Free · runs in your browser · works with ChatGPT and Claude
Your YouTube feed is not on your side.
See the channels eating your week and the clickbait aimed at you — then hand the whole picture to ChatGPT or Claude in one paste.
- Free, no account
- Runs in your browser
- Nothing uploaded
Your feed, decoded
298 videos · 75 days
Top channel: Nihongo Daily · 37 videos
Try it
What is this title doing?
- Promises a hidden truth
We read the title, never the video. A great video can carry a bait-styled title — that is the whole point.
Free extension
The same scorer, on your real feed
Free, no account, and it makes no network requests of its own.
Marks the bait
and names the technique, live on your feed
Hides the rest
Shorts, home feed, sidebar, end screens
Blocks channels
suggested from the ones you never finish
Spots rabbit holes
four videos in a row it chose, not you
What you walk away with
A prompt your AI can actually work with
Paste this into ChatGPT or Claude and it already knows your channels, your hours, your topics — and, just as importantly, what your export can’t tell it. No explaining yourself for twenty minutes first.
I'm sharing a summary of my own YouTube watch history, prepared by a tool that analysed it on my device. Please help me understand it. --- MY YOUTUBE DATA --- # My YouTube watch history 298 videos watched between 2026-04-12 and 2026-06-26 (75 days). Source: Google Takeout export (includes timestamps, but no watch durations) ## What this data can and cannot say - Topic labels were assigned by keyword matching on titles and channel names, not by watching anything. 77% of videos could be classified; the rest are genuinely unknown, not "other". Classification reads English only, so a non-English feed is under-counted rather than absent. - This export records WHICH videos were watched, not for how long. There is no watch-time or completion data at all. Do not estimate hours from video counts. - There is no per-video completion data, so nothing here says what was finished or abandoned. - This is the most recent 90 days, capped at 1000 videos. It is a sample of viewing, not a complete record, and it excludes anything watched while history was paused or in a different account. ## Overview - Videos: 298 - Distinct channels: 15 ## Topics - Language Learning (learning): 16% of videos, 37 videos - Music (entertainment): 16% of videos, 36 videos - Tech & Gadgets (mixed): 14% of videos, 31 videos - Gaming (entertainment): 13% of videos, 30 videos - Software & Programming (learning): 9% of videos, 20 videos - Fitness & Health (learning): 8% of videos, 19 videos - Food & Cooking (mixed): 7% of videos, 17 videos - Vlogs & Lifestyle (entertainment): 4% of videos, 9 videos - AI & Machine Learning (mixed): 3% of videos, 8 videos - Science & Engineering (learning): 3% of videos, 8 videos Of the classified portion: 38% sits in learning/professional categories, 36% in entertainment categories, 26% in categories that are genuinely either (news, food, travel, tech, cars). ## Channels - Nihongo Daily: 37 videos - Lantern Lo-Fi Music: 36 videos - Gadget Bench Reviews: 31 videos - Save Point Docs: 30 videos - Mega Challenge Co: 23 videos - Code in a Minute: 20 videos - Marcus: 20 videos - Deep Talk Clips: 18 videos - Small Kitchen Club: 17 videos - Studio B: 15 videos The feed leaned on 5 channels for 53% of your views, out of 15 it put in front of you. Nothing new broke into the rotation in the last 30 days. ## When - Busiest time of day: 7pm–10pm - Busiest day: Sunday - Hour-of-day distribution (0–23, weighted by videos): [7, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 63, 68, 68, 45, 46] - Weekday distribution (Sun–Sat): [93, 25, 26, 23, 23, 17, 91] ## Sittings - Sittings: 73 across 62 days with any viewing (1.2 per active day) - Typical sitting: 4.1 videos, median 29 minutes of elapsed time - Longest by count: 15 videos in one sitting - Longest by time: 123 minutes - Sittings starting between 10pm and 5am: 11 - A "sitting" is videos separated by gaps under 30 minutes. Elapsed time is measured from the first video's start to the last video's start, so it understates a sitting by the length of its final video and is 0 for a one-video sitting. ## Runs on one topic - Music: 8 videos over 92 minutes (88% on topic) e.g. "Jazz hop mix for a rainy afternoon"; "Lo-fi beats to study to"; "Full album stream, live at the studio" - Language Learning: 8 videos over 74 minutes (88% on topic) e.g. "Japanese vocabulary: at the konbini"; "Japanese listening practice for beginners"; "Shadowing practice, slow then natural" - Software & Programming: 8 videos over 93 minutes (75% on topic) e.g. "Python in 100 Seconds"; "The Reset"; "Python in 100 Seconds" - Music: 6 videos over 61 minutes (100% on topic) e.g. "Full album stream, live at the studio"; "Official music video - Night Bus"; "Full album stream, live at the studio" ## Recent vs earlier Comparing the older half of the window (149 videos) with the newer half (149 videos), as a share of classified videos: - Music: 10% → 21% (+11 points) - Tech & Gadgets: 17% → 10% (-7 points) - Food & Cooking: 11% → 4% (-7 points) - Language Learning: 19% → 13% (-6 points) ## Repeats - "Lo-fi beats to study to" (Lantern Lo-Fi Music) — 12 times - "Unboxing the new Steam Deck" (Gadget Bench Reviews) — 11 times - "Official music video - Night Bus" (Lantern Lo-Fi Music) — 10 times - "Japanese vocabulary: at the konbini" (Nihongo Daily) — 9 times - "Part 6" (Marcus) — 9 times ## Title style - 2% of titles (6 of 298) use attention-grabbing constructions: withheld payoffs, superlatives, shouted emphasis, stacked punctuation, thumbnail