In short: a community earns its keep by doing work the company would otherwise pay for. Decide the job first, design the community around it, and measure contribution instead of size.
Key points
- Member counts are easy to buy, easy to inflate with incentives and nearly impossible for outsiders to verify.
- When X revoked API access for pay-to-post apps in January 2026, those programmes shut down and linked tokens fell sharply.
- The five community jobs are test, teach, refer, build and govern; pick one or two.
- AI communities usually fit testing and teaching, as power users find failure cases faster than QA.
- Keep the community in channels whose records you control so it survives a platform policy change.
Why member count misleads
Member counts are easy to buy, easy to inflate with incentives and nearly impossible for an outsider to verify. Investors know this. Customers don't care. The only person a big number reliably impresses is the founder.
Incentive-driven growth also carries platform risk. When X revoked API access for apps that rewarded users for posting in January 2026, pay-to-post programmes built on it shut down and the tokens tied to them fell sharply (source). Any community strategy that depends on one platform's API can be switched off by that platform.
The five jobs
| Job | What the community does | What you measure | Fits best |
|---|---|---|---|
| Test | Uses early builds, reports bugs, gives feedback | Bugs found, feedback sessions, release sign-offs | Pre-launch products, AI tools |
| Teach | Answers questions, writes guides, onboards newcomers | Share of questions answered by members, guides written | Developer tools, complex DeFi |
| Refer | Brings customers, partners or hires | Referred sign-ups and their conversion | B2B AI, infrastructure |
| Build | Integrates, extends, ships on top of you | Integrations, plugins, forks used | Protocols, platforms, APIs |
| Govern | Makes real decisions with real consequences | Participation by people who aren't whales or the team | DAOs, protocols with live treasuries |
Pick one or two. A community asked to do all five does none of them well.
Web3 and AI, differently
Web3 communities are often holders first and users second. That's fine if you give holders a reason to become users: early access, a role in testing new markets, recognition in the product. It goes wrong when the only shared interest is price, because price is the one thing the team can't promise.
AI communities are usually users or developers. The job is often test and teach: power users find failure cases faster than your QA, and their prompt tricks become your best onboarding content. Treat them as an extension of the product team.
Designing for the job
- Channel: pick the place where the job gets done. Testing happens in a structured feedback channel or a private group, not a 10,000-person server.
- Rituals: a weekly office hour, a monthly build review, a release preview. Rhythm beats announcements.
- Recognition: name contributors publicly and specifically. In web3, onchain credentials or roles can make this portable; in AI, changelog credits and early feature access work well.
- Ownership: the community lives in channels you control the records of, so it survives a platform policy change.
The checklist
- [ ] One or two named jobs for the community
- [ ] A contribution metric for each job
- [ ] No growth tactic that depends on paying for posts
- [ ] A weekly ritual that moves the job forward
- [ ] A way to recognise contributors by name
Common questions
How should a startup measure community success?
Measure contribution to a defined job, not size. Depending on the job, that means bugs found, questions answered by members, referred sign-ups, integrations shipped, or governance participation by people who are not whales or the team.
Why is community member count a bad metric?
Member counts are easy to buy, easy to inflate with incentives and nearly impossible for an outsider to verify. The only person a big number reliably impresses is the founder.
What are the risks of incentive-driven community growth?
It carries platform risk. When X revoked API access for apps that rewarded users for posting in January 2026, pay-to-post programmes shut down and the tokens tied to them fell sharply.
How are web3 communities different from AI communities?
Web3 communities are often holders first and users second, so they need reasons to become users. AI communities are usually users or developers, and their job is often testing and teaching.