# AI-assisted transcription of YouTube videos on penile enlargement: analysis of their text quality and readability

## TL;DR

This study evaluates the quality and readability of YouTube transcripts about penile enlargement using AI tools, finding them generally clear but calling for more trustworthy content.

## Contribution

The first scientific analysis of AI-transcribed YouTube content on penile enlargement, focusing on English-language videos.

## Key findings

- AI-transcribed YouTube videos on penile enlargement are generally readable for a 6th-grade audience.
- Physicians produced the most videos, but no significant differences in text quality were found between groups.
- Health-related websites had better readability scores than non-healthcare videos.

## Abstract

Patients dealing with sensitive issues like penile enlargement (PE) might benefit from YouTube videos. Therefore, it is essential that the textual content of these videos is clear, trustworthy, and of high quality.

Are the AI-assisted acquired texts’ qualities and comprehensibilities of YouTube videos about PE enough and suitable for the patients?

On October 25, 2024, Google Trends analysis identified the 25 most searched phrases for “Penile enlargement.” Non-related terms were excluded, and the relevant keywords were searched on YouTube. Only content about PE included; excluding duplicates, non-English videos, YouTube shorts, those under 30 seconds, silent, and music-only videos. Videos were transcribed using Whisper AI, and their quality was assessed by M.F.Ş, E.C.T., and Ç.D. using the GQS (global quality scale) and DISCERN, the readability was evaluated via Flesch–Kincaid (FKGL and FKRE) measures. High assessor agreement was noted (Pearson r = 0.912). Videos were categorized by uploader, and metrics such as views, likes, comments, and duration were recorded. The Chi-square test was used for categorical variable comparisons; the Kruskal-Wallis H-Test was applied when normality and homoscedasticity were not met, with Bonferroni post hoc correction for multiple comparisons.

The mean DISCERN and GQS scores were 51.23 ± 13.1 and 3.32 ± 0.9, respectively. FKRE and FKGL scores were 73.12 ± 11.7 and 5.85 ± 2.1. Physicians (n = 67) produced the most videos, while academic institutions (n = 2) produced the least. No significant differences in text quality were found between groups (P = 0.067 and P = 0.051). Health-related websites exhibited lower FKRE compared to non-healthcare videos (P = 0.002), with a significant difference in FKGL as well (P = 0.019).

The video exhibited a high level of readability (indicating comprehensibility for almost a 6th-grade student). Text quality, view and like count of the videos uploaded by academic institutions was the highest.

In PE, YouTube video’s health information needs to be better quality and more trustworthy, according to our research. The language used in videos should be easier to understand.

This study is the first scientific analysis of YouTube video transcripts on PE using AI, focusing specifically on English content, which limits its applicability to non-English speakers and other platforms. Exclusions of silent and shorter videos may result in the omission of valuable information.

The need for better quality and trustworthiness in health-related YouTube information, especially for PE is essential. Content makers should stress clear, accessible language and minimize disinformation.