Ahrefs analyzed 1.4 million ChatGPT 5.2 prompts to show that 88% of cited URLs come from its “search” ref_type, while Reddit and YouTube sources rarely get credit despite heavy use. Citation decisions hinge less on snippets or dates and more on semantic similarity between query (and internal fanout sub-queries) and the page’s title or URL, plus natural-language slugs and content freshness.

• 88.46% of ChatGPT's citations come from the general "search" ref_type, while Reddit (67.8% of non-cited URLs), YouTube, news, and academic sources are pulled heavily but almost never cited.
• Snippets and publication dates looked like citation drivers at first (14.8% vs 4.36% snippets, 92.7% vs 36% pub_dates) but that gap disappears once Reddit's API-fed data is excluded, revealing it as a pipeline artifact rather than a real signal.
• Semantic similarity between page titles and the query is the real driver: cited URLs show higher cosine similarity to the original prompt (0.602 vs 0.484) and to ChatGPT's hidden internal "fanout" sub-queries (max match 0.656).
• Matching a page's title/URL to the model's likely sub-questions, not stuffing content with dates or snippets, is the most reliable lever for getting cited.