An undercover investigation by the investigative media outlet “Correctiv” has exposed the AfD’s use of an internal platform called “Alternita.” The system runs on commercial AI models from the United States such as ChatGPT (OpenAI), Gemini (Google), and Claude (Anthropic) and is used to generate social media posts, press releases, and graphics in the party’s style.
According to the party’s marketing materials, the platform’s aim is to help members quickly produce campaign content and to ensure a unified political line online. “You feed it a message, and out comes ready-made propaganda: posts, press releases, graphics, in minutes, in the party’s style,” Correctiv reports.
Results of undercover tests
Correctiv journalists, posing as AfD members, gained access to the tool. The platform also produced anti-LGBTQ messages during the tests. Among other things, a post was generated warning about a supposed “rainbow dictatorship.” This text was automatically linked to an AI-generated graphic depicting a burning rainbow flag.
Xenophobic content can also be produced so easily: after uploading migration-related articles, the AI, on request, generated increasingly provocative wording and called for an “immediate and comprehensive remigration” as well as a “citizenship moratorium.” The AI is not picky about sources: test material uploaded from a neo-Nazi publication was processed by the system without blocking or error messages to create new posts.
The AfD typically does not label AI propaganda
The investigation raises significant questions about labeling requirements and transparency in digital campaigning. Since the posts produced by Alternita are generally not labeled as AI-generated when published on social media, there is concern that voters may be deceived about the origin of the content. Moreover, the case demonstrates how the security barriers of commercial AI providers can be bypassed by integrating them into third-party platforms like Alternita to orchestrate targeted, polarizing campaigns — including hate speech that would ordinarily be prohibited.
Markus Beckedahl, the executive director of the Center for Digital Rights and Democracy, told Correctiv that the AfD’s use of this tool places it in a “legal gray area.” “What we’re seeing could also become the command center of a bot army that automatically pumps far-right content into social networks at the speed of one post per second,” he warned. The question remains whether OpenAI, Anthropic, and Google take their rules seriously and enforce them as they have done in other countries.
