Bernstein Analytics LLM Citation Ranking (Q3/2025)

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…

Published on: 16.10.2025
Reading time: 3 mins
Author: Julian Schibberges

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 sites the leading AI models ChatGPT, Gemini, Perplexity and Grok cite most often.

In the age of artificial intelligence, the relevance of media is being redefined. More and more users get their information directly from LLMs and visit editorial news sites less often. That is why the importance of media must in future also be measured by how often and how extensively AI models cite them. Traditional metrics such as circulation or reach fall short in this new environment.

Tagesschau in first place, followed by Spiegel

The new Bernstein Analytics ranking is based on an analysis of 750 prompts, written specifically on 25 relevant topics from the third quarter of 2025 in the areas of business/finance, domestic policy, foreign policy, culture and sports. Across a total of 3,000 responses, we evaluated which sources the models used to answer:

tagesschau.de
spiegel.de
fr.de

The rest of the top 10 are deutschlandfunk.de (4th), zdfheute.de (5th), zeit.de (6th), welt.de (7th), focus.de (8th), t-online.de (9th) and merkur.de (10th).

The next ranking will be published in mid-January 2026.

Notable preferences of the language models

The study reveals significant differences in how the individual LLMs select sources:

Tagesschau dominates Musk’s AI: The AI chatbot Grok cites tagesschau.de far more often than any other source. On average, the public broadcaster’s website is cited in 98 percent of prompts. Perplexity also favors content from tagesschau.de in its answers.
ChatGPT tends to cite conservative media, Gemini often left-liberal media: Looking at the split between conservative and left-liberal media in the top 20, there are clear differences between the models. ChatGPT uses media classified as conservative twice as often as left-liberal media (36% vs. 18%). For Gemini, the ratio is reversed: it uses three times as many left-liberal sources (10.5% vs. 32.5%), although it uses more neutral sources overall. Perplexity also leans left-liberal, but much less strongly (23.3% vs. 30.4%). Grok is the most balanced, with no significant difference.
Traditional journalistic media are cited rarely: The ranking only includes sources backed by a journalistic newsroom. German-language media make up only 26.35% of the sources used. This varies considerably between the models. Gemini does not use half of the top 20 media at all and relies on media sources in only 16% of cases overall. This sets it clearly apart from the other models, which use media sources in about a third of cases (ChatGPT: 27.7%; Grok: 28.1%; Perplexity: 32.3%).

Our official press release is available on presseportal.de.

Would you like to better understand how LLMs cover your topics? Feel free to contact Julian Schibberges. We are happy to help.