OtterAI will automatically transcribe audio files, but its results are far from perfect. Because of this, each Love Island episode had to be manually corrected, line-by-line. The process took about 90 minutes per episode, and the resulting transcript was still not perfect. Oftentimes, I didn’t include small, one-word remarks or exclamations. OtterAI also has a habit of switching speakers a few seconds late, so that the tail end of one person’s sentence might be shown as the beginning of someone else’s line. Usually, I didn’t correct these small mistakes, because I figured they affected all cast members equally, and therefore would not have a major impact on word count percentages. To test this assumption, I went through the transcript for one episode (Episode 26) and made it as accurate as possible. This process took about four hours, so I knew that I couldn’t do this for all the episodes.

I was very relieved to find that the word count differences between the 100% accurate transcript and the mostly accurate transcript were minimal. When I calculated percentages and rounded them to the nearest whole number, only one percentage point was different between the two transcripts. The graph below shows the word count differences between the two, and illustrates why I made the decision that using MOSTLY accurate transcripts would suffice for my data going forward.