CORRECTIV test: AI chatbots fake news from media outlets

Vienna, August 8, 2026. An investigation by CORRECTIV shows that popular AI chatbots such as ChatGPT, Google Gemini and Microsoft Copilot can be used to create fake news articles that appear to come from real, existing media outlets. Especially on sensitive topics, the testers were able to generate content that imitates established newsrooms, without the systems reliably preventing this.

Background

AI chatbots are used by millions of people every day, including to get information about current events. The major providers such as OpenAI, Google and Microsoft have built in safety mechanisms intended to prevent their systems from being used for disinformation or identity misuse. How effective these safeguards actually are has rarely been systematically tested until now.

CORRECTIV, a German nonprofit investigative newsroom, tested exactly that. The journalists asked the chatbots to write articles in the style and appearance of well-known media outlets, including fabricated quotes and source citations.

New Facts

According to the CORRECTIV investigation, the testers found it easiest with ChatGPT to produce fake articles that give the impression of coming from real news portals. The biggest security gaps were reportedly found in OpenAI's product, while Google and Microsoft showed stricter safeguards in some areas, but these were not flawless either.

According to the report, the testers were also able to generate content on socially sensitive topics that appears credible at first glance. The problem: such texts can easily spread on social networks and are hardly distinguishable from real reports by laypeople when the logo or layout of a well-known brand is imitated.

In its analysis, CORRECTIV emphasizes that although the chatbot providers have published guidelines against misuse, these are not consistently enforced in practice. The investigation thus provides, for the first time, concrete examples of how easily the reputation of established media brands can be misused for spreading false information.

Reactions

Official statements from OpenAI, Google or Microsoft regarding the specific CORRECTIV findings were not known at the time of publication. Media outlets whose names were used in the tests reportedly expressed concern, according to CORRECTIV, about the possibility that their brands could be used to spread false information without them being able to prevent it.

In Austria and Germany, media associations have long been calling for stricter rules on dealing with generative AI, particularly when it comes to the use of brands and editorial names. So far, there are no comprehensive legal requirements in place.

The findings from Germany can be directly applied to Austria, as the tested chatbots are just as available and widely used here. For the domestic media market, which is already struggling with declining trust in news, the investigation poses an additional challenge: if fakes can hardly be distinguished from genuine articles anymore, the pressure increases on readers to check sources more carefully, and on platforms as well as lawmakers in the EU to find technical and legal answers. The EU Commission is already working, as part of the AI Act, on labeling requirements for AI-generated content, though their practical implementation is still open.

Source: correctiv.org | Original Article