"Synthetic media" is the umbrella term for images, audio, video, and text that are generated or substantially altered by artificial intelligence. It covers a lot of ground, from a fully fabricated video of a public figure, to a real photograph with an AI-extended background, to a news summary written by a language model. For newsrooms, the category matters because the same techniques that power useful tools also power convincing fakes.

What counts as synthetic media

The label is broad on purpose. It includes deepfakes (AI-generated or face-swapped video), cloned voices, AI-generated images and illustrations, machine-written text, and hybrids that mix real and generated elements. The common thread is that a model, not a camera or a person alone, produced or changed the content.

How it is made

Most synthetic media today comes from a few families of models. Diffusion models generate images and video from text prompts. Large language models produce text. Voice models clone or synthesize speech from short samples. These systems have improved quickly, and the cost of producing a believable fake has fallen toward zero. We trace one fast-moving frontier in our report on voice cloning, and the video side in AI-generated video and how newsrooms should respond.

Why newsrooms care

Synthetic media cuts two ways. On one side it is a production tool, used for translation, summaries, and illustration. On the other it is a threat to the evidentiary value of a photograph or a recording. When any clip could be fabricated, the burden shifts to the newsroom to prove what is real. That is why detection and provenance have moved from niche concerns to core editorial infrastructure.

The question for editors is no longer "is this technology impressive." It is "can we tell, and can we show our readers, what is authentic."

Detection and provenance: two responses

There are two broad defenses, and they work best together. Detection tries to spot synthetic content after the fact by looking for statistical traces. It is useful but imperfect, as we document in why AI detectors disagree. Provenance takes the opposite approach: it attaches verifiable information about origin and edits to a file, so authenticity travels with the content. The leading standard here is C2PA, which we explain in our complete guide to content provenance. For a full toolkit, see our buyer's guide to deepfake detection tools.

A working policy for synthetic media

Newsrooms do not need to ban synthetic media to stay trustworthy. They need a policy. A practical one names the tools allowed, the disclosure required when AI is used, and the verification steps for incoming footage. The Partnership on AI has published responsible-practice guidance along these lines, and major wire services have built verification into their workflows, as we cover in how Reuters, AP, and AFP fight deepfakes.

Sources

  • Partnership on AI, Responsible Practices for Synthetic Media.
  • Coalition for Content Provenance and Authenticity (C2PA) technical standard.
  • Content Authenticity Initiative, guidance on provenance for publishers.