The cheapest free AI summary generator for YouTube is not a product. Copy the transcript out of YouTube's own panel, paste it into any free LLM chat, and ask for timestamped key points — that route has no daily cap, no upsell, and usually beats the one-box summarizers, because you control the prompt. Reach for Gemini Notebook when you need claims you can click back to the source, and always spot-check three bullets at their timestamps before you quote anything.
TL;DR
A free AI summary generator turns a YouTube link into a set of key points in about two minutes. The route that stays free forever is the one nobody sells you.
- Costs nothing, no ceiling. Copy YouTube’s transcript panel into a free LLM chat and prompt it yourself. Pasted text is the unmetered part of almost every free plan.
- Best free tool. Gemini Notebook, when the summary has to be checkable rather than just short.
- Fastest. A browser extension that summarizes from the watch page without a tab switch.
- Timestamps are opt-in. Models strip them unless you demand them in the prompt and leave them in the input.
- Long videos. Chunk by the video’s own chapters. Frame-reading models top out near 3 hours of video on a 1M-token context.
- Verify three bullets. The best model on Vectara’s summarization leaderboard still hallucinates 1.8% of the time.
What a free AI summary generator actually does with a YouTube link
It reads the caption track. That is the whole mechanism behind nearly every free YouTube summarizer, and knowing it tells you exactly when to trust one.
Paste a URL, and the tool fetches the video’s captions, hands that text to a language model, and prints what comes back. No frames, no audio, no picture.
Google’s Gemini Notebook documentation is blunt about it for YouTube sources: “Only the text transcript of the video is imported as a source.”
Two consequences follow, and both change how you use these tools.
First, the quality of your summary is capped by the quality of the captions, not by the model. A clear-audio conference talk summarizes beautifully. A noisy vlog with three people talking over each other does not.
Second, anything shown on screen and never spoken is missing before the model starts. Numbers on a slide, code in an editor, the price in a comparison table — none of it reaches the summarizer.
If you need the picture read rather than the speech, that is a different class of tool, and we cover how it works in our explainer on AI video readers.
Method 1: the transcript panel plus a free chat, which costs nothing
This is the honest answer, and it is the one the tool listicles skip because there is nothing to link.
YouTube already transcribed the video. Open the transcript, copy it, paste it into whatever free AI chat you already have open, and write your own prompt.
Here is the whole flow.
- On desktop, expand the video description and click Show transcript. A timestamped panel opens on the right.
- Leave timestamps on. You will want them in the output, and the model cannot produce what it never received.
- Click into the panel, select all, copy.
- Paste into a free chat with the prompt from the timestamps section below.
Four steps, about ninety seconds, zero dollars.
Why this beats most paid summarizers comes down to two things. You control the prompt, so you can ask for a decision memo instead of a generic paragraph. And free chat plans meter uploads and images far harder than they meter pasted text, so a transcript is the cheapest thing you can give a model.
The trade is that you do it manually, one video at a time. If you are summarizing forty videos a week, buy a tool. If you are summarizing four, do not.
One caveat worth naming. YouTube’s own help page warns that automatic captions “might misrepresent the spoken content due to mispronunciations, accents, dialects, or background noise” — so proper nouns and product names in the summary need checking against the video.
If the transcript itself is the deliverable rather than the input, the caption-export and local-Whisper routes are laid out in how to transcribe a YouTube video with AI.
Method 2: Gemini Notebook, for a summary you can check
Use this one when someone will act on the summary and you need to defend it.
Gemini Notebook takes a public YouTube URL as a source and grounds every answer in that source, showing you which passage a claim came from. That citation link is the feature, not the summary itself.
The free limits are real and published. Free accounts get up to 50 sources per notebook, a YouTube video of any length is accepted unless its caption file runs past 500,000 words, and imports fail if the video has no captions, no speech, or was uploaded less than 72 hours ago.
Practical use: add the video, then ask for a briefing rather than a summary. “What does this video claim about X, and what evidence is given for each claim?” returns something you can audit line by line.
There is a second free Google route worth knowing if you are comfortable with a slightly more technical interface. Google AI Studio and the Gemini API accept a YouTube URL inline with your prompt, and Google’s video understanding documentation confirms the free tier allows up to 8 hours of YouTube video per day, with up to 10 videos in a single request on Gemini 2.5 and later. Only public videos work.
