1. Purpose
Generative artificial intelligence is becoming a common tool in the creation of text, images, audio, video, and other digital content. AI can be used to conceive, retrieve, translate, organise, polish, or rewrite, and it can also take part in generating part or even all of the content. As AI reaches deeper into content production, readers, viewers, and users of content may no longer be able to tell from the finished work alone what role AI played in it. AICD (AI Content Disclosure Initiative) therefore proposes a simple, open, voluntary principle of content transparency:when artificial intelligence has in fact taken part in creating content, the creator or publisher may proactively disclose that fact to the public, and explain the manner and degree of AI involvement within a reasonable scope.
AICD is not opposed to AI creation, and it makes no value judgement about AI-generated content. What AICD advocates is not "banning AI", but giving readers additional information once AI has taken part in content production. We believe creators should be free to use AI tools, and that readers should equally have the right to understand how content was made; the two are not in conflict.
2. Voluntary nature
AICD is a wholly voluntary, non-mandatory, non-profit, open initiative. Any individual, author, journalist, researcher, publishing house, media outlet, blog, website, platform, community, or other publisher of content may decide for themselves whether to adopt AICD. Adoption is entirely at the discretion of the creator or publisher; AICD does not require any person or organisation to disclose AI use, and does not penalise, restrict, discriminate against, or speak negatively of any party for not adopting this initiative.
AICD does not constitute law, administrative regulation, industry supervisory rule, platform policy, or any enforceable contractual obligation. Where any government, platform, publisher, or other organisation requires AI content labelling under its own laws, policies, or business rules, those requirements are independent of AICD's voluntary status. Adopters of AICD may stop adopting it at any time, and may adjust their disclosure approach without affecting their normal creation and publishing activities.
3. The principle of proactive disclosure
The core requirement of adopting AICD is to inform proactively, clearly, and truthfully. When a creator confirms that artificial intelligence took part in creating the content, they should proactively provide the corresponding disclosure at the time of publication wherever possible, and should not deliberately hide that information where an ordinary reader would find it hard to notice. The disclosed information should be concise, easy to understand, and broadly consistent with the actual creation process.
The disclosure AICD advocates does not require creators to publish complete AI conversation logs, prompts, internal workflows, private information, account information, API keys, trade secrets, or any other information with no direct bearing on a reader's judgement of AI involvement. The purpose of disclosure is to state truthfully whether and how AI took part, not to require creators to expose their entire creation process.
Creators may disclose in natural language — for example, "AI was used to polish the language of this article", "parts of this article were produced with AI assistance", or "the first draft of this article was generated by AI and re-edited by the author" — or they may use the standardised levels and machine-readable data formats defined by AICD. Whichever form is used, the basic principle is that an ordinary reader can reasonably understand it.
4. Degrees of AI involvement
AICD does not recommend simply dividing digital content into two mutually exclusive categories, "AI-generated" and "human-created", because modern content production often involves both human and artificial intelligence participation. AICD therefore suggests describing the degree of AI involvement by level:
- A0 — No AI: No generative artificial intelligence was used for any substantive creation, editing, or generation of this content.
- A1 — AI-assisted: AI was used mainly for ideation, brainstorming, gathering material, classifying information, analysing problems, or other supporting work; the final content was produced primarily by a human.
- A2 — AI-edited: AI was used for grammar checking, polishing, translation, rewriting, headline optimisation, restructuring, or other editing work, while the main ideas, facts, and core expression came from the human creator.
- A3 — Partially AI-generated: AI directly generated part of the content, for example some sentences, paragraphs, headings, or summaries, after which a human author reviewed, revised, integrated, or rewrote it.
- A4 — Mostly AI-generated: The body of the content was generated by AI, and a human author carried out meaningful review, editing, revision, fact-checking, or restructuring, and takes responsibility for the published result.
- A5 — AI-generated: The content was mainly or entirely generated by AI; a human only made limited selections, arrangements, formatting, or publishing decisions.
These levels describe how far AI took part in the creation process; they do not evaluate content quality, the author's ability, or the value of the work. Different types of content may pick the closest level according to their actual circumstances; where a single level cannot accurately describe a complex process, publishers may use several markers or add explanatory text.
5. Human review and final responsibility
AICD encourages publishers to state whether the content was human-reviewed alongside the degree of AI involvement. For example, one may declare "AI involvement: A3; Human review: Yes", or "AI involvement: A4; the final content was reviewed and is the responsibility of the author".
AI-generated content may contain factual errors, fabricated information, logical errors, bias, and other problems. Using artificial intelligence therefore does not automatically transfer or remove the publisher's responsibility for the final work. For content that requires accuracy, expertise, or truthfulness, AICD encourages publishers to carry out human review and fact-checking as the circumstances warrant.
AICD does not require all AI-assisted content to undergo the same degree of human review, nor does it prescribe any particular AI tool or review process; the specific responsibilities and review standards may be determined by the nature of the content, the publishing context, and applicable law.
