The Milapulse AI Disclosure Standard
v0.1 — draft
We use AI to make things. We tell you exactly where.
Most media that uses AI tells you nothing, or says "AI-assisted", which means nothing. This is our alternative: a disclosure precise enough to be checked and narrow enough to be honest.
It applies to everything Milapulse publishes: video, writing, audio, images and social posts.
The principle
AI may do the labour. It may not do the thinking.
Every piece we publish exists because a human had something to say and went and found out whether it was true. Machines help us transcribe it, assemble it, caption it, translate it and cut it into the shapes each platform wants. They do not supply the argument, the reporting, the voice or the face.
The distinction that matters is not how much AI, but whose work is the thing you came for.
The three zones
Every production stage falls into one zone, and every published piece lists the zone of each of its stages.
🟢 Green — the machine does it, and that's fine
Real work that isn't the point. We use the best tool, log it, and don't apologise.
Transcribing our own recordings · assembling a rough cut from a transcript · subtitles and captions · translating our own script into Swahili, Kikuyu or Kamba (a native speaker reads it before it ships) · chapter markers, titles, descriptions, tags, alt text · finding candidate clips inside footage we shot · finding and ranking sources for us to read · arguing with our own drafts · building the tools that do all of the above.
🟡 Amber — the machine drafts, a human rewrites
Allowed, never shipped as-is. Every Amber stage names the human who worked on it afterwards, and our version history is the evidence.
A first script draft from our own notes, voice memos or transcripts, never from a topic prompt · structural edit suggestions on writing a human already wrote · a written companion drafted from our own video script · selecting stock or archive footage · pulling data out of documents we sourced · first drafts of social copy.
🔴 Red — we don't, ever
These are the promise. Break one and nothing else here is worth reading.
- The thesis, argument or opinion. If we didn't think it, we don't publish it.
- Synthetic voice or likeness, ours or anyone's. No voice clones, no digital doubles, not for a single pickup line.
- Generated footage or images presented as a real place, event or person. If you see Nairobi, someone stood in Nairobi. AI-rendered "Africa" is one of the most common fakes on the internet, and we live here.
- Fabricated quotes, statistics or sources. Every factual claim traces to something a human opened and read.
- AI-generated comments, replies, followers or engagement. When you talk to us, you're talking to us.
- AI-written or AI-"covered" music. We licence real music from real musicians.
- Undisclosed use of any of the above. Silence is the only real violation.
The ledger
Every piece ships with a machine-readable record, the AI ledger, listing each stage, its zone, the tools used, and what a human did afterwards. It is rendered at the foot of every article as "How this was made" and linked from every video description.
The ledger is written while the work happens, as a by-product of our pipeline, not reconstructed afterwards. Our publishing process refuses to release a piece whose ledger is missing or incomplete. We built that check first, on purpose: a promise that depends on remembering to keep it is not a promise.
We also keep, and will produce on request:
- the list of every source behind every external claim
- the version history of the script, showing what a first draft looked like and what survived
- hashes of the original footage
What this standard does not claim
- It is not a guarantee of accuracy. We will get things wrong. The ledger says how something was made, not that it is correct. When we're wrong we'll say so in public and date the correction.
- It is self-attested. Nobody audits us. What we offer instead is claims precise enough that a reader could catch us out, and a paper trail we have committed to keeping.
- It is not a rule for anyone else. Plenty of honest work uses AI differently. This is where we drew the line, published so you can hold us to it.
- It will change. The tools are moving. This is v0.1, versioned, and every piece records the version it was published under. If we loosen something, you will be able to see that we did.
Why publish this at all
Two reasons. The second is the honest one.
First: audiences are about to get very good at spotting machine-made substance, and unforgiving about it. Being specific now is cheaper than being trusted later.
Second: it constrains us. It is much harder to let a machine quietly start writing the argument when you have published a document saying you don't, and built a pipeline that records what you did.
Adopt it
Offered under CC BY 4.0. Take it, fork it, tighten it, argue with it. A disclosure standard one publisher uses is a marketing badge; one that several use is infrastructure.
If you adopt it, tell us and we'll link you.
Milapulse is a small independent publisher in Nairobi, Kenya, running on its own infrastructure. The platform it publishes on is built on the idea that a feed should end: no infinite scroll, no engagement ranking, friction on broadcasting rather than on reading. This document is the same idea applied to how the work gets made.
Version history
| Version | Date | Change |
|---|---|---|
| 0.1 | draft | first draft |