The untapped share of a Skool community sits somewhere between 76% and 95% of the roster, which is ordinary for any online space that has been measured. On its own that number is not actionable, because it lumps together two groups of people who need completely different responses — and only one of those groups is a real opportunity.
The definition
Show-up rate is the share of a community's roster still seen on Skool at all, across a sampling window — as opposed to merely silent in your community.
Its complement is the share who have gone from the platform entirely. We report the positive direction because that is the number an owner can do something about, and because a metric named after failure gets quoted by people trying to sell you a fix for it.
Why it matters more than it sounds
Every engagement strategy assumes its target is reachable. Show-up rate is the measurement that tells you what fraction of your silent majority actually is.
- A member on Skool and ignoring your community can be reached. Something you do could change their behaviour.
- A member gone from Skool cannot. No post, no DM, no campaign will move them, and counting them in your denominator means measuring yourself against a target that cannot be hit.
That split does not currently exist in any published analysis of the platform, and it is the reason this measurement was built.
How it is measured, and why the method is the hard part
Skool shows which members of a public community are online right now. That fact makes a naive version of this metric extremely easy to compute and completely worthless.
The share of members shown online swings roughly eightfold across the hours of the day. A reading taken at 9pm in one timezone and a reading taken at 4am are not two measurements of the same thing. Any statistic built on one reading is a measurement of the hour it was taken. We published a version of that error earlier this year, retracted it, and the retraction is what produced the design below.
The design. The same communities are read every 7 hours. Because 7 and 24 share no common factor, the reading slots walk through all 24 hours of the day, so every community receives an identical hour distribution by construction — the confound is eliminated structurally rather than adjusted for statistically after the fact.
Distinct member identities are then unioned across all readings. A member seen once counts as present. The union only grows, so it cannot be biased by which hour any individual reading landed on.
A known limit. The anonymous route returns at most 30 identities per reading and ignores pagination. Where more than 30 members are online at once, that reading is truncated and the resulting union is a floor rather than an estimate. How much that matters depends on whether the visible 30 turn over between readings, which is a thing we can test rather than assume — and it does turn over almost completely, so repeated readings keep reaching new people rather than circling the same faces.
The run was extended, and why that is worth saying. It was designed as 24 readings over a week. Partway through, the count was still finding large numbers of people it had not seen before — it had not settled. The pre-registered rule for that situation says to keep sampling rather than publish an estimate, so collection was extended to 40 readings and now finishes on 29 August. Nothing about the method changed — it runs longer, that is all. A count that stops when you hoped it would finish, rather than when it actually finishes, is not a measurement.
The checks this number has to pass before it is published
Five were fixed before any data existed:
- Hour-effect null. Odd and even readings must agree on unioned reach, while a positive control confirms that single readings genuinely do swing by hour.
- Split-half reliability. Two independent half-samples must agree closely on the ranking of communities.
- Saturation. New identities appearing in the final readings must fall below a threshold. If they do not, the answer is a floor and more sampling — never an extrapolation.
- Negative control. A set of control communities must return approximately zero. Anything else means the route is not returning what we think it is, and everything downstream is invalid.
- Rotation test. Whether the 30-identity window rotates between readings. If it does not, large communities receive floors only, and any claim about size is dropped — because the cap inflates apparent absence in exactly the direction that story would want.
A sixth check was added partway through, once it became clear that the final three readings would land on unusually busy hours and could fail the third check for reasons that have nothing to do with whether the count had finished. It measures a full seven-reading cycle instead, which covers every hour of the day exactly once. It was fixed before the readings it judges, and it is reported alongside the original whatever either of them says.
The figure publishes in the State of Skool report if the checks pass. If one fails we will say which, and publish a floor instead of an estimate — per size band, so a band that settles cleanly is not held back by one that does not. If it fails badly enough, there will be no number at all, and the section will say so rather than quietly estimating.
What to do in the meantime
The distinction is useful before any platform-wide figure exists, and you can act on it inside your own community today. A member who is quiet with you but posting elsewhere on Skool this week is a completely different situation from one whose account has gone quiet everywhere. The first is reachable — something you do could change their mind. The second is not, by any post, message or campaign.
Most owners treat both as the same failure and measure themselves against a target that was never available. Separating them usually makes the real job smaller than it looked.