A video of a politician, a photograph of a disaster, a voice note from a relative asking for money — each can now be fabricated convincingly, and each has been. Governments and platforms are responding with labelling rules for synthetic media: requirements that AI-generated or heavily manipulated content be identified as such, visibly and in its underlying data.
The rules are arriving from several directions. The European Union’s AI Act phases in transparency duties for synthetic content and deepfakes. Other jurisdictions have election-specific laws, and major platforms have committed to labelling material produced by their own AI tools. As our global news coverage tracks, the challenge is making labels consistent enough that a user can trust what a label — or its absence — means.
Three Layers of Identification
The first layer is disclosure by the creator or the tool: a checkbox when uploading, or automatic marking when content is generated inside a platform’s own software. The second layer is metadata — standards such as C2PA embed a tamper-evident record of where content came from and how it was edited, sometimes called content credentials.
The third layer is detection, where platforms scan for signs of synthesis in material uploaded without disclosure. Guidance from bodies such as UNESCO stresses that no single layer suffices: metadata can be stripped, detectors err, and honest labellers are not the problem. Combined, however, they raise the cost of deception and give journalists and investigators evidence to work with.
What Labels Can and Cannot Do
A label is information, not a verdict. Much synthetic media is harmless — film effects, accessibility tools, satire, advertising — and labelling it protects legitimate creators as much as audiences. Problems arise when labels are so common that users ignore them, or so inconsistently applied that unlabelled fakes borrow credibility from the system.
Enforcement against genuinely malicious actors, often operating across borders or on fringe services, remains the hardest part. Rules bind the large platforms where most people encounter media, which matters, but determined disinformation campaigns adapt quickly, cropping out watermarks or re-recording content to break its metadata chain.
News organisations face their own version of the task. Many now publish verification policies, label AI-assisted illustrations, and preserve original files from photographers so their work can be distinguished from fabrications. In an environment of doubt, provenance becomes part of a newsroom’s product.
The Election Test
Elections concentrate the risk, because a fake released hours before voting may do its damage before any correction spreads. Several countries now require expedited labelling or removal of deceptive synthetic election material, and platforms run dedicated operations centres during major votes. Civil-society groups monitor whether those promises survive contact with a real campaign.
Researchers caution against expecting labelling alone to restore trust. Media literacy — teaching people to check sources, reverse-search images and pause before sharing emotional content — remains the slow, essential complement to any technical rule.
Platforms are also under pressure to show their working. Researchers and regulators increasingly ask for data on how much synthetic content is labelled, how often labels are appealed, and whether detection treats languages and regions equally. A system that works well for English-language video but poorly elsewhere would export risk to the places least able to challenge it. Independent audits, public reports and researcher access are therefore becoming part of the labelling debate itself, not optional extras. That transparency will decide whether users treat labels as meaningful signals or as background noise.
The Bottom Line
Synthetic media labelling is becoming a baseline duty for platforms rather than a voluntary extra: disclose generation, preserve provenance, detect what is hidden, and explain decisions. Done well, it defends the value of authentic reporting — the same asset that reader-funded outlets depend on, as our article on local news cooperatives shows.

