They mean it kindly. They are beaming at you. And somewhere behind your professional smile, a small part of you is screaming. For multiple reasons.

One because we’re terrible at taking compliments.

Two… because you know what runs itself: a rock. A parking lot. A PDF. Communities do not run themselves. They run because someone is running them, and that someone has spent years getting good enough at the job that it no longer looks like a job. Congratulations. You have achieved professional invisibility.

This is not a metrics problem, exactly. We’ve been told for years that the solution is better measurement: track response times, post volume, engagement rates, NPS scores. And look, metrics matter. But dashboards haven’t solved the core issue, because the core issue was never about data. It’s alwways been translation.

The invisibility of community labor isn’t a measurement problem. It’s a legibility problem. The solution isn’t more dashboards; it’s learning to speak your organization’s language instead of asking your organization to learn yours.

Think about the building superintendent. In a well-run building, you never think about the super. The heat works. The pipes don’t rattle. The lobby is clean. The super is invisible by design, because the super is doing their job. But the moment the boiler breaks, suddenly everyone knows exactly who they are and what they do and why it matters.

Community managers are the super. Except we don’t usually have to deal with heating, cooling, or rattling pipes literally and unless things really go sideways leadership doesn’t always fully register that someone is down there running it. The better you perform, the more you disappear. It is a genuinely perverse incentive structure and I think about it more than is probably healthy.

So what do you do with that?

First, stop documenting only what happened and start additionally documenting what didn’t. A post count tells your leadership how many conversations occurred. It says nothing about the conflict that got quietly defused before it became a thread, the new member who almost churned after an awkward first post but didn’t, the question that got answered in four minutes instead of four days. Near-misses are data. Start treating them that way.

Second: stop asking leadership to learn community vocabulary. They won’t. Not because they’re incapable, but because they already have a vocabulary that works fine for them. If your organization is obsessed with churn, your community work is a retention play. If they care about pipeline, your users are warm leads. If they track support ticket deflection, your peer-to-peer answer threads are saving someone real money. The translation is your job. I know that’s annoying. But alas… it’s also true.

Third, make your labor visible before you need it to be. The worst time to explain what community management actually involves is during a budget review or, worse, a reduction in force. By then you’re playing defense. The case for your work - and YOU - should be ambient, woven into regular reporting, mentioned casually in team meetings, present in the room before anyone thinks to question whether you should be.

Here’s the part that’s hardest to say: a lot of us were trained, explicitly or culturally, to make hard work look effortless. To handle the crisis gracefully, clean up quietly, and never let them see you sweat. That norm made us better at the job and worse at keeping it. The community that “runs itself” is also the community that leadership looks at and thinks, hm, do we really need a full-time person or team for this?

The answer is yes. Obviously yes. But invisible labor doesn’t argue for itself. Don’t hide in the boiler room. Go introduce yourself. Tell someone what’s down there.

Well, as most of us know by now we (or I) didn’t do anything. It was just our AI overlords moving in… noisily, clumsily, and without much warning. Not much we could do about that, and so we moved on with life and they moved on with their hallucinations.

But then… then we starting noticing the problems.

Us humans love quick dopamine hits and nothing hits quite like smugly thinking you’ve solved your problem super fast with a few key words and smashing the enter key. And man did Google’s AI Overviews seem to deliver initially.

In reality, LLMs are kind of just glorified search engines except they’re dialed in to sort of make educated guesses on what words likely surround the word in question. We talked about this in Your Community Isn’t Messy. It’s Training Data.

And all of that is well and good as the internet has many words to consume and frankly us humans could use a hand there. However, my team and I starting noticing that Google’s AI Overviews were surfacing subtly wrong answers about our product. And lucky for us, we didn’t panic. We got curious.

The pattern was interesting: a customer would search something reasonable - like whether they could book multiple appointments in a single session - and the AI Overview at the top of the results page would respond with a confident, slightly off answer. Not a hallucination exactly. More like the model had found a thread with 21 replies, half of them contradicting each other, and landed on the most cautious interpretation it could find. I mean… AI… same. I get it.

That gave us a hypothesis: If community content is what these models are pulling from, but community content is inherently conversational and messy, then the fix isn’t to produce less community content. It’s to give the models something better to work with as a sort of hub model.

So we ran an experiment.

The Setup

We audited our highest-traffic content (top-viewed threads, most common queries, biggest search drivers, highest impressions) and looked at what Google was actually surfacing and what AI Overviews were doing with it. The failure mode turned out to be pretty consistent. Someone posts “can’t do this yet, but it will soon”. A human reads that and understands timing, context, and subtle nuance. A model reads it and files it under “can’t do this” because it craves binary inputs and outputs. Multiply that across hundreds of pieces of user-generated content (UGC) and you’ve accidentally given Google’s AI a reason to be confidently wrong about your product.

