Free tool mines YouTube transcripts for audience pain points

A developer released a free, open-source tool called pain-point-miner that automatically scans YouTube channel transcripts and extracts ranked lists of what audiences ask and struggle with. The system works in two steps: indexer.py scans a folder of transcript .md files and builds a lightweight index, and pain_point_extractor.py reads that index, sends each transcript through Claude (Haiku) to pull out questions, pain points, and desired outcomes, then aggregates everything into one ranked report. Example output shows "Top Questions (Most Asked)" and "Top Pain Points (Most Expressed)" ranked by mention count across videos—for instance, a question about rebuilding processes around AI capabilities mentioned in 4 videos, or a pain point about existing business models built on inefficiencies mentioned in 6 videos. The creator, who spent 25 years in enterprise IT before encountering Claude Code, frames the tool as automating the scanning phase of a skill he applied for two decades in operations: reading frustration and finding the real question under the noise. He emphasizes that the tool does not replace human judgment. Setup requires cloning the repository, installing the Anthropic Python library, obtaining an API key from https://console.anthropic.com/settings/keys, and populating a local transcripts/ folder with your own .md files in the structure transcripts/<group-name>/<channel-name>/*.md. Each transcript file needs a plain text header (title, channel, date) followed by the transcript body; the creator notes that sample transcripts are not included because the test videos were copyrighted content. Running the tool involves two Python commands: one to build the index and one to run the extractor, specifying the group name. Cost is minimal—a few cents per run for a handful of files using Haiku, paid directly to Anthropic with nothing routing through the developer. A known limitation emerged during testing: one of three test transcripts failed with a truncated-JSON error because Haiku ran out of token budget mid-response. The workaround is to either raise max_tokens in the _pass1_extract() function or accept the occasional skip, since the aggregation step still completes successfully on whatever succeeded.

The tool addresses a practical problem in creator and product research: identifying what audiences genuinely care about from the noise of video comments and discussions. Rather than manually scrolling through comments or guessing at pain points, the system automates the extraction phase by feeding transcripts to Claude (Haiku), which then aggregates and ranks findings across multiple videos. The creator's background in enterprise IT—25 years of reading room-full frustration and finding the real question under the noise—informs the design philosophy: automation handles the scanning, but human judgment remains central to interpreting what actually matters. The tool's output format (ranked lists with mention counts) reflects this: it surfaces patterns and frequency, not verdicts. The cost structure (a few cents per run on Haiku, paid directly to Anthropic) keeps the barrier low for creators or teams exploring audience feedback at scale.