We read the comment sections of 47 YouTube videos, 2,990 comments on three very different topics (copywriting, a new AI model launch, and the origins of COVID-19), to find out what checking the comments before you watch actually gets you, and where it misleads.
Across 18 videos about the same AI model launch, 7 comment sections complained about the model burning through usage limits and 8 said it felt worse than the version before, under titles like "Beats EVERYTHING!" and "Is INSANE". Comments also caught padding ("starts at 2:55", which the transcript confirms), "new" principles experienced marketers recognized as decades old, and two creator self-corrections in 47 videos.
Social proof was the most widely shared advice across 21 copywriting videos: 17 of the 21 transcripts mention it. But the question viewers kept asking, how to show social proof with no customers yet, came up under two videos (five times under one) and none of the 21 transcripts appeared to address it. A gap like that only shows up when you compare what an audience asks with what a whole topic's worth of creators said.
The seven COVID-origins videos took different positions, but every comment section leaned the same way, including under a public broadcaster's video concluding the opposite. A comment section reflects who watches and comments on that topic on YouTube, not the strength of the evidence in the video, so the same comments read as a rebuttal under one video and an endorsement under the next.
On one popular AI channel, 72 of the 84 channel replies we sampled were signed by an AI agent the creator uses to answer comments. Scam and spam bots barely appeared in what YouTube shows publicly.
Video Triage and the right-click Rate tool now add a short "From the comments" note to a video's card: what several separate viewers flagged, where viewers say the content starts, and any real correction the creator posted. On debated topics the viewer line comes with a note that comment sections there lean one way whatever the video concludes, every note is labeled as unfiltered viewer comments, and signed bot replies are filtered out first. The quality scores still come from the transcript.
47 videos from three public Tube-U research reports; the top 100 comments per video in YouTube's relevance order (200 for the COVID set) plus the channel's own replies, read in September 2026. An AI model summarized each comment section under fixed rules (a flag needed at least two separate comments; disagreeing with a video's side didn't count as the video being wrong). Unanswered questions were checked against the transcripts by keyword, so they are a strong hint, not proof.