That 10-videos-per-request allowance is the underrated part. It turns “summarize this video” into “compare what these six videos say about the same topic”, which no one-box summarizer offers for free.
Method 3: browser extensions, when you want it without leaving the page
Extensions win on friction. You are already on the watch page, and the summary appears beside the player without a copy-paste round trip.
The category leader by installs is YouTube Summary with ChatGPT & Claude, which lists 1,000,000 users and a 4.3-star rating from 4,000 ratings on the Chrome Web Store as of September 2026. It pulls the transcript, offers summaries with or without timestamps, and lets you pick which model does the work.
Understand what most of these extensions are, though. They are a transcript scraper plus a pre-written prompt, sent to a chat you are already signed into.
That is genuinely useful — it automates Method 1. It also means the summary quality is the model’s, not the extension’s, and the extension’s canned prompt is usually worse than one you would write yourself.
Two things to check before installing any of them. What permissions it asks for, and whether it sends your transcript through the developer’s server or straight to the model provider. A summarizer that only needs the current tab does not need read access to every site you visit.
Method 4: dedicated summarizers, and where “free” actually stops
One-box tools are the fastest thing to try and the easiest to get burned by. Paste a URL, click a button, read the summary.
The pattern to watch for: the landing page says free and no login, while the pricing page tells a different story. NoteGPT markets a free no-login summarizer and batch processing, but its pricing page (checked September 2026) starts at a $9/month Pro plan with 1,000 basic quotas, 100 premium credits, 5 YouTube videos processed at once, and a 120-minute ceiling on videos without subtitles.
Mindgrasp’s YouTube summarizer page advertises “100% Free with No Sign-Up” for unlimited videos of any length and publishes no cap at all — which is a different problem, because an unpublished limit is one you discover by hitting it mid-task.
Neither is a reason to avoid these tools. It is a reason to run one test video before you build a workflow on top of one.
Three questions answer whether a given free summarizer is worth your time.
- Does it publish a numeric limit? A stated cap beats a generous-sounding claim.
- Does it output timestamps? Without them the summary is unverifiable.
- Does it choke on your longest typical video? Test with your worst case, not a five-minute clip.
If you want these ranked against each other on output quality rather than described in passing, that is our comparison of AI video readers and summarizers. For summarizing things that are not video at all, the general ranking sits in free AI summary generators and the document-specific one in free AI PDF summary generators.
The prompt that forces timestamped key points
Timestamps are the single biggest upgrade to a YouTube summary, and almost every tool drops them by default. Summarizing means compressing, and the timestamp is the first thing compression throws away.
You get them back by demanding them and by making sure they were in the input. Copy the transcript with timestamps visible, then use a prompt with all four of these parts.
Below is the timestamped transcript of a YouTube video.
Write 8-12 key points in the order they occur.
Rules:
- Start every bullet with the timestamp of the transcript line it
came from, in [MM:SS] form. Never invent or estimate a timestamp.
- One idea per bullet. No preamble, no conclusion paragraph.
- If the speaker states a number, a name, or a specific
recommendation, keep it exactly as said.
- If something is unclear in the transcript, write
[unclear] rather than guessing.
Transcript:
<paste here>Each rule is load-bearing. “Never invent a timestamp” blocks the most common failure, where a model produces plausible-looking times that point nowhere.
“Keep numbers exactly as said” stops the quiet rounding that turns 37% into “roughly a third”. And the [unclear] instruction gives the model a legal way to admit a gap instead of filling it.
Change the output shape by swapping only the second line. Ask for “a chapter list, one line each” for navigation, “the three decisions the speaker recommends, in order” for a how-to, or “every claim and the evidence given for it, as a two-column table” when you are evaluating an argument.
Reading a video and questioning it are related but separate jobs — if you want Q&A and full-text reading rather than a summary artifact, that workflow is in how to read any YouTube video with AI.
How to summarize a video longer than an hour
Length breaks free summarizers in two different ways, and which one hits you depends on what the tool reads.
Transcript-based tools scale well. Text is cheap. A 60-minute talk at normal speaking pace is roughly 9,000 words, which any modern model swallows without effort.
Tools that actually watch the video hit a wall sooner. Google’s documentation samples video at 1 frame per second by default and costs it at roughly 100 tokens per second of video at low media resolution. On a 1M-token context that lands at about 3 hours at low resolution, or roughly 1 hour at high resolution.
