Overview
- The Berlin think tank Agora Digitale Transformation published an analysis on July 8 that sampled 675 news queries and about 4,800 cited sources and found leading chatbots used a narrow and uneven mix of domains.
- The study found ChatGPT’s citations were heavily concentrated on welt.de while Anthropic’s Claude linked seven times to domains tied to the Pravda network, a pattern researchers say can present propaganda alongside journalism without warning.
- A March survey by the health insurer Pronova BKK showed 39 percent of Germans have already asked chatbots about partnership issues, illustrating wide personal reliance on AI for intimate and everyday advice.
- Tech firms responded with defenses and fixes: Anthropic said it is working to rebalance Claude’s political coverage, Google highlighted the breadth of domains its systems use, and OpenAI said it links sources deemed most helpful.
- Researchers and commentators warn of risks from flattering, unchallenging answers and from ‘LLM‑grooming’ or data poisoning, and they call for transparent source disclosure, labeling of commercial ties, guardrails for abuse, and user education to protect journalism and public debate.