Thumbnails are chosen at fingernail size

You spent an hour on it. You picked the frame, you cut the subject out, you nudged the text until it sat properly, and on your screen it looked good, because on your screen it was the size of a paperback.

A thumbnail may appear small on a phone, beside competing images, while someone is doing something else. It can also appear larger on desktop or television. Check the contexts your audience uses.

The fix: shrink it to the size it will be seen at and put it in a row with others before you decide. Every time.

This is not a design skill problem. It is a viewing conditions problem, and it is fixable this afternoon with nothing but the file you already have.

Where is your thumbnail being chosen?

Often at a small size, in a list, with little time to make its subject clear.

Those conditions are absent when you design. You look at the image on its own, large, for as long as you like, already knowing what the video is about. Every one of those advantages makes the image look better to you than it will ever look to anybody else, and none can be switched off by concentrating harder.

The losses run in one direction. Detail goes first, then contrast between similar tones, then anything that needed reading rather than recognising. What survives is shape and colour, which is why a thumbnail that works is almost always simpler than its designer wanted it to be.

Why does it look fine on your monitor?

Because on your monitor you are not choosing. You are approving.

Approving and choosing use different information. Approving asks whether the thing is well made, the crop tidy, the text spelled correctly. Choosing is cruder: scanning for whichever image resolves fastest into a recognisable thing.

So your version is being assessed on craft and theirs on legibility. An image can pass the first test comfortably and fail the second completely, and you will never notice, because you are looking at it in the conditions where it passes.

How do you look at it the way they will?

Shrink it before you decide, every time, without exception.

Export the image and open it on your phone. Not the desktop preview at half size. The device itself, the list itself, alongside the others it will sit beside. It takes about a minute and it is the most effective habit in this subject.

Squinting is a rough check for dominant shapes and contrast. It does not reproduce a phone display, so use it alongside an actual small-size preview.

Why does it matter what sits next to it?

Because you are never being compared with your last video. You are being compared with whatever happens to be around you.

Check the thumbnail alone for clarity, then among relevant neighbouring images for context. A screenshot can give you a useful comparison, although recommendations differ between viewers.

If neighbouring thumbnails use similar colours, try a different treatment and see whether it makes your subject clearer. Standing apart helps only if the image still represents the video.

Shapes that survive being small

Three working rules, and they are ours. They come from watching which thumbnails hold together once they have shrunk, and they are offered as a starting shape you should break as soon as you have a reason to.

Four words on the image, or fewer. Not the video title, which has room. The words sitting on the picture. A fifth is usually a qualifier that made sense on your monitor and is a grey smear by the time anybody sees it.

A face, if there is one, at about a third of the frame. Smaller and the expression stops being readable, which is the only thing a face is there for. Larger and it pushes out the subject, and a face with nothing around it tells a stranger nothing about what they would be watching.

Corners left clear. The platform puts things there without asking: durations, progress bars, watched badges, its own labels. Anything in a corner is underneath something you cannot see while you are designing.

None of the three is a rule in the sense that breaking it is wrong. They are the shape that survives after everything fine has been thrown away, which is why they read as arbitrary on a monitor and obvious on a phone.

Title and image as a pair

They arrive together, they are read together, and most people write them as though only one of them has to work.

Check whether title and image reinforce a clear promise. Repetition can help recognition; complementary detail can add interest. The problem is ambiguity or a pair that implies a different video.

What you want is two halves of one thought. The image raises the question and the title answers part of it, or the image shows the result and the title says what it cost. Read them out loud together as one sentence. If the sentence repeats itself, or does not parse, that is the fix, and it is a writing fix rather than a design one.

What do you do with the thumbnails you have already published?

Take your six most recent, put them on your phone, and sort them by how quickly you can tell what each one is from the list.

The bottom two are your instruction, and the job is not to redesign them. Look at what the top two do that the bottom two do not, in the crudest terms: fewer things in frame, more separation from the background, a colour that stands apart. Write down what you find, because it is specific to your subject and nobody could have told you in advance.

Then the decision that changes the next year of the channel. Every thumbnail from now on gets shrunk and put in a row before it is published, or it does not. That workflow choice is the only one here that compounds.

Sources and evidence

What here is independent research and what is our own working method.

Standards and platform facts. A study of thumbnails from branded YouTube videos, published in Decision Support Systems and covering 3,745 videos, found real relationships between a thumbnail's visual attributes, including complexity, how many subjects appear, faces and basic image properties, and how well the video performed. That supports the idea that these attributes matter. It does not validate the specific defaults below, four words or a face at about a third of the frame, so we hold those as a starting shape to break rather than a rule proven by the research.

Our method. Shrinking every thumbnail to the size and setting it will actually be seen at, squinting at it, and putting it in a row against recent uploads before publishing, is this studio's habit regardless of the exact numbers, because it is the viewing condition itself that decides what survives, not a rule of thumb.

References.

• Decision Support Systems. Thumbnail visual attributes and view-through (3,745 branded videos). sciencedirect.com. Thumbnails: complexity, subject count, faces and image properties have real relationships with performance (not simply 'more is better'; industry moderates). Does NOT validate the exact defaults (four words / one-third face); keep those as channel-overridable starting points.