a dependency — a dependency of testing weid, not of using it. Nothing in the convention needs Node. To run it:

npm install playwright && npx playwright install chromium
node test.mjs

It renders both demos in a real browser and asserts the margin notes above 1100px, the return to the flow below it, the three line styles, the JavaScript-free toggle, both dark-theme hooks, print, and that the notes are still legible with the stylesheet blocked.

The markup is not written by hand

Nobody is going to tag 3,000 words by hand. The point is that the markup comes out of the writing session itself: the model knows which sentence was yours because you just said it. That information exists while the conversation is alive and not after — the model cannot recover it either.

So the producer comes in two forms:

• PROMPT.md — portable. Copy and paste it into any LLM.
• skills/weid/ — a skill, for runtimes that support them (Claude Code, among others). Install it with a symlink into the skills folder and it activates on its own when you are writing something that will be published. It is the strong form: a prompt is something you have to remember to paste.

The skill composes with voice skills, it does not replace them: if you are writing with a persona built from your own texts, weid detects it and declares it in data-weid-voice.

Relationship to existing standards

weid does not compete with them, it complements them:

• ai-disclosure (W3C AI Content Disclosure Community Group) solves the document level. weid uses it as is, with a data- prefix so it validates today. Its three values — human-only, ai-assisted, ai-autonomous — are the baseline; weid subdivides the middle one.
• IPTC digital source types and C2PA cover images and cryptographic provenance.
• Article 50 of the EU AI Act makes machine-readable marking an obligation.

None of them marks the sentence. That is the gap.

What weid is not

• Not a detector. It records what the author declares. It does not verify it.
• Not proof. And the distinction is in the name: weid is evidence — what stays in plain sight — not proof. A mark shows; it does not demonstrate. Anyone who wants to lie with weid can, just as they can lie in a byline. Honest disclosure serves those who want to be honest; the rest is cryptography, and that is what C2PA is for.
• Not a compliance certificate. It helps; it does not discharge the obligation.
• Not a percentage. No "17% human". The unit is the claim, not the ratio: a single marked sentence can carry the whole idea.

License

MIT. Use it, fork it, rename it.