The most honest user research in your category runs continuously, unpaid, on camera: reviewers using products — yours and your rivals' — while narrating exactly what they think. "Why is this setting buried here." "This took me three tries." "Nobody will find this feature." It's think-aloud protocol, the gold standard of usability research, performed publicly at scale — and PM teams pay agencies for lab versions of it while the free corpus grows unread.

The pain points in those narrations are roadmap input of the best kind: unprompted, specific, and demonstrated rather than recalled. A user saying they struggled in a survey is data; a reviewer struggling on camera while explaining why is evidence.

deep_search finds the friction moments across your category: transcript-level search means "videos where reviewers struggle with <workflow>" returns those exact narrations, evidence-timestamped — across your product, your rivals', and the category's generic tooling.

library_import_youtube_video and get_youtube_transcript turn the richest sessions into analyzable text: import, pull the verbatim narration, and harvest the pain inventory — what they struggled with, what they said while struggling, what they tried first (your users' mental model, documented). The output reads like a usability-lab report, sourced from sessions nobody had to schedule, recruit or pay for.