Cheap to count

Every metric is an instruction. You think you are choosing what to measure; you are actually choosing what to make. Pick a number to steer by and you have told the work what it has to become in order to move that number. Choose a cheap-to-count metric — impressions, reach, views — and you have issued a quiet instruction that the work should be cheap. Not cheap in dollars. Cheap in what it costs to earn: easy to produce, easy to inflate, easy to move without anything happening on the other side of the screen.

This is not an essay about which metric is better. We have written that one, more than once. This is an essay about what happens the instant you pick — about the selection pressure a measurement choice exerts on everything downstream of it, before a single piece is made. Impression-led content optimizes for what is cheap to count, not what is hard to earn. And it does so not because the people making it are lazy or cynical, but because the number they were handed can only see the cheap thing. What the number cannot see, the work stops doing.

Cheap to count is a category, not a complaint

Start with what "cheap to count" actually means, because the phrase has to do real work and it is not a slur.

A metric is cheap to count when its growth can be purchased independent of whether the work was worth anyone's attention. Impressions are the pure case. You can always make impressions larger — more spend, more posts, more surfaces, more reformatting of the same thing into more shapes. The ceiling is your budget, not your quality. The same is true of reach, of views, of publish count, of follower totals. Every one of them will rise if you pour enough money and enough output into the top of the funnel, and every one of them will rise whether or not the thing you made was any good. That is not an accusation. It is a structural property. These numbers are trivially inflatable by construction, and the triviality is the whole point: a number you can always move is a number that tells you almost nothing when it moves.

Hard to earn is the mirror property. A metric is hard to earn when it moves only when the work did its job inside a real person — when no amount of spend, on its own, can lift it. You do not get to that number by buying more of the top of the funnel. You get to it by being good enough that a person kept going. The contrast between the two classes is not a matter of degree, one number slightly noisier than the other. It is a difference in kind. One class can be manufactured from the outside; the other can only be earned from the inside. That line — cheap to count on one side, hard to earn on the other — sorts every metric you will ever be offered, and the sort is the argument.

Notice what has not happened yet. I have not told you a single number about how anyone's content actually performs. I do not need to. The claim is not that cheap metrics correlate with worse outcomes in some dataset; the claim is that they are inflatable by construction, and that this property alone is enough to bend the work. You do not measure your way to this conclusion. You reason your way to it from what the metrics are.

The number rewrites the brief

Here is the mechanism, and it is quieter and more automatic than the usual "what gets measured gets managed" slogan admits.

When you install a metric as the thing you steer by, you have not just added a gauge to the dashboard. You have rewritten the brief for everyone working under it — including yourself. Because the honest way to hit a target is to do more of whatever moves it, and people are good at finding out what moves it. If the number that decides whether your work "worked" is impressions, then the fastest, most reliable way to make your work succeed is to produce the cheapest possible unit that generates an impression. Not because anyone decided to make disposable content. Because the reward function they were given pays out fastest for exactly that, and a reward function is an instruction whether or not anyone reads it aloud.

This is how noise gets manufactured on purpose. Most content is noise, and almost none of it was made by someone who set out to make noise. It is the residue of a reward function that pays for volume and is blind to signal — cheap units produced at the rate the counter rewards, each one individually defensible, all of them together indistinguishable from the din. The noise is not a moral failure. It is what a cheap-to-count metric prints when you leave it running.

This is why the choice of metric is upstream of everything. The metric you choose is an instruction to the work; choose a cheap-to-count metric and you have instructed the work to be cheap. The instruction does not arrive as a memo. It arrives as a thousand small local decisions — make it shorter, make it faster, make it fit the format that travels, ship the version that clears the counter today — each of which is individually rational and collectively fatal to anything hard to earn. The hard-to-earn thing does not get argued out of the plan. It quietly drops out of the objective, because nobody is counting it, and a thing nobody counts is a thing the system cannot feel itself losing.

This is the cause; the forgettable content is the symptom

We have described the downstream of this before. A while ago we argued that volume metrics select for content that is forgettable on purpose — that when output is the number, the selection pressure runs toward the cheapest possible unit of output, and forgettable is what cheap looks like at scale. That essay was about the output of the bad reward function. This one is about its cause.