So the first question is which kind of tool you are holding. If it takes a pasted transcript, paste the whole thing and stop worrying.
If it watches, chunk it. Three rules make chunking work.
- Split on the video’s own chapters, not on equal time slices. Chapter boundaries are topic boundaries, so each chunk stays coherent.
- Use the same prompt on every chunk. Different prompts produce different bullet styles, and stitching them reads like three people wrote it.
- Summarize the summaries last. Feed the chunk outputs back in and ask for a single 10-bullet version. Keep the chunk-level detail as an appendix.
Where the tool supports it, clipping is better than chunking. Google’s API takes start_offset and end_offset so the model reads a defined window instead of the whole file — useful when you only need the Q&A at the end of a two-hour stream.
One more trick for pasted transcripts: strip the timestamps from everything except every fifth line before pasting a very long one. You keep enough anchors to verify against, and you cut the input size meaningfully.
When the video has no captions
No captions, no summary. That is the failure mode behind most “this tool didn’t work” complaints, and the error messages rarely explain it.
Gemini Notebook names three separate causes: the video has no caption file, the video contains no speech at all, or it was uploaded less than 72 hours ago and has not finished processing.
Each has a different fix.
Fresh upload. Wait. Auto-captions usually land within a day, and this resolves more failures than people expect.
Audio exists, captions do not. Run speech-to-text yourself and summarize the transcript you get back. That path, including a local pipeline, is in our YouTube transcription guide.
No speech at all. A silent screen recording, a timelapse, a music-led montage — no transcription helps, because there is nothing to transcribe. You need a tool that samples frames, and the ones with real free tiers are listed in free AI video reader tools.
How to check a summary before you trust it
Free summaries are cheap enough to be worth generating and cheap enough to be wrong. Both are true at once.
Vectara’s Hallucination Leaderboard measures exactly this task — models summarizing documents, scored by Vectara’s HHEM-2.3 evaluator across a set of more than 7,700 articles. In its 11 May 2026 update the best model, Antgroup’s Finix S1 32B, hallucinated on 1.8% of summaries, with OpenAI’s GPT-5.4-Nano at 3.1% and Google’s Gemini-2.5-Flash-Lite at 3.3%.
Those are the leaders. Whatever model sits behind a free one-box summarizer is unnamed, probably cheaper, and probably worse.
The check takes ninety seconds and catches nearly everything.
- Spot-check three timestamps. Pick the two most surprising bullets and one number, click through to those moments in the video, and confirm the speaker said it. Fabricated timestamps announce themselves instantly.
- Re-read every number. Figures are where compression does the most damage. If a bullet says “cut costs by half” and the speaker said “cut costs by 40 to 60 percent depending on volume”, the summary is technically wrong in the direction that matters.
- Ask the model what it left out. “What did you omit from this transcript that a careful reader would want?” often surfaces a caveat or a counterargument the summary flattened.
- Never quote from a summary. Quote from the transcript at the timestamp. The summary is an index, not a source.
Run the same summary through a second free tool if the stakes are high. Two independent models agreeing on a claim is weak evidence, but two disagreeing is a very strong signal to go check.
Which free method to use, by job
| Your job | Use | Why |
|---|---|---|
| One video, want it right | Transcript panel + free chat | You write the prompt, so you get the shape you need |
| Summary someone else will act on | Gemini Notebook | Grounded answers you can click back to the source |
| Skimming ten videos before choosing one | Gemini API free tier | Up to 10 videos per request, 8 hours per day free |
| While you are already watching | Browser extension | No tab switch, transcript scraped automatically |
| Occasional, no setup at all | One-box summarizer | Fastest first result, unpublished limits |
| Video has no captions | Transcribe first, then summarize | Nothing to read otherwise |
The short version
Start with the transcript panel and a free chat. It costs nothing, has no daily cap, and beats most paid summarizers because you control the prompt.
Move to Gemini Notebook when the summary needs to be defensible, and to the Gemini free tier when you are triaging a stack of videos rather than reading one.
Always leave timestamps in the input and demand them in the output. A bullet without a timestamp cannot be checked, and an unverifiable summary is a rumour with better formatting.
Then spend ninety seconds on three spot-checks. That habit is the difference between a free AI summary generator being a real tool and being a confident guess.