Bernstein Analytics LLM Citation Ranking (Q4/2025)
Bernstein Analytics GmbH presents its second quarterly analysis of how large language models use sources.
The relevance of traditional media in the age of AI
Bernstein Analytics GmbH published its second quarterly analysis of how large language models use sources. The LLM Citation Ranking for Q4/2025 shows which traditional German news sites the AI models ChatGPT, Gemini, Perplexity and Grok cite most often. This edition of the ranking adds Google’s “AI Mode.” As a growing part of Google Search, it reaches a broad audience.
Artificial intelligence is changing the relevance of traditional media. More and more users get their information directly from LLMs and visit editorial news sites less often. As a result, media outlets become more important the more often and the more extensively AI models cite them. Traditional metrics such as circulation or reach no longer capture this development.
Tagesschau and Spiegel defend their top positions
In the fourth quarter of 2025, Bernstein Analytics GmbH analyzed 750 prompts on 25 relevant topics from business, domestic and foreign policy, culture and sports. The evaluation of 3,750 responses shows which sources the language models use to answer:
LLM Citation Ranking: Results for Q4/2025
The results for Q1/2026 are available in the LLM Citation Ranking Q1 2026.
Notable preferences of the language models
The study shows clear differences in how the individual LLMs select their sources:
Shifts between the rankings: There are 2 newcomers in the top 10 (n-tv.de, taz.de). welt.de (rank 11; -5) and t-online.de (rank 12; -3) dropped out. The results show a stable media landscape with clear shifts within the top 10.
LLM Citation Ranking: Changes from Q3 to Q4 2025
Which sources LLMs use when asked about the most-discussed topics across news sections in Q4/2025.
A correction of “reach-driven portals”? fr.de falls from 3rd place in Q3/2025 to 7th place. Critics have previously pointed to fr.de’s heavy SEO optimization. However, the results do not suggest a deliberate correction by the model providers. The stable 10th place of merkur.de, a sister outlet in the Ippen group, also shows this. In fact, the models use fr.de more often in absolute terms than in the previous quarter, but they use other media even more.
Grok loves media: Elon Musk’s AI chatbot Grok cites German media by far the most often. Almost 40% of its sources are media outlets, and just over 36% of these are public service media. Perplexity also uses media more than average at 31%. ChatGPT (23.6%), Gemini (19.7%) and AI Mode (19.8%) use them much less often. Grok cites (public service) media particularly often, which contradicts its developer’s critical stance toward the media.
LLM Citation Ranking: Share of German media among sources
How often German media were cited as a source in responses
Left-liberal shift: With the exception of Grok, all models in the Q4/2025 ranking use media classified as left-liberal more often than in the previous quarter. At the same time, Gemini increasingly uses sources classified as conservative. Overall, all models cite left-liberal media (22–29%) more often than conservative media (12–16%).
LLM Citation Ranking: Political preferences
A media outlet’s “bias” is based on ratings from various professional and academic projects.
LLM Citation Ranking: Comparison of political bias, Q3 vs. Q4 2025
A media outlet’s “bias” is based on ratings from various professional and academic projects.
About Bernstein Analytics GmbH
Bernstein Analytics GmbH supports its clients by systematically preparing information for strategic decision-making. Its focus is on researching, analyzing and evaluating data to accurately assess risks and opportunities in complex regulatory and communications environments. A core area is the study of narratives in AI language models (LLMs) and strategic advice on how to improve visibility in these systems.
“Tagesschau” remains the most frequently cited source by language models, but the rankings behind it have shifted: welt.de has jumped from 11th to 2nd place.…
Expect your stakeholders to have AI summarize your documents. Bottom line up front Your first reader is always AI. Every briefing and every document you…
Insights from the LLM Citation Ranking Bottom line up front Your current media strategy falls short. AI chatbots such as ChatGPT and Gemini act as…
Bernstein Analytics GmbH has published a comprehensive analysis of how large language models (LLMs) use sources. The LLM Citation Ranking shows which traditional German news…
Can a multi-billion-euro transport project, backed by two state governments and the federal government, fail because of the Dahme-Spreewald district administration? It can, as the…
Sometimes things don’t work out the way you think. While a current legislative procedure edged closer to the finish line in the European parliament, one…