A SISTRIX analysis of 3.8 million German-language responses to ChatGPT found that citation patterns changed after ChatGPT started using GPT-5.5.
The analysis, published by SISTRIX founder Johannes Beus, tracked 38 daily samples of 100,000 ChatGPT responses. SISTRIX compared quotes four days before the change and four days after it.
The company describes the event as a “ChatGPT Core Update,” comparing it to how Google’s Core Updates change search rankings. SISTRIX also notes that the data show correlation, not evidence that model changes caused the quote movement.
To be clear, SISTRIX sells Quick Tracking, a product used to collect data for analysis.
What the Data Shows
On a typical day during follow-up, the quotation patterns vary by 1-2%. During the May 22-23 model indicator change, that number jumped to 47%.
The average number of sources identified per response also decreased, from 30.9 before the change to 28.4 after.
A SISTRIX comparison of 800,000 responses across pre- and post-change windows shows a pattern of winners and losers.
The winners
Reddit had the biggest gain, adding 7,007 citations with 10,000 replies each for a 59% increase. Reddit was already the most cited site before the change.
German mainstream publishers have benefited across the board. Welt.de increased by 99%, faz.net by 124%, and bild.de by 83%. German broadcasting and sports platforms have also been achieved. Dazn.com is up 383%, sky.de 157%, and kicker.de 357%.
Special tools and resources are also available. Mapbox.com and openstreetmap.org are each up 83%, and justwatch.com is up 624%.
The losers
International integration platforms have seen the biggest decline. It’s actually down 47%, Tripadvisor 53%, and Glassdoor 37-52% in national versions. Expedia and Rome2rio are each down 60%.
Global technology platforms also lost ground. YouTube dropped 18%, Wikipedia 14%, and Google.com 22%. Facebook and LinkedIn are each down about 20%.
According to a list of German job and comparison sites, Kununu has dropped by 46%.
Industrial Pattern
The analysis finds three winning categories in the German data: German media, Reddit, and special tools all gained. In contrast, the three missing categories included: international aggregators, global technology forums, and German job sites.
Beus interprets the pattern as localization. For German language questions, ChatGPT answers contained original German sources after the change, including older publishers and German service products.
Whether this was the goal of the model cannot be determined from the data, according to SISTRIX.
Reddit is the exception. Despite the fact that most of them speak English, it has continued to grow. This suggests that user-generated discussions across topics continue to be perceived as highly relevant.
Why This Matters
The findings add to growing evidence that ChatGPT’s citation behavior varies by model. SEJ compiled Resoneo data showing that the GPT-5.3 mutation reduced the identified domains per response. Separate analysis showed that the default and premium ChatGPT models already generate different citation patterns.
The SISTRIX data now shows how citation patterns have changed for a visible version change in a large sample.
Looking Forward
The analysis covers the German market only. SISTRIX notes that similar patterns have been observed in other markets but has not yet published that data. The company says it will continue to track quote patterns through its Prompt Tracking product.
Featured Image: The Hypostasis/Shutterstock
