Audience insight · YouTube

Separating prompted responses from audience discussion

A Beast Philanthropy video featuring The Phoenix drew 50,338 comments. In an AI-assisted analysis, I counted responses to a donation-linked call to action and mapped the themes in the wider discussion.

The work

Sole author and analyst. I treated the comment section as primary audience data: I normalized spelling, punctuation, capitalization, and spacing variants of the video’s “Together We Rise” call to action and counted 2,270 instances, roughly 4.5% of all comments. I then mapped the recurring discussion themes.

The phrase was incentivized. Beast Philanthropy and Stand Together donated a dollar for every comment carrying it, matched two to one up to $200,000. The 2,270 figure counts normalized phrase instances in the comment corpus. It is not a count of unique participants or a viewer-conversion rate.

What 50,338 comments actually said
ThemeCharacter
Recovery milestones and peer supportDominant: sobriety counts, encouragement, “one day at a time”
Dignity and empowermentRecurring: the approach treats people as capable, not as patients
Call-to-action response (incentivized)2,270 normalized “Together We Rise” phrase instances (~4.5% of comments); a dollar was donated per comment
SkepticismA minority: addiction-as-choice arguments, met with pushback in threads
Belief in scalabilityRepeated: viewers see the model as expandable and want a role in it
Theme map, qualitative by design. Sentiment is characterized qualitatively because the percentage-level coding methodology isn’t documented with a reproducible evidence trail.

Method and limits

The retained analysis documents the corpus size, the normalized call-to-action count, and the qualitative theme map. It does not preserve a reproducible extraction, deduplication, or coding log. I therefore do not present percentage-level sentiment estimates or compare the call-to-action share with a platform engagement benchmark. If I rebuilt it, I would preserve the source data, normalization rules, coding schema, and an auditable analysis script.

What it established

The prompted phrase appeared 2,270 times, while the unprompted comments repeatedly returned to sobriety milestones, peer support, dignity, and disagreement over whether addiction is a choice.

The sharpest pattern was in the replies: some commenters argued that addiction is a choice, and others pushed back without being asked. That is the thread I would build a stronger methodology around next time. I would also preserve the evidence trail underneath the theme map so each conclusion could be defended line by line instead of summarized.

Attribution Sole author. Comment count as of March 2025.