6. AI tool information
Publishers adopting AICD may voluntarily disclose the artificial intelligence tools, models, or services they used, for example:
AI involvement: A2 — AI-edited AI tool: a generative AI model Purpose: language polishing and restructuring Human review: Yes
Tool names are not a mandatory disclosure item under AICD. Creators may complete a basic disclosure without revealing specific models, providers, or internal technical details. AICD is especially opposed to making "you must publish the specific model name" a condition of adopting this initiative, because different creators may be constrained by privacy, trade secrets, terms of service, or other practical conditions.
7. Truthfulness and good faith
The effectiveness of AICD rests on the good faith of its participants. Adopters should disclose according to the actual creation process as far as possible, should not deliberately mark clearly AI-generated content as "no AI used", and should not mislead readers by exaggerating or falsifying AI labels.
At the same time, AICD does not require creators to trace AI use they cannot reasonably confirm, nor does it require creators to prove that they did not use a particular AI tool. AICD is a proactive disclosure system, not an AI detection system.
Therefore, content carrying an "AICD A0" marker only means that the publisher declares that no AI within the scope of that definition was used in its creation; it does not mean that the maintainers of AICD, a platform, or any third party has technically certified that claim. Likewise, content carrying an "AICD A4" or "A5" marker only means that the publisher has proactively disclosed the degree of AI involvement; it does not mean that AICD certifies the quality of the content.
8. Participation is not a judgement of value
AICD does not hold that human creation is inherently better than AI-assisted creation, nor that AI-generated content is inherently inferior to human creation. The degree of AI involvement should not in itself be the sole basis for judging the truthfulness, quality, artistic value, authorial ability, or social value of content.
AICD's aim is not to build "AI content discrimination", but to build a mechanism for transparency about the creation process. Readers may decide for themselves whether to take this information into account, and publishers, platforms, or other organisations may decide under their own rules how to use it; AICD itself makes no value judgement about any mode of creation.
9. Privacy, freedom, and reasonable disclosure
AICD argues for a reasonable balance between transparency and freedom, and between informed understanding and privacy. Content creators should not be asked to reveal private conversations, prompts, personally identifying information, trade secrets, unpublished research material, or other information unrelated to AI content disclosure simply because they adopt AICD.
The "proactive notification" AICD requires should be understood as a reasonable disclosure of the fact of AI involvement, not as an unlimited exposure of the creation process. Different contexts — personal blogs, news media, academic publications, commercial advertising, literary works, social media, and so on — may adopt disclosure of differing degrees.
10. Openness and interoperability
AICD encourages implementing AI content disclosure in an open-standard way. Beyond the marker visible to ordinary readers, a unified machine-readable format can be provided in the future so that browsers, search engines, content management systems, publishing platforms, AI systems, and other software can automatically read a piece of content's AI involvement information.
On the technical side, AICD may adopt HTML Metadata, JSON, JSON-LD, HTTP Header, or other suitable data formats, and encourages compatibility with existing and future standards and initiatives for digital content provenance and authenticity, including but not limited to C2PA, Content Credentials, and W3C.
AICD does not require anyone to reinvent existing digital content provenance technology. For complex needs such as digital signatures, provenance proofs, and content integrity verification, compatibility with mature open standards should be preferred; AICD's focus remains on providing a simple, low-threshold way for ordinary creators to actually disclose AI content.
11. How to adopt
Any individual or organisation may adopt AICD; no registration, payment, or authorisation from any body is required. Adopters may declare on a website, in an article, a publication, or other digital content:
AICD 1.0 · AI involvement: A2 (AI-edited) · Human review: Yes
Or in a more compact form:
AI Disclosure: A2
For implementations that support machine reading, standardised fields may record whether AI took part, the degree of involvement, the purpose, whether there was human review, and optional tool information.
Anyone may develop plug-ins, themes, content management systems, browser extensions, verifiers, search tools, or other software that supports AICD. Any website may adopt AICD in its own services free of charge.
12. Open evolution
AICD belongs to no single commercial organisation, and does not aim to control the market for AI content disclosure. This initiative welcomes opinions, revision proposals, and technical implementations from creators, developers, publishers, researchers, platforms, and the public.
Future versions may, according to actual use, add new types of AI involvement, machine-readable formats, digital signatures, content provenance proofs, verification mechanisms, and ways of aligning with international standards. Any extension should preserve AICD's core principles as far as possible: voluntary, proactive, truthful, transparent, low-threshold, respectful of privacy, respectful of creative freedom, and free of value discrimination against AI creation.
13. Core statement
In the end, AICD advocates only one simple thing:
If artificial intelligence took part in creating content, creators who are willing to follow this initiative can tell their readers so.
We do not ask for AI to be banned, we do not ask for all content to be labelled, we do not ask creators to expose their private creation process, and we do not ask anyone to accept AI detection.
What we hope to build is a digital content culture based on voluntariness and good faith:
Creators have the freedom to use tools, readers have the right to understand how things are made; artificial intelligence may take part in creation, and the creation process can remain transparent.
Create freely. Disclose voluntarily.