“This was made by AI”: designing for the AI Act’s transparency rules

Imagine chatting with a customer support agent for ten minutes, only to find out, at the very end, that you were talking to a machine. Or scrolling past a video of a public figure saying something outrageous, with no hint that it never happened. Since 2 August 2026, the EU wants those moments to come with a warning. Article 50 of the AI Act sets transparency rules for chatbots, synthetic media and deepfakes, and the law is clear on what has to be disclosed. On how it should look, it says almost nothing. That part lands on designers, and it is a trickier problem than it seems.

Four situations, one question

Article 50 covers four cases:

• People must be told when they are interacting with an AI system rather than a human, unless it is obvious from the context.
• Providers of generative systems must mark their outputs in a machine-readable way, so synthetic audio, images, video and text can be detected.
• Deployers using emotion recognition or biometric categorisation must inform the people exposed to them.
• Anyone publishing deepfakes, or AI-generated text meant to inform the public on matters of public interest, must disclose that the content is artificially generated or manipulated.

The calendar has a twist. The Digital Omnibus left Article 50 in place, but gave generative systems already on the market before 2 August 2026 until 2 December 2026 to meet the machine-readable marking requirement. Deployer duties, like visible labels on deepfakes, were not postponed. The Commission has also published its Guidelines on Article 50 and a voluntary Code of Practice on AI-generated content, the best available guide to what “good” looks like.

Buried in the text is the line designers should underline: information must be clear, distinguishable and given at the latest at first interaction or exposure. That is not a legal footnote. It is a UX brief.

A disclosure nobody sees is not a disclosure

The fastest way to comply on paper and fail in practice is to hide the message. A line in the terms of service, a grey note in the footer, a tooltip nobody hovers over: none of them does the job. Disclosure belongs where the decision is made: at the start of a chat, next to the generated image, before the video plays.

We apply the same logic on a multilingual site we’re currently developing. The moment a visitor switches language, a short notice appears to say the text was translated by a machine-learning plugin. It is not a compliance claim, since whether machine translation falls under Article 50 is open to interpretation, but it shows the principle at work: tell people at the point of action, not in a footer.

A few rules make the difference:

• Say it early, keep it visible. A chatbot should introduce itself in the first message and stay identifiable throughout, not just once at the top of a long thread.
• Speak plainly. “Created with AI” beats any technical or euphemistic phrasing. Nobody should have to decode a label.
• Never rely on colour or icon alone. Labels need real text, solid contrast and an accessible name, so they reach screen reader users too.
• Speak the medium’s language. A badge on an image, a spoken notice in audio, a caption on video: the disclosure should live where the content lives.

And remember there are two layers. Watermarks and provenance metadata handle machine detection, but they vanish easily when content is screenshotted, compressed or reposted. The visible label and the technical mark are not alternatives: they are two halves of the same system.

From patch to system

This is where our daily work comes in. At COMMpla we design and build digital products where interface, content and code are shaped together, and transparency fits the same pattern as any other recurring element. Defined once as a component in the design system, with variants, placement rules and copy tokens, an AI label stays consistent across web, mobile and whatever surface comes next. Left to each team, it becomes five different badges with five different meanings.

There is a balance to strike. Label everything and labels become noise, and users stop reading them, which defeats the point. Decide what truly needs disclosure, test the wording with real users, and check that people understand what a label means, not just that they saw it. Where the legal boundary is blurry, as with assistive editing or creative and satirical work, bring in legal counsel early: this article is design guidance, not legal advice.

The takeaway

The AI Act’s transparency rules turn honesty into an interface requirement. Teams that treat them as a design challenge, with early placement, plain language, accessible labels and a reusable component, will ship products that are compliant and more trustworthy. Teams that treat them as a checkbox will end up with a disclosure that exists on paper and goes unnoticed on screen.
If you’re building AI features and want your interface to say the right thing at the right time, our UI/UX design and development teams can help. Contact us!

Giulia Causarano | Web designer & developer