A Cited Page Can Still Add Almost Nothing to the Answer
A public dataset of 602 prompts finds Perplexity cites the most sources, while ChatGPT's fewer citations carry a higher average influence score.
Direct answer
The paper separates citation selection from citation absorption. In its descriptive table, mean citations per prompt are 6.88 for ChatGPT, 12.06 for Google, and 16.35 for Perplexity, while mean fetched-page influence is highest for ChatGPT at 0.2713.
An arXiv HTML paper dated 29 April 2026, numbered 2604.25707, argues that a citation and a contribution are different events. Citation selection is the moment a platform chooses a source. Citation absorption is the moment that source supplies language, evidence, structure, or a fact to the answer. The authors say the public geo-citation-lab set holds 602 controlled prompts, 21,143 valid search-layer citations, 23,745 citation-level feature records, and 18,151 fetched pages.
Their descriptive table splits breadth from depth. Mean citations per prompt are 6.88 on ChatGPT, 12.06 on Google, and 16.35 on Perplexity. Mean influence among fetched pages runs the other way: 0.2713 for ChatGPT, 0.0584 for Google, and 0.0646 for Perplexity. Perplexity lists more sources. ChatGPT's shorter list, in this snapshot, is the one with the higher average influence score.
A formatting habit that did not help
The same table says question-and-answer pages had mean influence of 0.0947, against 0.1005 for pages that were not in that format, a relative difference of minus 5.74 percent. The authors treat that as a negative finding: Q&A packaging alone did not raise absorption. They also warn that the numbers are descriptive. They say they are not claiming the features caused the engines to cite a page.
Source: arXiv 2604.25707.
FAQ
- Did question-and-answer pages score higher?
- The paper says Q&A pages had mean influence 0.0947 versus 0.1005 for other pages, a relative difference of about minus 5.74 percent.
- How large is the set?
- The abstract says 602 controlled prompts, 21,143 valid search-layer citations, and 18,151 fetched pages across ChatGPT, Google AI Overview or Gemini, and Perplexity.