Use a YouTube transcript API to vet a creator before you sponsor them.

There are three routes. YouTube's own Data API has a captions endpoint, but it is built for channel owners: downloading a track needs a signed-in account with rights to the video. A free open-source Python library reads the captions any viewer can see, and its documentation warns that YouTube often blocks requests from cloud servers. Paid APIs read public captions per call.

This run used two paid transcript tools through one MCP connection, so the agent could pull a transcript and read it in the same chat. Each returned the full text with a timestamp on every line.

Views and subscribers tell you reach. A transcript tells you what you would be buying: where the ad read sits, how long it runs, how the creator introduces it and which brands already sponsor the channel. You can read that in a minute of text instead of scrubbing through two videos.

The agent searched each transcript for the word sponsor and read the lines around it. In the productivity video the read starts at 4:26 with "By the way, if you're anything like me, you might be using an iPad", names Paperlike at 4:36 and hands back at 5:44 with "Let's now get back to the video." It runs 82 seconds of a 26-minute video.

In the 51-minute coaching video the read starts at 7:20, runs 109 seconds and ends with a link in the description. Both reads open with "by the way", name the sponsor out loud and come before the 10-minute mark. A sponsor can expect the same slot, and can see that an AI product already bought one.

• Who already sponsors the channel, from the last few transcripts.
• Where the read sits, and how far into the video.
• How long it runs and how it is introduced.
• Views per video against subscribers.

The channel shows 6.7M subscribers, a figure YouTube rounds. The last 30 videos had a median of 165,154 views, 2.5% of subscribers. 3 passed 500,000 views and 6 stayed under 100,000. View counts are totals to date, so the newest video, 3 days old with 166,399 views, is still counting. Price a slot on the median, not on the best video.

The channel and the 30 videos cost $0.0022 each. The two transcripts cost $0.0041 and $0.0047. The whole run cost $0.0132. Ten transcripts from one channel would cost about $0.043. The free library costs nothing per call, but you run it, and it can get blocked.

Both transcripts were auto-generated, so names can come out wrong. The second one heard the sponsor as "Perplexity Compute" twice before "Perplexity Computer". A video with no captions returns no text. When YouTube refuses the first transcript tool, a slower fallback is billed even if it finds nothing. The run read two of 30 videos and did not pull comments, audience location or watch time, which only the channel owner sees.