About Summarizer.tube

What we make

Summarizer.tube takes a YouTube video URL, extracts the video's caption track (manual subtitles when available, auto-generated otherwise), feeds the transcript to a large language model, and returns two things: a paragraph-length summary of the main argument, and a bulleted list of key takeaways — each tagged with a clickable timestamp that seeks directly to that moment in the YouTube player.

You can use the website at summarizer.tube (paste a URL into the input), or install our free Chrome extension to summarize directly on the YouTube watch page. Everything else — chat with the video, bookmarks, history — is built around those two core outputs.

Why we built it

YouTube hosts roughly 800 million videos and the average viewer watches about 48 minutes of content per day. Most of that time is not well spent. Long tutorials bury key insights in filler. Lectures repeat for emphasis. Conference talks meander before they reach a conclusion. We wanted a tool that strips a video down to its argument in 30 seconds — without forcing a sign-up, without harvesting browsing data, without aggressive upsells.

Existing tools either lock the basic feature behind a paywall, require a Chrome extension you cannot avoid, or wrap the AI in a UI that feels designed to convert rather than to inform. We built Summarizer.tube to be the opposite: a free tool that does one thing well, an extension that's optional, and a public summary archive at /summary that anyone can read without an account.

How it actually works

1. Transcript extraction. When you submit a URL, our backend asks YouTube for the video's caption track. Manual captions take priority; if none exist, we use YouTube's auto-generated captions. Videos with no captions at all cannot be summarized — the AI works from text only.
2. Language detection. The caption track's declared language code is used. If the video has multiple tracks, we pick the primary one. This determines what language your summary comes back in.
3. Summarization. The transcript is sent to a large language model with a structured prompt that asks for a paragraph summary and a list of key points with timestamps. We rotate across multiple AI providers for reliability and cost.
4. Timestamp anchoring. Each key point includes the timestamp of the relevant moment in the source video. The model uses the transcript's built-in time markers to anchor its claims. The clickable timestamp seeks the YouTube player directly to that point.
5. Public archive. Successful summaries are stored in our public archive at /summary so others can find them via search and so AI assistants can cite them. If you don't want a video in the public archive, contact us — we remove it.

Where it falls short

We list this honestly because every other summarizer pretends to be perfect and that's why people stop trusting them.

• Visual content is invisible to the AI. We work from text only. Diagrams, code shown on screen, gestures, body language, b-roll, charts — none of these reach the summarization step. For demonstration-heavy videos this is a real gap.
• Auto-generated captions degrade on technical content and accents. Specialized terminology gets misrecognized; the summary inherits those errors.
• Very long videos may be truncated. Models have context limits. Videos past about 5 hours get their transcript trimmed to fit. The summary still covers the available content but may underweight the tail end.
• The AI can hallucinate, especially on niche topics. We do not currently fact-check. If a video contains a wrong claim, the summary may repeat it confidently. Treat summaries as a triage tool, not a citation source for important decisions.
• No video without captions. If a creator disables captions and YouTube auto-captions also fail, we cannot summarize the video. This is rare but real.

What we keep about you

• Your local history stays in your browser. We do not have it. You can export or delete it from the profile page if you sign in.
• Server-side bookmarks exist only if you sign in and explicitly bookmark a summary. Otherwise we don't store anything per-user.
• The summaries themselves are stored in the public archive — the videoId, the title, the summary, and the key points. Anyone can find them at /summary.
• Daily quota counting uses an IP-based fingerprint with a 24-hour sliding window. We don't link this to any identity unless you sign in.
• Payments via Paddle for Pro subscribers. Paddle handles card details — we never see them. We see only your subscription status.

Full details: privacy policy.

The team

Summarizer.tube is built by a small indie team. We are not VC-funded, not affiliated with YouTube or Google, and not part of a larger media company. Every feature shipped exists because we built it ourselves — if it breaks, we fix it.

Want to know who wrote a specific blog post or feature? Email us — we read every message.