An analysis of 60 Perplexity AI responses revealed that the search-based AI model frequently cites paragraphs ranging from 30 to 80 words in length. The study compared 113 cited paragraphs against 4390 non-cited ones extracted from 79 source pages.
The median length of a cited paragraph was 41 words, compared to 23 words for non-cited ones. Three-quarters of the cited paragraphs exceeded 29 words, while three-quarters of the non-cited paragraphs were shorter than 36 words.
The research found that starting paragraphs with transition words like 'therefore' or 'this' does not affect the likelihood of being cited. However, ensuring that the subject is explicitly named within the paragraph rather than relying on headers improves citation chances.
The author suggests that content creators should merge short fragments of 10 to 20 words into complete thoughts. A Python script was developed to evaluate text against these criteria, focusing on paragraph length, explicit subject naming, and the presence of numerical data.
