Embedding a hidden, machine-readable signal in AI-generated images, video, audio, or text so platforms and regulators can tell it was made by AI, now a legal requirement for content reaching the EU.
Also known as AI content marking, machine-readable marking, digital watermarking
AI watermarking is a signal built into a piece of generated content that software can detect even when a person cannot. It might be a pattern woven into the pixels of an image, a near-inaudible signature in audio, or statistical fingerprints in how words were chosen. It usually travels alongside metadata standards such as C2PA content credentials, which record how a file was made.
The watermark itself is different from a visible label. A label is a note a viewer can read (“made with AI”). A watermark is for machines: platforms, ad servers, fact checkers, and regulators. Under Article 50 of the EU AI Act, providers of generative AI tools must mark their output in a machine-readable way, and the grace period for systems already on the market runs out on December 2, 2026. Separate rules require deployers to visibly disclose deepfakes and certain AI-generated text.
Watermarking moves from a tooling detail to something on a client’s compliance checklist.
The tool stack has to keep the mark intact. Watermarks can be stripped by cropping, compressing, screenshotting, or re-exporting through another app. An agency post-production pipeline that quietly removes them creates a problem for the client, even if the original generator did its job.
Clients will ask what was AI-made. Brands running campaigns in Europe need a record of which assets were generated or heavily edited with AI, which tools were used, and whether the marking survived. That is easier to log at the moment of production than to reconstruct later.
It is not the same as disclosure. A watermark covers the machine side. Whether an ad also needs a visible “AI-generated” label depends on what it depicts and where it runs. Agencies should keep the two decisions separate in their process.
A global agency producing a holiday campaign for a retailer generates hero imagery in one tool, upscales it in a second, and finishes it in Photoshop. Ahead of the EU launch, the producer runs the final files through a content credentials checker and finds the upscaler stripped both the watermark and the C2PA record. The team swaps the upscaler for one that preserves credentials, re-exports the finals, and adds a column to the asset tracker noting the generator, the edits, and whether marking was verified. Legal signs off on the EU flight with the tracker attached.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.