LLMs aren’t great with grey area. When content is ambiguous, model confidence drops. When confidence drops, answers get worse. We wanted to see if we could fix that upstream.

We chose to focus on Google Search and AI Overviews specifically because the data is accessible (hello, Google Search Console), the surface is one we knew our users were actually hitting, and it was feasible to test and measure in a reasonable timeframe. It doesn’t capture everything - other search engines, direct LLM queries in tools like Claude, ChatGPT, or Gemini are a different problem - but it was a smart, practical place to start. And honestly, it’s probably a smart place for other community teams to start too.

The Build

We did something a little meta: we asked AI what kind of content it actually wanted to consume. Clear titles. Direct answers. Step-by-step formats. Headers. FAQs. Links to authoritative sources. Then we built a template around exactly that and created an AI agent workflow to do the heavy lifting. It would read the original UGC thread (from URL), cross-reference our help center, apply the template, and output a clean document ready for human-in-the-loop review before publication.

The result is what we’re calling “Asked + Answered” articles: structured Q&A with clear solutions and resolution summaries, purpose-built to become the preferred ingestion target for AI Overviews and public models. We published them into a dedicated, low-profile forum inside our community; not actively promoted, but indexable, findable, and fully controlled by us. And most importantly: still human friendly.

And the original user generated content (UGC)? The original UGC stays and serves in a few ways. It still powers discovery. Customers find us through those threads constantly. But now when Google goes looking for what we can and can’t do, it has a better source to pull from. And, it augments the AI optimized answer, which builds confidence and additional context.

UGC as discovery engine. Optimized content as ingestion target. Both doing their jobs.

What We Found

The hypothesis held… and then some.

More than 50 optimized articles are now fully ingested by public AI systems. Google impressions for community grew roughly 37% from before the initiative to after, and about 90% year-over-year, well above our historical average. We hit the largest month ever for page views, unique visitors, and impressions in the history of the community… while new topic creation actually declined. We theorize the growth is coming from better discovery and utilization of existing content, not just more volume.

The traffic pattern tells a clear story: UGC continues to drive topic discovery. Optimized content improves ingestion and answer accuracy in the AI Overviews that prospects and customers are landing on every day.

The Bigger Opportunity

AI models and search algorithms generally tend to rank community pages above static help center content. That means community teams are sitting on a strategic asset that most organizations aren’t fully leveraging yet. The opportunity isn’t just “write better content.” It’s “write content that makes models confident enough to repeat it correctly while continuing to serve your human audience” - and then verify that they actually do… both.

If you’re thinking about what this looks like for your own community: start with Google. The data is there, the surface is familiar, and the wins are measurable. It starts with an audit, a template, and a willingness to publish content that’s more structured than your usual style. The Asked + Answered articles don’t read like traditional community. They read like reference or help docs. That’s the point.

But the payoff is that when someone searches a question about your product, they get an answer you’d actually stand behind. Your community becomes the authoritative source, not the ambiguous one. And that makes you - the steward of both humans and AI knowledge - pretty darn valuable.

That’s a pretty good place to land.

PS: Big thanks to my colleague David Hartman who ran with this pilot, operationalized it, and continues to ensure Calendly Community is at the right place at the right time with the right answer every single day. He’s becoming quite the AI pro. Watch out world!

It’s not moderation. It’s not event planning. It’s not advocacy stuff. It’s the person who goes first. The one who posts before anyone else has a chance to, who asks the question everyone’s thinking but nobody wants to be caught asking, who walks onto the empty dance floor and starts jamming while the rest of the room pretends to check their phone.

In community work, I call this “being first weirdo”. And it is quietly one of the most important things we do.

Let me allow these shirtless dudes below to demonstrate (things I didn’t think I’d write, but alas).

Watch what happens in that clip (not my video; cannot find original for credit either). One guy, shirtless, in a largely empty stadium section, decides to have fun with it. He waves his shirt. He commits. And then, slowly and then all at once, other dudes join. Not because anyone told them to. Not because it was safe. Because some dude made it safe by going first and surviving. And, actually having fun with it which seems like a great outcome.

And yeah, that’s the job. That’s what I do.

THE END. (kidding, please keep reading)

But here’s the dealio: The first weirdo isn’t performing bravery. They’re performing permission.

When you post the first reply in a brand new forum, you’re not just adding content. You’re telling everyone watching that it’s okay to exist here, that participation won’t be met with silence or worse. You’re saying that the room is inhabited by someone who seems normal enough - or ahem… even stranger than them! You’re absorbing the social risk of going first so everyone else’s activation energy drops just enough for them to follow.

This is a structural intervention dressed up as a personality quirk. The industry tends to celebrate it as a character trait, something about being “willing to be vulnerable” or “authentically engaging” but that framing obscures what’s actually happening. Someone has to go first. You’re deciding it’s going to be you. That’s not a vibe. That’s a deliberate act with real stakes.

