Which Short-Form Metrics Are Vanity and Which Are Signal

Every analytics screen offers dozens of numbers. Most are decoration: they move, they feel like progress, and they change nothing about what you make next week. A small set earns its place, because each one tells you something you could not have guessed and points at a specific change.

The test is easy to apply and slightly brutal. For any metric, ask what you would do differently next video if it doubled. If the honest answer is "nothing", it is vanity. Keep watching it if it motivates you, but do not let it steer.

Views measure distribution, not quality

A view is a delivery event. It says the platform put the video in front of someone, nothing more. The same video can get nine hundred views or ninety thousand depending on the hour it went out, the account's recent history, and whether one early viewer shared it.

Because views are decided largely by the feed, they are a poor per-video judgement and a decent per-month trend. Watch the monthly shape with a stable format and ignore the single outlier in either direction. A lucky video is not evidence of a method, and an unlucky one is not evidence against it.

Likes are the cheapest thing a viewer can give

A like costs one tap and commits the viewer to nothing. It is a weak measure of interest and a reasonable measure of agreement. Over many videos, likes per thousand views can show whether a format is landing emotionally. As a per-video signal it will send you in circles, because people like videos they abandoned at second four.

Followers are a historical record, not an audience

The public count essentially only goes up, so it hides churn. It also mixes audiences: a thousand followers gained from one off-topic video are not a thousand people who want your usual work, and their indifference will drag down the early performance of everything you publish afterwards.

What matters is conversion. Follows per thousand views tells you whether a video made people want more of you. Returning viewers tells you whether they came back. Both survive a lucky week; neither can be bought with a single viral post.

The four metrics that predict the next video

Four numbers are enough to run a short-form channel on:

• Percentage viewed, matched by length. The only fair measure of whether a video held attention.
• Shares per thousand views. A share puts the video in a private conversation, which costs the viewer something.
• Saves per thousand views. A save says the viewer expects to need this again.
• Follows, profile visits or link clicks per thousand views, depending on what the account is for.

Shares and saves are the ones worth protecting, because they are the only engagement that costs the viewer social capital. Someone can like a video out of politeness; nobody sends it to a friend to be polite.

Ratios beat totals

Totals scale with distribution, so they reward luck. Ratios describe the video.

Take a video with a hundred thousand views and twenty shares. Now take one with four thousand views and forty shares. The second is the better video by a wide margin, and the first is the one that will feel like the success. If you only look at totals, you will spend the next month trying to repeat the accident instead of the method.

Total watch time flatters long videos

Watch time is duration multiplied by retention. A three-minute video watched to thirty percent produces more watch time than a twenty-five-second video watched to seventy percent, so if watch time is the target you will be pulled toward longer and looser, which is the opposite of what short-form rewards.

Percentage viewed keeps comparisons honest across lengths. Keep both figures if you like, but read the percentage.

A four-number weekly screen

One line per week, four numbers, no more:

1. Views, as a weekly total.
2. Percentage viewed, averaged across the week's videos of similar length.
3. Shares per thousand views.
4. Follows per thousand views.

Four numbers on one line fits on a screen and can be compared to last week at a glance. Twenty numbers cannot be compared to anything, which is why dashboards with twenty numbers change no decisions.

When a rising number is bad news

Progress is not always up. Three cases worth recognising:

Watch time rises while percentage viewed falls. You made longer videos that held people less. The headline improved and the product got worse.

Followers rise fast from a topic you do not make. The new audience arrived for something else. Every subsequent video is now judged by people who did not ask for it, which suppresses the distribution of your real work.

Views rise on a format you dread making. A format you dislike is a format you will abandon in six weeks, taking its numbers with it. Treat the temptation as a warning.

Low volume makes ratios noisy

Honesty about the limits: at a few hundred views, one share is a large swing in shares per thousand. Below a few thousand views per video you are reading tea leaves on percentages, and the fix is not a better dashboard but more posts and a longer window. Aggregate weekly, and treat any difference under about twenty percent as unproven until it repeats.

A worked comparison

Two videos from the same month, same length, same topic.

Video A: 120,000 views, 190 new followers, 14 shares, 22 saves, 38 percent viewed. Video B: 9,000 views, 210 new followers, 96 shares, 140 saves, 61 percent viewed.

Totals say A won by an order of magnitude. Ratios say B is the better video: it held nearly two thirds of its audience, and it produced more follows from a thirteenth of the views. A managed roughly one and a half follows per thousand views; B managed twenty-three.

The decision that follows is not "make another A". It is "find out what B did and do it again", because B is the video that would compound if it had A's distribution. A was a delivery event. B was a product.

Reporting numbers to a sponsor without misleading them

When a brand asks for your analytics, they will usually ask for averages. Give them medians and a range instead, and say so plainly.

An average views figure on a small account is dominated by one outlier, so it is both true and useless: it describes a month that will not repeat. "Median 4,200 views, range 900 to 120,000, with the high figure driven by one video the feed picked up" survives scrutiny and starts a better conversation than a number that quietly over-promises.

The same applies to completion. Report percentage viewed for videos of comparable length, not a single figure across fifteen-second and three-minute videos, because the second will always look worse and the client will conclude the audience is bored.

Being the person who explains the limitation is also a commercial advantage. It is the opposite of what brands expect, and it moves the next conversation from metrics to the work.

That is also why hook testing at small scale is designed around sequential tests rather than clean A/B splits, and why the monthly review is a better unit than the week.