Turn A YouTube Video Into A Newsletter Draft And A Week Of Notes

Audio up to three hours can be transcribed. MP3 and WAV are cut into ten-minute pieces, so a long episode in either format works at any size up to that length. M4A, Ogg, WebM and FLAC files cannot be cut, so they are sent in one piece and need to fit in a single request of about 24 MB. If a file is too large in one of those formats, an MP3 or WAV copy of the same audio will go through.

Short sources come back ready with the text. Long audio is transcribed in the background: you get a transcript id, and your assistant checks it every 20 seconds or so until it is ready. Long episodes can take several minutes.

By default the transcript returns up to 60,000 characters of text, and you can ask for up to 400,000. The result also carries the title, the source link, the duration where it was measured, and the language when it was detected.

Here is what actually comes back. On October 18 I tested the direct audio path with the LibriVox reading of the Gettysburg Address, a public-domain recording, by passing the link to its MP3 file. The audio is 158 seconds long, and the transcript came back as 1,660 characters of plain, punctuated text: the LibriVox credit, the title and date, then the speech itself, starting "Four score and seven years ago, our fathers brought forth on this continent a new nation, conceived in liberty, and dedicated to the proposition that all men are created equal."

There were no timestamps to strip and no line breaks every few words. Sentences, commas and capitals were already in place, so the text could go straight into a draft. I chose a public-domain recording on purpose, for the reason in the last section of this page: only transcribe what you have the right to reuse.

Once the transcript is in the conversation, ask your assistant to write the article. It reads your voice profile first, writes from what the transcript actually says, and saves the result as a draft in your Narrareach Drafts. Nothing is scheduled or published at this point.

A good brief helps. Tell it which section of the talk to build around, who the newsletter is for, and what to leave out. A transcript is spoken language, so the draft should restructure it into an argument rather than tidy it line by line.

Open the draft, add the parts only you can write, then schedule it to Substack when it is ready. This is the same AI content repurposing workflow Narrareach uses for articles and Notes, starting from audio instead of text.

The same transcript can feed the Notes generator. Your assistant passes the transcript as the source text with the episode title, and Narrareach writes up to 10 Notes per request in your voice, each with a suggested posting time. Generated Notes can target Substack, LinkedIn, X, Bluesky, Threads and Facebook, and a batch can be written in English, Brazilian Portuguese, Spanish, French, German or Italian.

Generation saves and schedules nothing. You read the batch, keep the ones that sound right, and only then does your assistant add the approved ones to your queue. If you already repurpose a long piece as Substack Notes, this is the same step with a transcript as the source.

Notes generated through an assistant have their own allowance of 20 requests a day. Approved Notes go into the posting slots you set when scheduling Substack Notes in Narrareach.

Transcripts are part of Revenue Writer and above. Revenue Writer includes 100 YouTube transcripts and 120 audio minutes a day. Full Agentic Mode includes 300 YouTube transcripts and 600 audio minutes a day, and the 7-day trial uses the Full Agentic Mode allowance. Allowances reset at midnight UTC.

A public video or episode that someone has already transcribed comes back from the cache and does not count against your allowance. Once your audio minutes for the day are used up, podcast feed and Apple Podcasts links are refused until the reset, while cached YouTube videos, direct audio links and your own uploads still work. In a team workspace, transcripts use your own plan and allowance, not the workspace owner's.

For automation, the REST endpoint POST /api/v1/transcripts and the Get Transcript step in the n8n node need Full Agentic Mode and an automation token with the research:write permission. Tokens with research:read can only check on a transcript that is already running. That sits beside the content scheduling API you would use to publish the result.

Narrareach does not check who owns a video or episode. Transcribe your own recordings, guest appearances where you agreed the material could be reused, or content whose creator gave you permission. Quoting a short passage with credit is different from republishing someone else's talk as your article.

The assistant is told to quote only what the transcript says. If a line in the draft is not in the transcript, cut it or rewrite it as your own view.