Why 'minutes watched' is the wrong number for learners

Open any watch-time tracker and it will tell you a number: minutes watched. It is the easiest number to produce and, for a language learner, close to the least useful.

The short answer. “Minutes watched” is the wrong number because it measures how long a video was open, and progress in a language depends on three things that number cannot see: whether the audio was in the language you are learning, whether the video was actually playing rather than paused, buffering or showing an ad, and whether you understood enough of it for it to count as comprehensible input. A tracker that reports minutes watched puts an English recap, an ad break and a forgotten tab into the same total as forty minutes of well-chosen listening. The number that matters to a learner is hours of input in the target language that actually played. It is smaller and harder to produce, but it is the one every immersion roadmap and every “how long does it take” estimate is talking about, and compared against those estimates a padded total makes a working method look like a failing one.

Three things it hides

Whether the audio was in your target language. A learner’s viewing is mixed. Recaps in their own language, a video with commentary in a third one, a tutorial they clicked out of curiosity. “Minutes watched” adds them all up as if every minute pulled in the same direction.

Whether the video was actually playing. An open tab isn’t listening. Ads aren’t input. A player that buffered for a minute didn’t deliver a minute. Wall-clock trackers count all of this, and the padding is invisible because it’s spread evenly across every day.

Whether you followed it. Forty minutes of native news you caught a third of, and forty minutes of learner content you followed nearly all of, are the same forty minutes to a tracker and nothing like the same thing to your brain. This one no tool can measure for you, but a tool that reports the other two honestly at least leaves you a clean number to attach your judgement to.

Where the padding comes from

It helps to be specific about what a wall-clock number contains, because each piece looks too small to matter.

Ads are the obvious one. A pre-roll and a mid-roll on every video add a few percent to every total, and because they land on every day equally, no day ever looks wrong. Pauses are the next. A learner pauses more than a casual viewer does, to look a word up, to replay a sentence, to type something into a dictionary, and a timer keeps running through every one of those. Skipping is a third: jump past a five-minute introduction and a tracker that credits the player’s new position has just given you five minutes of listening you did not do. Then there is autoplay, which runs on into the next video after you have left the room, and the tab left open overnight, which is the single largest error most people’s logs contain.

The mistake also runs the other way. Browsers slow down timers in tabs you cannot see, so a tracker built on a timer loses most of the ten minutes you listened while cooking. A wall-clock total is therefore wrong in both directions at once, and there is no way to tell from the number which way it is wrong today. A tracker that reads the player’s own position avoids both problems: a paused, buffering or advertising player does not move, and a hidden one does.

Video length is not listening time

There is a subtler version of the same error. Some logs record the length of the video rather than the time you spent on it. For a learner who changes playback speed, which most do, the two diverge immediately. An hour-long video at 2× is thirty minutes of listening, not sixty; a thirty-minute video slowed to 0.75× is forty. Content time flatters you when you speed up and short-changes you when you slow down, and the roadmaps are written in hours of your life, not hours of footage. The honest record is the elapsed time, with the playback rate noted beside it so the difference is explained rather than mysterious.

What a better number looks like

The number worth keeping is hours of input in the target language that actually played. It is smaller than minutes watched, sometimes a lot smaller, which is exactly why people don’t like producing it. But it’s the number every roadmap and every “how long does it take” estimate is talking about, and comparing a padded total to those estimates is how learners conclude the method has failed them when it hasn’t.

Getting there needs three things a tracker rarely does:

1. Count from the player’s own clock, so pauses, ads and a sleeping laptop don’t count, and a hidden tab does.
2. Listen to the audio and decide what language it’s in, and hold back time in a language you already speak instead of adding it in.
3. Show the held-back time next to the total, so the parts still add up and nothing looks like it vanished.

Recorded is not the same as counted

That last point matters more than it sounds. A tool that silently deletes your English minutes is indistinguishable from a bug. A tool that says “52 minutes of Japanese, and 14 minutes of English not counted” is telling you something true about your day. The distinction, recorded but not counted, is the whole design of Kiku Hours, and it’s why the panel never shows a number it can’t explain.

It also changes behaviour in a way a bare total never does. When the English recap you clicked on shows up in the day’s list with its fourteen minutes marked as not counted, you notice. Nobody is scolded and nothing is lost, but the next click tends to be different. A total that quietly absorbed those minutes would have taught you nothing.

Unlabelled is not counted

A tool that listens for the language will sometimes not be sure. A speaker under loud music, a video with almost no speech in it, a clip you glanced at for twenty seconds before moving on. The tempting design is to guess, because a guess produces a tidy label, and the second-most tempting is to count the time anyway and leave the label blank. Both pad the total. Kiku Hours does neither: a video’s time counts only once the video has a language, and a video that never gets one never counts. The minutes are not thrown away, because the label belongs to the video rather than to the moment: the first half minute of every video is unlabelled while the detector listens, and it joins the total the instant the label lands, as does a return visit to a video labelled last week. What stays out is the time nothing could be said about, which is exactly the time that should not be in a number called comprehensible input.

What to do with the number once you have it

An honest total is only useful if you do something with it, and the three uses are the same for every learner.

Compare it with the roadmaps. The hour counts people quote describe what listening feels like at 300, 600 and 1,000 hours, and they are calibrated on honest hours. If your total is honest and your ears match the description, the method is working and you can relax. If your total is padded, the comparison is meaningless, and it is the padding, not the method, that failed.

Set a goal you can meet on a bad day. Streaks are built on floors, not ceilings, and thirty minutes you actually listened to is worth more than an hour that was mostly a tab.

Look at months, not days. Listening comprehension improves too slowly to feel from the inside, so a day’s number tells you almost nothing and a calendar of the year tells you nearly everything. That is where a trustworthy count earns its keep: it lets you see this month against three months ago and believe the difference.

The point isn’t a smaller number

It’s a number you can trust next week. Progress in listening is slow and mostly invisible from the inside; the only reliable way to see it is to compare this month to three months ago. That comparison is only meaningful if the units didn’t change, and “minutes watched” changes its meaning every time your viewing mix does. Hours of real input doesn’t.

This is the number Kiku Hours is built to produce. It is a free Chrome extension with no account that reads time from the player rather than a clock, ignores ads, pauses and seeks, keeps counting in a hidden tab, and credits the minutes a video actually took rather than its length. It listens to each video’s audio to identify the language, across 106 languages, holds back time in a language you already speak with the amount shown, and leaves an unclear video counted but unlabelled. Every interval is stored separately in your browser, with the video it came from, so the total can always be inspected, exported, backed up and restored. It works on YouTube, Twitch, Kick, Vimeo, Dailymotion, TED, Bilibili and Niconico, in Google Chrome or Firefox on a computer, and setting it up takes two minutes.