How to build an AI chapter and show notes generator for YouTube

The problem it solves

Creators and podcasters publish weekly and skip the metadata because it's tedious. That costs them search traffic, suggested-video placement and viewer retention. Editors and agencies doing it manually spend 20 to 40 minutes per video. A generator that produces accurate, timestamped drafts removes most of that work.

What to include in your first version

• ■Timestamped chapters in YouTube's required format, starting at 0:00
• ■An SEO description with target keywords and links
• ■Podcast-style show notes with key points and resources mentioned
• ■Brand voice settings per channel
• ■One-click copy or publishing to YouTube, WordPress or a podcast host

How to build it, step by step

1. 01

   Fetch the transcript

   Call /transcribe with format.timestamp and format.paragraphs so you have both readable text and start times.

2. 02

   Detect topic changes

   Send the paragraphs with their timestamps to an LLM and ask it to group them into five to ten chapters, using only start times that exist in the data.

3. 03

   Write the description and notes

   Use a second prompt to write the description and show notes from the chapter list, including the creator's keywords and links.

4. 04

   Review and publish

   Show the drafts in an editor, let the creator adjust them, then copy or publish through the platform's API.

Starter code

This idea is built on the "Recipe 1: One video in, structured output out" pattern. Swap in your own channels, prompts and storage.

import requests

API = "https://www.youtubetranscript.dev/api/v2"
HEADERS = {"Authorization": "Bearer yt_sk_live_YOUR_KEY"}

res = requests.post(f"{API}/transcribe", headers=HEADERS, json={
    "video": "https://www.youtube.com/watch?v=VIDEO_ID",
    "format": {"timestamp": True, "paragraphs": True},
}).json()

transcript = res["data"]["transcript"]
print(transcript["text"][:500])        # plain text for your LLM prompt
for seg in transcript["segments"][:3]:  # timestamps for deep links
    print(seg["start"], seg["text"])

Why it's worth building

Charge creators a monthly subscription based on videos per month, or add it as a paid feature inside an existing editing, scheduling or podcast hosting tool. Agencies will pay for team seats and brand voice presets.

Mistakes to avoid

• ✕Letting the model invent timestamps: constrain it to start times from the transcript segments.
• ✕Producing generic descriptions: feed in the channel's keywords and past descriptions as examples.
• ✕Forgetting YouTube's rules: the first chapter must start at 0:00 and chapters need at least three entries.

Questions about this idea