47 YouTube channels use AI presenters to report German politics

A woman in a widely shared video calmly analyses the election in Saxony-Anhalt. She has a face, a voice, a channel and more than 1,800 videos behind her, if you count the whole network. She does not exist.

The non-profit Centre for Monitoring, Analysis and Strategy (CeMAS) published the research paper "This Expert Does Not Exist" on 14 September 2026, identifying 47 German-language YouTube channels that use AI-generated presenters to discuss politics and geopolitics. CeMAS identified the channels in July 2026, analysed 38 of them in full and documented more than 1,800 associated videos. Euronews reported the findings on 21 September, counting 1,876 relevant videos published between December 2025 and August 2026, with more than 66,000 subscribers, about 7.55 million views and nearly 30,000 comments.

German domestic politics appeared in 795 of the videos counted by Euronews, alongside a heavier focus on the wars in Iran and Ukraine. The report describes the German-language videos as portraying the country in a profound political and economic crisis, with the government of Chancellor Friedrich Merz held responsible. CeMAS describes indications of a coordinated pattern in the channels' registration and activity windows, their content and their presentation, rather than proof of a single operator, and says its reading of user interaction shows the videos reached real viewers.

What actually gives them away

The tell is not the picture. Researchers found the giveaways in the details around it: translation errors in the narration, voices that change from video to video, and thumbnails featuring a different avatar from the one who appears in the video. Those signals are getting harder to spot, which is the finding that matters for anyone who watches video online rather than makes it.

Disclosure is not absent. Many of the videos carry a label saying the content was altered or synthetically generated. The researchers' complaint is that the label does not say what was generated: it does not distinguish a video where a machine made the presenter from a video where a clip was colour-corrected. A viewer who reads it learns that something was touched, not that the person talking is software.

Why the format works

An AI presenter costs almost nothing to run and can publish on a schedule no person would sustain. That is the same economics that made synthetic voices and avatars the default for faceless channels in every language, and it has been good for people making honest channels. What the German network shows is the other use of the same tool: an invented authority, calibrated to be plausible rather than convincing under inspection, and published fast enough that fact-checking arrives late.

For a creator watching this beat, three things follow. First, the tools are not the story; distribution is. The channels grew on the ordinary recommendation mechanics every channel uses. Second, the disclosure label is doing the work that a policy document says it does, not the work a viewer assumes it does. And third, the same generated-presenter stack that carries a political claim can carry a product review, a tutorial or a channel's own brand, which is why the question of who is on camera is now part of the format decision rather than an afterthought.

CeMAS's paper is a research paper rather than a peer-reviewed study, its 38 analysed channels are a subset of the 47 it identified, and the per-video counts above are Euronews's.