Indig: How ChatGPT cite sources


Key Takeaways:
Indig analyzed over 21.000 ChatGPT citations from different industries in depth to cover how AI cites sources based on content length and which pages/domains are picked etc.
His findings / statements:
Text with a question mark is 2x more likely to be cited (18.9% vs. 9.4%)
43.2% of sources cited by ChatGPT rank in Google position #1
ChatGPT retrieves about 6x more pages than it cites
Top 10 domains take 46% of all citations in a topic, Top 30 at 67%
While pages with lower than 500 words are cited 2.5x articles with 2,000 to 5,000 words are cited 3.6x and with >5,000 words more than 4x --> especially small pages tend to answer questions in a way that LLMs can based on their training data
58%/67% of sources are cited once / in one of the prompts
His recommendations:
Cover topics broadly, answer a cluster as a domain - matters more than domain authority
Write comprehensively - more words for a page correlate with higher citations & multi-intent pages perform better than a bunch of single-intent pages
Put your most citable claims and data in the first 30% of the page to ensure citations (summarize)
Write conversation style using questions and answers
Write definitively, do not use vague language
Use entities / terminology
Form a balanced sentiment: pros and cons, tend to write objectively
Try to write in simple language (as far as the topic allows; compared to other texts)
Summarize & conclude - try to achieve high information density






Sources:

![ICYMI Metehan: Statistics & Fact Checking in AI Chats and beyond the Web [Experiment]](https://static.wixstatic.com/media/24c590_b0a8e1392c1c47f28a0a46a67d59df71~mv2.jpg/v1/fill/w_980,h_653,al_c,q_85,usm_0.66_1.00_0.01,enc_avif,quality_auto/24c590_b0a8e1392c1c47f28a0a46a67d59df71~mv2.jpg)
