How to Summarize Long YouTube Videos With AI

Feed a ninety-minute podcast to a one-click summarizer and you will usually get back a confident, well-written summary of its first fifteen minutes — with nothing marking where the rest went. That silent truncation is the central problem with a YouTube video summary with AI on long content. The fix is unglamorous and reliable: split the transcript into sections, summarize each one, then summarize those summaries. It forces the model to read every part of the video instead of skimming what fits.

Why one-click tools fail on long videos

Short clips are easy; a 90-minute podcast is a different problem. Link-based summarizers hit three walls:

• Context limits. Long transcripts run tens of thousands of words. Tools that grab everything and dump it in often overflow the model, so text gets cut.
• Silent truncation. Many services quietly summarize only the opening minutes. You get a confident summary of the opening, written as though it covered the whole thing.
• Lost detail. Even when the whole thing fits, models tend to skim the middle. The effect is documented: Liu et al.'s "Lost in the Middle" found model accuracy is highest when the relevant passage sits at the very start or very end of a long input and sags in between — which is exactly the shape of a one-shot summary of a two-hour podcast.

The result is a summary that looks complete but isn't. For long content, you need a method that forces the model to read every section.

Get and split the transcript

Start by pulling the full transcript with the free YouTube Transcript extension. It loads the entire transcript on the watch page, even for multi-hour videos, and lets you copy it as clean text. Once you have it:

1. Break the text into sections of roughly 2,000 to 4,000 words.
2. Split at natural boundaries where possible — chapter markers, topic changes, or a new speaker in an interview.
3. Number the sections so you can reassemble them in order later.

Natural breaks beat arbitrary word counts because each chunk stays coherent, which produces a cleaner section summary. If pulling the transcript is new to you, see how to get a YouTube video transcript.

Summarize each section

Run every section through your AI tool with a consistent prompt:

• "This is section 3 of a YouTube transcript. Summarize it in 4 to 6 bullets. Only use what's in this text. Keep numbers and names."

Keep the format identical across sections so the pieces combine cleanly. Bullets work better than prose here because they compress well and are easy to merge. Save each section summary as you go. Claude and Gemini handle bigger chunks per pass; ChatGPT works fine with smaller sections. We compare the models in AI summarize a YouTube video.

Summarize the summaries

Now combine. Paste all the section summaries into one final prompt:

• "Here are ordered section summaries of one long YouTube video. Combine them into a single summary with a 3-sentence TL;DR, then key takeaways grouped by theme. Don't add anything not present in the sections."

This two-pass structure is what makes long summaries accurate. The first pass guarantees every part of the video is read and condensed. The second pass finds the throughline across sections without ever dropping the middle. For format options once you're here, see creating a summary from the transcript.

One check before you trust the result

After the final pass, compare the summary against the transcript's last few hundred words. Truncation is the most common failure in long-video summarising and the hardest to notice, because a summary that covers three quarters of a talk reads exactly as convincingly as one that covers all of it. If the ending is represented, the chunking held.

Where to cut the transcript

Chunk boundaries matter more than chunk size. A split in the middle of an argument gives the model half a reasoning chain, and it will summarise that half confidently, stating a conclusion without the evidence or evidence without the conclusion.

Cut at topic changes instead. In practice these are easy to find: a speaker announcing what comes next, a chapter marker if the creator added them, or a question in an interview. If you have to split an unbroken stretch, overlap the chunks by a paragraph so the reasoning appears whole in at least one of them.

What a two-pass summary loses

It is worth being clear about the trade-off rather than pretending there isn't one. Summarising summaries compresses twice, and the second pass only ever sees the first pass's output. A detail that seemed minor within its own section, such as an aside that only matters because of something said forty minutes later, is gone before the final pass can connect them.

Two things reduce this. Ask the section summaries to preserve names, numbers, and direct quotes verbatim, so the specifics survive compression. And keep the full transcript rather than discarding it once the summary exists, because the moment you need a connection the summary missed, searching the original takes seconds.

Time saved and accuracy gained

The method sounds like more work than clicking a button, but the math favors it:

• Time saved: a 90-minute podcast takes 90 minutes to watch. The split-and-summarize pass takes 10 to 15 minutes and gives you searchable notes.
• Accuracy: because each section is read in full, nothing gets skipped. You can spot-check any claim against the transcript and jump to the exact moment with a click.
• Reusability: the section summaries are useful on their own — chapter notes, timestamps, and pull quotes fall out for free.

Once the long summary exists, the transcript stays useful. Search inside the video to find a specific claim, or repurpose the content into a blog post. For the complete framework, start with the pillar guide to summarizing YouTube videos.