What does transparency for AI generated content look like in practice - an Australian take

Picture this: your next amazing blog post is locked and ready for publication – and you’re ahead of schedule thanks to time saved through use of AI. You know you should be ‘transparent’ about that use, but how do you actually go about it? What does ‘transparency’ even look like?

‘Transparency’ is a frequently used term described as a core part of artificial intelligence (AI) governance, and responsible AI more broadly. In principle, transparency is about ensuring that individuals are aware of when they interact with AI systems or consume AI-generated content. As AI-generated or augmented content looks more realistic and is harder to distinguish from human-made material, transparency is increasingly essential for building public trust, and protecting people and communities from AI-generated content which may be false or misleading or which can drive or enable trends of misinformation, fraud, identify theft, defamation and false advertising.

But what does transparency actually look like in practice?  

On 28 November 2025, the Australian Government released voluntary guidance: Being clear about AI‑generated content (Guidance). The Government described this Guidance as providing “up‑to‑date best practice approaches to AI-generated content transparency based on the latest research and international governance trends”. The Guidance supplements Australia’s earlier Voluntary AI Standards (see our previous article) and Guidance for AI Adoption: Implementation practices. 

To what extent does the Guidance change the way we understand and practise transparency with AI-generated content? Here are three of our initial observations. 

1. There’s more to transparency than just adding labels. 

The approach to transparency in relation to AI-generated content has been compared to an analogy of disclosing applicable terms and conditions on advertising materials (for example, including fine print on a billboard that the advertised offering is subject to further terms and conditions). Similarly, it is common practice for AI-generated content creators to include a label on the AI-generated content that discloses that it was AI-generated (for example, as a fine-print label in the corner of an AI-generated content image). While the Guidance recognises labelling as one transparency technique, it also recommends two other techniques, being:

• Watermarking: a way to embed information into digital content so you can trace its origin or verify its authenticity. Watermarks can be visible or invisible. Visible watermarks can appear as semi-transparent overlays in images and videos or audible disclosures in audio content. Invisible watermarks embed hidden data that requires special tools to detect and verify.
• Metadata recording: descriptive information about a piece of content that is generally included with the content file. It can include many details about a piece of content, like who created it and when, and whether it has been edited.

Watermarking and metadata recording can provide more useful and robust transparency protection on top of labelling which can be susceptible to unauthorised edits (e.g. cropping, photoshop removal, etc).  

While the Guidance is non-binding, its recognition of watermarking and metadata recording may assist in setting a market standard for transparency, which may in turn drive consumer expectations and even inform qualitative legal tests where transparency is a practical consideration.

Understanding the various techniques is important; however, determining the appropriate technique to apply to each piece of AI-generated content presents a separate challenge. 

2. There is no ‘one-size-fits-all’ approach to transparency.

At a conceptual level, it is important to understand AI-generated content not only in terms of different media formats (e.g. text, image, video, audio, etc), but also in terms of the degree to which AI is used. The Guidance considers this on a helpful spectrum:

• AI-assisted content: AI helps with tasks like spelling, grammar checks, or simple photo touch-ups (e.g. red-eye removal).
• AI-enhanced content: AI refines material based on user input, such as fact-checking, major edits or rewrites (e.g. for tone), or removing elements from images.
• Fully AI-generated content: AI creates content from scratch with little human involvement, like generating videos from images and scripts or new artwork from verbal instructions.

Understanding these categories can help with assessing transparency risks, as risks vary by AI-generated content type. Which transparency mechanism you select to use (i.e. labelling, watermarking, and metadata recording in order of strength) should generally match the potential risk level.  

The chart and table below (from pages 20 and 31 of the Guidance) show that greater transparency is needed if AI-generated content can cause negative impacts with limited human oversight. For example, stronger transparency mechanisms (e.g. metadata recording and watermarking) may be needed for AI-generated content that is capable of being used in a clinical setting which could lead to a risk of misdiagnosis, or in a recruitment process which could lead to a breach of employment law. Critically, transparency in these contexts may be in addition to any other regulatory approvals required and on its own transparency does not ensure compliance with other applicable rules and regulations. In contrast, only minimal transparency may be needed (or even none at all) when AI-generated content has low potential impact, or there is only low AI involvement (e.g. grammar corrections in casual emails, or automated brightness adjustments to a personal photo). 

Source: “Being clear about AI-generated content: A guide for business” licensed under Creative Commons Attribution 4.0 International Licence CC BY 4.0”.