WHAT IS A MARKDOWN FILE

The short answer

A markdown file is a plain text file that both humans and machines can read. It has no formatting buttons, no hidden structure, no proprietary encoding. It is just text — words and symbols arranged in a way that a person can understand by reading it and a computer program can understand by parsing it. The .md extension signals that the file uses a lightweight formatting convention called Markdown, but the key property is simpler than that: it is readable, parseable, and deployable by anything.

In the age of AI agents — automated systems that browse, retrieve, and act on web content at scale — that property is not incidental. It is the whole point.

What makes a file “machine-readable”?

Not all file formats are equal in the eyes of an automated system. A Word document (.docx) contains your text inside a compressed archive of XML files with proprietary structure. A PDF wraps content in a format designed for printing, not for parsing. An image of a contract is visible to a human but invisible to a machine without optical character recognition.

A plain text file — which is what a .md file is — contains only characters. No compression. No proprietary wrapper. Any program, in any language, on any operating system, can open it and read exactly what is written. When you put a cip.md file at the root of your domain, every AI system that visits your website can find it, open it, and read your rights declarations without any special software, any conversion step, or any ambiguity about whether the parsing worked correctly.

This is why robots.txt — the original machine-readable web standard — is a plain text file. It has worked for thirty years across every crawler, every search engine, and every AI system because it is impossible to misread. cip.md follows the same logic, extended from controlling access to declaring rights.

What Markdown formatting adds

Markdown is a set of simple conventions that add structure to plain text without making it unreadable by humans:

• A line starting with # is a heading
• A line starting with - is a bullet point
• Text wrapped in **double asterisks** is bold
• A line in the format Key: Value is a structured data pair

In the CIP declaration files, the key-value pair format is the most important:

CIP-Training-Ingestion:   Prohibited
CIP-NILP-Voice-Clone:     Prohibited
CIP-TDM-Opt-Out:          true

These lines are simultaneously human-readable (a person can understand them immediately) and machine-parseable (an AI system can extract the field name and its value in a single operation). No ambiguity. No interpretation. No hidden metadata. Just a field and a value, separated by a colon.

Why this matters in the age of AI agents

AI agents are programs that operate autonomously — browsing websites, retrieving content, making decisions, and taking actions without a human in the loop at each step. They are already widespread: search indexers, training data scrapers, content retrieval pipelines, and agentic AI assistants that browse the web on behalf of users.

An AI agent visiting your website faces a question at every piece of content it encounters: what am I allowed to do with this? For a long time, the only answer available was robots.txt— a file that could say “do not index this” but could not say anything about rights, consent, or compensation.

The CIP declaration files extend this vocabulary to the full range of questions that matter for creative content in the AI era:

A well-designed AI agent — and a CIP Platform Certified platform — is required to look for and respect these declarations before deciding what to do with content it encounters. The declaration file is not a technical nicety. It is the machine-readable equivalent of a notice on a gate.

The CIP declaration file

CIP uses a single declaration file: cip.md. It is a plain-text Markdown file that declares rights, consent, and certification status in machine-readable form.

How the files are found by AI systems

An AI system looking for rights declarations on your domain follows a predictable lookup path:

1. Check robots.txt for a CIP: reference pointing to cip.md
2. Check your HTML <head> for a <link rel="cip-rights"> tag
3. Attempt direct access at yourdomain.com/cip.md

The CIP standard requires all three signals for full discoverability.

CIP: https://yourdomain.com/cip.md

<link rel="cip-rights" href="/cip.md" type="text/plain" />

https://yourdomain.com/cip.md

The legal weight of a machine-readable declaration

A cip.md file is not just a technical convention. It is a machine-readable expression of rights that already subsist in your content under law. The file does not create those rights — copyright, moral rights, neighbouring rights, NILP rights — they arose automatically at the moment of creation. What the file does is make those rights legible in the medium where AI systems operate.

Under UK law — specifically the Data (Use and Access) Act 2025 — a rights holder who has applied a machine-readable TDM opt-out signal has a legally recognised basis to object to commercial AI training use of their content.

Under EU law — EU AI Act Article 53 — providers of general-purpose AI models must respect TDM opt-outs declared by rights holders.

In the United States, an explicit machine-readable opt-out significantly strengthens the rights holder's position under fair use analysis by establishing that the use was not authorised and that the holder had taken active steps to prevent it.

The file is small. The legal significance is not.

What AI agents are obligated to do

A CIP Platform Certified platform is required to:

1. Query the Rights Registry at ingestion time for any content it encounters
2. Read the CDR linked to that content
3. Respect CIP-Training-Ingestion: Prohibited by excluding the content from training corpora
4. Respect CIP-NILP-Voice-Clone: Prohibited by excluding voice data from synthesis training
5. Attach a Provenance Certificate to any outputs derived from content it did ingest under licence

An AI agent operating on a certified platform that ignores a cip.md declaration is not just committing a technical error. It is creating a subsisting rights failure — a legal liability — at the point of ingestion.

For each CIP audience