emoji. - Examples: "You Won't Believe What Happened Next" - This measures title STYLE, not video quality. A rigorous documentary called "The Truth About Sugar" scores as bait-styled. It reads English only. --- END OF MY YOUTUBE DATA --- ## What I want Identify where my viewing time is actually concentrated. Rank the biggest consumers by topic and by channel, using watch time if the data has it and video counts if it doesn't (say which you used). For each one, note whether it looks like deliberate viewing or drift, and what in the data tells you that. Separate 'large' from 'not worth it' — they are not the same thing, and a hobby I love being my biggest category is not a problem. ## How to answer - Use only the figures given above. Do not estimate, extrapolate, or invent numbers, and do not fill in the gaps described under 'What this data can and cannot say'. - Where the data can't answer something, say so plainly and say what would be needed, rather than guessing. - Be specific: name the actual channels, topics and figures from the data instead of speaking generally. - Describe what the recommendation feed served and what got watched. Don't moralise about my discipline or willpower, and don't assume watching less is automatically the goal. - Do not make psychological, medical or diagnostic claims. No assessments of addiction, attention disorders, dopamine, or mental health. - Topic labels came from keyword matching and are imperfect. Treat a single surprising label as possible noise; treat a consistent pattern as signal.
Built from 298 videos across 15 channels over 75 days · worked out in your browser, in about a second
Yours names your channels, not these. Free, no account — and like this sample, it will tell your AI which 23% of your titles it couldn’t read.
How it works
Four steps to a better question
Add the extension
The free extension reads your watch history in one click — including how much of each video you finished, which Google’s own export leaves out. It stays on afterwards to mark the bait and hide what you never asked for. No extension? Google Takeout works too.
It works itself out in your browser
Topics, channels, sittings, the runs you fall into, the hours you watch. All of it is keyword matching and arithmetic on your own machine — which is why there is no server to send it to and no account to make.
Look at what it found
Your dashboard: what the feed served, which handful of channels own most of your week, when you actually watch, and how much of it we could read at all.
Hand it to your own AI
Pick a question — where the time goes, what you're really learning, a 30-day plan — and copy the prompt. It carries your figures and the limits of your data, so ChatGPT or Claude answers from evidence instead of guessing. We never see that conversation.
On a phone, or not on Chrome? Start with Google Takeout — ask for history only and the .zip usually lands in minutes. Drop it in as it arrives.
The two routes see different things. Takeout knows when you watch, down to the hour. The extension knows how much of each video you finished. Neither is worse — your dashboard says plainly which questions your file can answer, and your prompt tells your AI the same thing so it doesn’t invent the rest.
What we never get
We can’t lose what we don’t have
You never log in
There is no “connect your account” button anywhere. We never ask for your Google password, and we never touch your account.
There is no account here either
No sign-up, no profile, no password to forget. Nothing to log into, so nothing to be locked out of or leaked from.
It's free to use
No card field, no subscription, and nothing sold on the side. Your history is analysed on your own machine, not on ours.
So there is no database of what anyone watched.
Your history is read in your own browser and never leaves it. There is no upload step, no database, and no copy of it here — we could not hand it over, sell it or lose it, because we never receive it. It stays in this browser so your dashboard survives a refresh, and one button forgets it.
Personalized
You decide what “a good feed” means
Two questions before we read your history. Entirely skippable.
You tell us
“I use YouTube for…”
- Learning something specific
- Unwinding after work
- Background noise
- + 5 more
“I’m working toward…”
- Learning a skill or language
- A career move or new job
- Getting my attention back
- + 5 more
Plus one optional line of your own.
It answers
Your feed is half helping
Getting in the way
- Roughly a third of what the feed offers you midweek is productivity and reaction content you bail on inside a minute
Most stats pages grade your feed against nothing. Answer these and your AI gets told where you said you were going — so the question becomes whether the algorithm took you there, never whether you had the willpower.
What it finds
Your clicks and your taste are at war
One of these got finished. One got under a minute. Call it.
One of these got watched to the end. The other got under a minute. Call it.
1/3
The feed only listens to one of them.
Your dashboard names your own titles and channels — and your prompt hands them to ChatGPT or Claude, so you can ask what to do about it.
FAQ
Common questions
Nothing today — there's no account, no subscription and no card field. The analysis is arithmetic and keyword matching running in your own browser, so there's very little for us to charge for. The last step hands you a prompt to paste into ChatGPT or Claude, which is an assistant you already have. We may add paid features later; if we do, we'll say plainly what's paid and what isn't rather than quietly moving the line.
