A false start worth showing
The first pass at this looked spectacular. Communities whose card contained a number had a median of 100 members against 55 for those that didn’t — an 82% lift. Mentioning a dollar figure: +77%. Saying “free”: +50%.
All of it was an artefact.
Compare communities within the same size band — so that size is roughly held constant — and every one of those effects collapses:
| Feature | Apparent lift in members | After controlling for size |
|---|---|---|
| Contains a number | +82% | +0% to +7% |
| Promises a result | +5% | −2% to +4% |
| Topic framing | −2% | −3% to 0% |
| Asks a question | −19% | −6% to −4% |
The lifts were describing which size bands those communities happened to sit in, not anything the words were doing.
It is worth stating the correction to the obvious explanation too. The tempting story is reverse causation — a community with five thousand members writes “join 5,000 members” and a new one can’t. That turns out not to be the mechanism: of 8,245 cards containing a digit, only 101 cite a member count. The numbers are prices, timeframes and ages. The effect was composition, not self-citation.
So the honest headline is that packaging does not appear to drive membership at all. Which would be the end of a short article, except that it moves something else.
What packaging does predict
Switch the outcome from how many members to how active they are — posts per member, compared only within size bands — and the same features that produced nothing on size produce clear, consistent signal.
| Card feature | 20–49 members | 50–249 | 250–999 | 1,000+ |
|---|---|---|---|---|
| Asks a question | +41% | +41% | +12% | — |
| Contains emoji | +37% | +18% | +10% | +10% |
| Addresses "you" | +14% | +3% | +5% | +10% |
| Promises a result | +0% | +7% | +6% | +11% |
| Under 90 characters | −19% | −17% | −7% | −8% |
| "Learn / discover / explore" | −25% | −16% | −13% | −13% |
Every row keeps its direction across every band where the sample allows a comparison. That consistency is what makes this worth reporting — four independent slices of the platform agreeing is a much better signal than one large number.
The worst thing you can write is the most natural thing to write
Topic framing — learn about, discover, explore, all things X — is used by 17.2% of communities and is associated with the lowest activity per member at every single size, from −25% in the smallest communities to −13% in the largest.
This is the default way to describe a community, because it is the honest answer to “what is this about.” It is also, apparently, the framing that produces the quietest room.
The plausible reading — and it is a reading, not a finding — is that it sets the wrong expectation. A card promising that you will learn about something recruits people who intend to consume material. A card that asks you a question, or names a problem you have, or addresses you directly, recruits people who expect to say something. Consumption does not generate posts.
That would also explain why the effect is strongest in the smallest communities, where a handful of members determines whether the place feels alive, and weakest at scale where volume carries it.
Brevity is not a virtue here
The median card runs 127 characters against a ceiling of about 196 — so most owners are using roughly two thirds of the space available, and a quarter use under 90 characters.
Those short cards correlate with 7–19% lower posting per member. There is no support here for the instinct to be punchy. The cards that precede active communities use the room.
What to actually do
- Stop optimising the card for growth. On this evidence it does not do that job, and time spent rewriting it to attract more members is time spent on the wrong surface. Traffic comes from elsewhere.
- Do optimise it for who shows up. That is the job it appears to do, and it is the harder problem — a quiet community is a worse outcome than a small one.
- Cut "learn about". It is the single most consistent negative in the data and it is probably what your card currently says.
- Ask a question, or name a person. The strongest positive signal, and almost nobody does it — only 2.3% of cards contain a question mark.
- Use the space. If you are under 90 characters you are in the band that under-performs at every size.
Caveats
This is the card, not the whole About page. The card is the compressed pitch that appears in listings and advertising. The full page carries a cover image, a video, proof and a much longer body, none of which is measured here — and owners consistently report the cover image as the single biggest lever on conversion, which this analysis cannot speak to at all.
Posts per member is a crude proxy for health. It rewards chatty communities and penalises ones built around live calls, courses or one-to-one work, which may be thriving silently.
Correlation, and the causal story is genuinely uncertain. Emoji almost certainly do not cause engagement; they signal an informal owner, and the owner is doing the work. The same doubt applies to every row. What the table supports is that these features travel with activity — which is useful for diagnosis even where it is useless for cause.
Sampling. These are communities observable on the platform, which skews toward larger ones, and the analysis is restricted to those with at least twenty members so that a posts-per-member ratio means anything. Question marks are the thinnest slice at 867 communities overall.