Coverage
The last chapter should start in the final fifth of the video. That is the rule this pipeline enforces; use it on any tool.
Guide · 2–5 hour courses, lectures and podcasts
Long videos are where a summary saves the most time, and where most summaries quietly fail: the first hour is covered in detail and the rest is a sentence. This is the method we use, the checks that tell you whether any summary covered the whole video, and a real four-hour example.
It has to be public. Members-only, private and age-restricted videos have no transcript anyone outside can fetch. A video without captions isn't a dead end: the audio can be transcribed instead, it just takes longer.
The failure that matters on a long video is a summary built from the first hour. Here the full transcript goes to the model in one request, prefixed with the video's total length, and the instructions require chapters spread across all of it. No splitting into chunks that get summarized separately and stitched back together.
Look at the last chapter's timestamp and compare it with the video's length. If a 4-hour video's chapters stop at 1:10:00, the summary skipped most of it, whatever it says.
Each key point carries the timestamp where the claim was made. Read the ten lines, and open the video at the two or three moments you actually need.
A summary compresses four hours into a few hundred words. For an exact figure, name or quote, search the full transcript on the same page rather than relying on the paraphrase.
freeCodeCamp.org’s Python course for beginners runs 4h 26m. Its transcript was about 82,000 tokens, and the summary below came back in 29 seconds. It is unedited output — the same example you can open on the homepage.
TL;DR: A comprehensive beginner-friendly Python course covering everything from installation to classes and inheritance.
The chapters keep a steady pace through the last hour — classes, inheritance and the interpreter — instead of stopping at the beginner material. That is the coverage check passing.
The last chapter should start in the final fifth of the video. That is the rule this pipeline enforces; use it on any tool.
Key-point timestamps should come from across the video, not cluster in the opening half hour.
“Covers dictionaries” is a topic label. “Dictionary keys must be unique and immutable” is a key point you can act on.
Click two chapter links. Each should land on the moment described. Timestamps copied from the transcript do; invented ones drift.
No, on any plan. Long videos take longer to process — the 4-hour example below took about half a minute — but they are not cut off or summarized from a sample.
Yes: 3 new summaries every rolling 30 days with no account, and a long video counts the same as a short one. A video someone has already summarized costs nothing.
You can, if you fetch the transcript yourself and it fits. A four-hour transcript is tens of thousands of words, and the result has no clickable timestamps unless you ask for them and check them. The method above still applies: check the last chapter against the video's length.
That is a different job — one summary per video plus a report across the series. See the playlist summarizer.