A tutorial from three years ago may not work any more. How to check before you start

You find a tutorial, it looks right, and six minutes in the button it tells you to click is not on your screen. By then you have paid the whole cost of the video and got nothing, which is the specific failure that makes stale tutorials worse than merely bad ones: a bad tutorial announces itself early, and a stale one waits.

The instinctive fix is to sort by date. It is a bad fix, and it is bad in an expensive direction — the best explanation of most things is usually not the newest one, so filtering by recency systematically throws away the good material to avoid a problem that only affects part of the library.

First decide whether your question has a version number

This is the whole method, and everything below is detail.

Some questions are about how a thing works: why an algorithm is fast, what happened in 1919, how the perfect tense is formed, why a bridge stands up. Answers to those questions do not expire. A well-made explanation from years ago is not a compromise, it is usually the better option, because it has had time to be found and recommended by people who watched the whole thing.

Other questions are about how to do a specific thing in a specific product, what something costs today, or what the current state of a field is. Those answers have a version number attached whether or not anyone wrote it down, and they go stale on a schedule set by whoever ships the next release.

Almost all of the harm comes from treating the second kind as if it were the first. Our own reading of the library separates these cleanly: material tied to a moment scores very differently from material that is not, and the split follows the subject far more than the video.

The checks, cheapest first

Read the title for a version number, even an implied one. “How to set up X” is version-bound in a way “How X works” is not. A product name in the title is a date stamp in disguise.

Look at what the first frames show. If the video opens on an interface, you are in version-bound territory and the date now matters. If it opens on a diagram, a whiteboard or a face, you are usually not.

Sort the comments by newest, not by top. This is the highest-value thirty seconds available to you. The top comments are from the week the video was published and they are all positive; the recent ones are from people in your position, and if the tutorial has stopped working, one of them has already said so. The catch is coverage — it only exists on videos with an active comment section, and on a quiet video there is nothing to read.

Check whether the thing being taught has a canonical home. For anything with official documentation, a two-minute look at the current documentation tells you whether the shape of the process still matches what the video is about to show you. You are not reading it to learn the process. You are reading it to see whether the steps still have the same names.

Prefer the video that explains why over the one that lists steps. A tutorial that tells you what it is doing at each stage survives a change in the interface, because you can find the new equivalent of the thing it described. A tutorial that is purely a sequence of clicks breaks completely the first time one of them moves.

What the date is actually good for

Nothing said here means the date is useless. It means the date is a second-order check that most people apply first.

Once you have decided your question is version-bound, the date becomes the most important thing on the page and you should treat it strictly — for a fast-moving product, a video from two years ago is a different product. Once you have decided it is not, the date is nearly free information and you should be willing to watch something old without a second thought.

The mistake is not using dates. It is using them uniformly, on every video, as a proxy for quality.

The failure mode nobody warns you about

There is a second kind of stale that the date will never catch: material that was never version-bound and simply turned out to be wrong. Nutrition advice, productivity systems, investing rules of thumb — these do not break when a menu moves, they erode as the evidence behind them changes, and a video making a claim that has quietly stopped being supported looks exactly like a video making one that has not.

We do not solve that one. Our measurement is of how much a video says per minute and how tied that material is to a moment, not of whether the claims are true — a limit worth stating plainly, because it is the limit people most often assume we have crossed. If a video’s problem is that it is confidently wrong rather than out of date, nothing on this page will catch it and neither will we.

The order to work in

Decide whether the question has a version number. If it does not, ignore the date and pick the best explanation you can find, however old. If it does, check the date, then sort the comments by newest, then glance at the official documentation for the current names of things. If a video passes all three, it is very likely fine.

And before any of that: know how long the video takes to start. The median measured video does not begin delivering until 1:00, and one in ten does not start until 3:46 — which is a long time to spend discovering that a tutorial was out of date all along.