The relationship is strictly upstream and downstream, and it is worth naming so the two do not blur. Choosing a cheap-to-count metric is the decision; forgettable content is what that decision produces once it has run for a while through a real team with real deadlines. You do not get the forgettable output without first making the measurement choice that rewards it. So if you have read the symptom and wondered where it comes from — why capable people who want to make good work keep producing forgettable work anyway — this is the origin. It is not a talent problem or a taste problem. It is installed one layer up, at the moment somebody decided which number the work would be judged by.

Why the cheap number is the attractive one

If cheap-to-count metrics bend the work, why does everyone keep choosing them? Because cheap to count is also cheap to report, cheap to defend, and cheap to celebrate — and those are real pressures on a real team.

We have written about the number everyone knows is wrong — the headline figure teams report even while half-aware it is not measuring what they care about. The reason the wrong number survives that half-awareness is exactly its cheapness. It is available today, it always went up, it fits in a slide, and it never embarrasses anyone in a meeting. A hard-to-earn number has none of those comforts. It is slow, it can sit still while you spend, and it can be low in public. Between a number that flatters on demand and a number that occasionally tells you your work did not land, most incentive structures will select the flatterer — not because anyone is fooled, but because the flatterer is easier to live with. The attraction of the cheap metric is not that people believe it. It is that it asks nothing of them.

The number with no purchase path

Set against all of this is a single reader-side number that behaves the other way. Completion is the only reader-side metric that cannot be inflated by spend.

Read that as a property, not a boast. Every other reader-side number has a purchase path: you can buy the impression, buy the click with a headline built to mislead, buy the open with a subject line that over-promises, buy the start of a read. What you cannot buy is the finish, because finishing is not a decision made at the door. It is a verdict rendered continuously, sentence by sentence, and the only thing that keeps a reader rendering it in your favor is that the next sentence was worth the last one. There is no budget line for that. The internet rewards what people finish, not what brands publish — the return visit, the forward, the slow reason to come back attach to the work a reader completed, not to the work you shipped.

I want to be careful about what that does and does not claim. It does not claim that finish-rate is therefore the metric you should adopt, or walk through why it is the right one to build on — we made that affirmative case in full elsewhere, and this essay is deliberately upstream of it. The point here is narrower and structural: completion is an example of a hard-to-earn number, and its un-buyability is what qualifies it. Whether you build on it is the next essay's question. Whether you can even see it is this one's — and if the number you chose to steer by is cheap to count, you cannot, because the cheap number was never watching the reader in the first place.

The gate we hold

It would be cheap of us, in an essay about cheap measurement, to describe a hard standard and then not sit under one. So, plainly: we measure completion, not impressions. The gate we hold ourselves to is a 65% finish-rate on long-form editorial — median per piece, on a four-week rolling window, sustained across two months before it counts as met.

That is stated as the discipline we impose on ourselves, present tense, not as a result we are claiming to have banked. The whole number travels together on purpose: the median rather than the average, so one outlier cannot launder a body of work that did not finish; the four-week window, because a single piece's completion is noisy; the two sustained months, because a standard worth trusting has to survive time rather than spike once. We name it here for one reason only — to show that we count the hard-to-earn thing rather than the cheap one. Not to argue that this is the number you must keep. Simply to make concrete what it looks like to steer by something you cannot buy your way to.

Choose the instruction on purpose

The trap is easy to describe and hard to escape, because the cheap metric is always the one already on the dashboard. It is there by default, it is free, and it will rise the moment you do anything at all. So the work drifts toward it, not by decision but by gravity, and one day you look up and the whole operation is optimized for a number that was never watching the reader.

The escape is not a better dashboard. It is remembering that the number is an instruction, and choosing it as deliberately as you would choose anything else that decides what you make. Cheap to count will always be available, and it will always flatter you, and it will always be beside the point. Hard to earn will always be slower, quieter, and more willing to tell you the truth. You are going to be steered by one of them. The only real choice is which — and you make that choice, whether you know it or not, the moment you decide what to count.