No — every number is calculated directly from your own file, not guessed. The video counts, channel rankings, sittings, hours and topic shares all come out of your history. Nothing is written by an AI at any point: we don't hand you a personality, we hand you your own figures and the prompt that lets your AI reason about them.
Usually none. Our free Chrome extension reads your history and hands it straight to the site — there’s no file to export or upload. It picks up the watch-time details (how much of each video you actually watched) that a plain Google export doesn’t include. A Google Takeout export works too, and we’ll walk you through either one on the upload screen.
Your watch history never leaves your browser. There is no upload step and no server that could read it: the extension hands your history to this site inside your own browser, the analysis runs on your device, and your dashboard is drawn from that. Your history is saved in this browser only, so the page still works when you come back, and one button on the dashboard forgets it. The one moment anything leaves is when you choose to paste the prompt into ChatGPT or Claude — that's you sending a message to them, under their privacy policy, and we're not part of it. You can see the exact text first, and turn off real video titles if you'd rather send only the numbers.
Yes, using Google Takeout. The extension needs a desktop Chromium browser (Chrome, Edge, Brave, Arc, Opera or Vivaldi), but Takeout works on any device: request the export on your phone, and when Google emails you the .zip, open it here exactly as it arrived — no unzipping, no digging for a file inside it. Tick history on its own and the export is usually ready in a few minutes; the long waits people warn about come from leaving the whole account selected. Neither route is the lesser one — they see different things. Takeout records when you watch, down to the hour, which the extension can't see. The extension records how much of each video you finished, which Takeout can't see. Your dashboard says plainly which questions your file can answer, and your prompt tells your AI the same thing so it doesn't fill in the gaps with guesses.
You don’t — Google Takeout works. The extension is just faster (one click, no waiting on an export) and it reads how much of each video you actually finished, which only appears on your YouTube history page and which Takeout can’t give you. It also does plenty that has nothing to do with the dashboard: it marks bait-styled titles on your feed and names the technique they use, hides Shorts, the home feed, recommendations, comments and end screens, blocks channels you never finish, warns you when four videos in a row were chosen for you rather than by you, and keeps a daily attention budget. The extension is free, and it makes no network requests of its own.
Only if you ask it to. Everything is off until you switch it on, apart from marking bait-styled titles. It can hide Shorts, the home feed, watch-page recommendations, comments and end screens, drain thumbnail colour, block channels, and replace the wall of recommendations at the end of a video with a choice. It will never close a tab, block a video, or stop you watching something — the prompts are sentences with a Dismiss on them. Two features remember anything at all (a weekly count and a daily duration), neither switches on before showing you what it keeps, and one button erases both.
It lives in this browser, so it's there when you come back — no account, nothing on our end to log into. It clears itself after 90 days, which is the window the analysis covers anyway, and there's a ‘Forget my history on this device’ button on the dashboard that wipes it straight away. Clearing your browser data removes it too. It's your copy: we have no copy to delete, because we never had one.
A dashboard and a prompt. The dashboard shows what the feed served you: your topics, the channels that own most of your week, when you watch, the runs where you fell into one subject, what you re-watched, and how much of your history we could read at all. The prompt is the part you take elsewhere — pick a question (where the time goes, what you're really learning, a 30-day plan, or type your own) and copy a block of text that carries your figures, your goals if you gave them, and the honest limits of your data. Paste it into ChatGPT or Claude and it can answer properly instead of guessing.
Because your own assistant is better placed to. You almost certainly already have ChatGPT or Claude, it already knows how you like to be talked to, and you can keep arguing with it — which a fixed, one-shot report never let you do. So we do the part that has to happen on your machine: reading your history, working out the numbers, and writing them up as context worth pasting. The hard part was never the answer, it was getting your actual data in front of the thing answering.
No — but the prompts get sharper if you do. After you upload, we ask two quick questions: what you feel you use YouTube for, and what you're working toward. You pick from a short list, and there's one optional line for anything in your own words. Skip it and everything still works. Answer it and two extra questions unlock — comparing what the feed delivered against what you said you wanted — and your goals ride along in the prompt so your AI knows what it's measuring against. The comparison is about the feed, not about your willpower.
The same thing that happens to your watch data: they stay in your browser. They're saved next to your history so your dashboard remembers them, they're deleted together with it, and they never reach a database, a log line or our usage analytics — we record only whether the questions were answered, never what was picked. The one thing to know: anything you type in the optional line becomes part of the prompt, so if you paste that prompt into ChatGPT or Claude, you're sending it to them. Keep anything genuinely sensitive out of it.
Your most recent 90 days, capped at your 1,000 most recent videos. That keeps it about what your feed is doing now rather than two years ago — and keeps the prompt short enough to paste.
Large files are fine, including a Takeout .zip that runs to tens of gigabytes because you left photos and Drive selected. We read the archive by byte range rather than loading it, pull out just the history file, and analyze your 1,000 most recent videos from the last 90 days — usually in a few seconds.
Your turn
See what it has been doing
Free, no account, nothing uploaded. Your history is read in this browser and stays there.