Shelf life: how long the material in a video stays true, by subject

Everyone has clicked a tutorial, followed it for six minutes, and found that the button it tells you to press is not there any more. The obvious defence is to check the date and prefer the recent one. That defence is aimed at the wrong variable.

We read every measured video for shelf life: how far the material is tied to a particular moment, version or price, as against how far it would still be correct in a few years. Across 3,137 videos, the reading is driven overwhelmingly by subject, and barely at all by anything a viewer can see on the results page.

A note on the numbers. Unlike the rest of this site, the figures on this page are frozen at the date above rather than recalculated on each visit. A report people cite should not move under them. When we run this measurement again it will be published as a new edition with a new date.

The subject decides it

Science at 84.8, history at 83.1 and engineering at 78.7 sit at the top; tech news at 37.9 sits nineteen points below the next subject up. That gap between tech news and travel is the largest single step anywhere in the chart, and it is the shape of the whole result: one subject is about things that changed this week, and everything else is not.

Programming, the subject people worry about most when they think about stale tutorials, lands mid-table at 70.1. Machine learning, which has a reputation for moving too fast to document, lands slightly above it at 73.8. Both are well above finance at 64.1 and design at 62.3, neither of which is usually thought of as a fast-moving field.

The pattern behind the ordering is not the pace of the field. It is how much of the material is tied to a specific version of a specific thing. A video about how a sorting algorithm works does not go out of date; a video about how to do something in a particular application’s current menu does. Programming contains a great deal of the first kind, which is why it scores better than its reputation.

The spread matters more than the average

An average hides the thing you actually need, which is whether a particular subject is a coin flip or a safe bet.

Three quarters of measured science videos score above 80. Three quarters of measured tech-news videos score below 48. Those two distributions barely overlap: within tech news, the top quarter of videos is roughly where the bottom quarter of science begins.

Programming and machine learning are the interesting middle. Their medians are high — 76 and 80 — but their lower quartiles are 62 and 65, which means a quarter of the videos in each carry material that is meaningfully tied to a moment. In those subjects the question “is this still current” is a real question with a real answer either way, which is exactly when it is worth asking.

News and politics is the widest spread on the table: a lower quartile of 45 and an upper of 80. Some of it is analysis of things that do not change and some of it is a report on a week, and from the outside the two look identical.

What does not predict it

Two things people reach for as proxies do not work.

Difficulty moves the reading by four points across its entire range. Introductory videos average 69.6, advanced videos 71.2 — and advanced is fractionally higher, which is the opposite of the common assumption that deeper material dates faster. Advanced material tends to be about principles, and principles keep.

Runtime moves it even less. Four of the five bands sit within 1.2 points of each other. The hour-plus band is higher at 81.1, on forty-nine videos, and we would not build anything on that.

What does move it is the kind of video: explanatory material averages 71.3, news material 25.5. That is a forty-six-point split, and it is the same finding as the subject chart seen from a different angle.

What this means when you are choosing

The practical rule that follows is not “prefer recent videos”. It is decide first whether your question is version-bound, and let that decide how much the date matters.

If you are asking how something works — a mechanism, a proof, a piece of history, a grammar rule — the date is close to irrelevant and a five-year-old video from someone who explains it well beats a fresh one that does not.

If you are asking how to do a specific thing in a specific product, or what something costs, or what the current state of a fast-moving field is, then the date is the first thing to check and the video’s own quality is the second.

The subjects where this matters most, by our reading, are tech news, travel, business and news commentary. The subjects where it matters least are science, history and engineering — where a good explanation from years ago is very likely still a good explanation.

What this measurement does not say

It is not the video’s age. We do not read publication dates, and nothing on this page is a claim about how old anything is. This is a reading of how tied the material is to a moment, which is a different question and the more useful one — a fresh video about a product release is dated on the day it is posted, and a five-year-old explanation of a stable idea is not.

It is not a check that the video still works. A high shelf-life reading says the material is the kind that lasts. It does not verify that the specific steps in a specific tutorial still run today.

It is not accuracy. We measure how durable the material is, not whether it was right in the first place.

How this was measured, and what it does not say

Shelf life is one of the readings behind the value score, on the same 0–100 scale as the rest. The reading is ours. Method and its limits are in what value density means.

It is read on 3,137 of the 3,243 measured videos. The rest carry a value score without this particular reading and are excluded from every figure here rather than counted as average.

Four subjects rest on fewer than seventy videos — comedy, DIY, travel and gaming. Read their position as approximate; the ends of the chart are where the finding is.

The sample is not YouTube. It is English-language explanatory video across twenty subjects, assembled to build a recommendation library rather than to be representative of the platform.

The figures are frozen at 1 September 2026. Scores are recalculated as the underlying measurement improves, so the live numbers on the shelves will already have moved slightly from these. This edition will not be edited; the next measurement will be published separately.