And sometimes… you misread the room entirely.

Sometimes you wave your shirt and the section stays empty. You’ve posted into the void, modeled a behavior nobody wanted to replicate, started a thread that got two replies including your own and your moms (hi mom! why haven’t you been liking my substack posts!?!). You’ve danced alone to music nobody else heard. This is also part of the job, and the community management industry is hilariously bad at talking about it, because it doesn’t make for a great case study.

However - good news: misreading the room is data. When nobody joins, you’ve learned something real about what this particular group of people won’t do in public yet, or maybe ever. That’s not nothing. It’s just emotionally expensive nothing. (I’m still working on reframing that one in the moment, for what it’s worth. I am an introvert who needs significant wine in her system to be the first on the literal dance floor.)

The wry truth is that going first means occasionally (read: often) going alone. You budget for it. You shake it off. You go first again next time and the next time and the time after that, because that’s what the role requires.

What the stadium clip captures, though, is the other outcome. The one that reminds you why you keep doing this even when you spent some time creating multiple profiles on your community just to not look like a lunatic talking to yourself. It’s not just that the section fills up. It’s the speed of it. It’s watching the logic of “this looks fun” spread faster than any announcement or nudge or carefully crafted call-to-action ever could. It’s the moment when the thing you started stops being yours and becomes theirs.

That transfer of ownership is the whole point. You go first not to lead forever but to make leading unnecessary. A community that still needs you to be the first weirdo two years in isn’t a community that’s grown, it’s a community that’s waiting. The goal is to go first often enough, and model it openly enough, that other people start to see themselves as someone who could do it too.

And eventually, someone does. They post first. They beat you to the dance floor. They ask the weird question. They walk onto the empty floor and they freaking own it.

And maybe it’s weirdos recognizing other weirds, but you’ll recognize them immediately. They’re doing the thing you taught them without knowing you taught them. They didn’t need a workshop or a framework. They just needed to see someone survive going first.

So here’s the challenge: find your empty section. It doesn’t have to be dramatic (you can keep your literal shirt on). It doesn’t have to work. Post the thing you’ve been waiting for someone else to post. Start the thread. Ask the question. Wave the metaphorical shirt.

Someone’s watching. They just need to see you survive it first.

But do you? Like community manager to community manager… do you?

We’ve been measuring community engagement the same way since the forum era, and honestly, most of it made sense then… when the platform was the community, when showing up to post was the whole point, when activity volume was a reasonable proxy for value. That world still exists in some corners. But for a lot of us, the community has quietly grown into something more complex than a post count can hold. We chatted about this earlier in From Destination to Infrastructure.

So here’s the uncomfortable question: What if I forced you to strip out every default metric your platform hands you - the ones you didn’t choose, you just inherited, or some leader insisted that’s how it was done - and had to build your measurement framework from scratch? What would you actually want to know?

The most honest answer to that question is also the most useful thing you can do for your community strategy.

Historically, what we’ve reached for is traffic, membership, and engagement. How many people showed up, how many joined, how many did something while they were there. These made sense as a starting framework. They’re super visible, they’re easily trackable, and in the early days of community-building they gave you a reasonable read on momentum. A growing member count meant you were doing something right. A spike in posts meant the conversation was alive. Direct traffic told you people were coming back on purpose, not just stumbling in from a search.

The problem isn’t that these metrics are wrong. It’s that they’re incomplete in ways we’ve mostly agreed not to talk about. Membership counts don’t tell you if the right people joined or if they ever found what they came for. Engagement volume doesn’t distinguish between a genuinely helpful conversation and a three-page thread that went nowhere useful. And direct traffic… well, direct traffic just means someone knew your URL. It doesn’t mean anything worked. In fact in the day and age of AI, it might signal that the right information isn’t being ingested.

So what would actually work?

Chances are you didn’t come here to read empty musings. At least I think not. You came here for some real world nerd perspective.

And while I don’t have all the answers, nor can I claim to have absolutely perfected an even near answer - I’ve been working on an approach I call ACE: Audience, Consumption, Engagement. It may be a bit biased towards the support communities I tend to run, but it has elements applicable to all communities.

Audience isn’t about how many members you have. (I retroactively apologize for the ferocious eye rolls I’ve given that metric when asked and the violent eye rolls I’ll continue to give it.) It’s about whether the people who need your collective knowledge can actually find it. Discovery is the metric. Can someone land on your community from a search - internal or external - and reach something useful without already knowing exactly what to look for? If your answer is “our content is all in there somewhere,” that’s not a measurement strategy. That’s a hope, a prayer, a concept of a plan.

Consumption is where things get interesting - or at least I think things get interesting (says the nerd delighted by CSV files). Did the knowledge seeker find it, and did it actually land? Did it change what they did next? The behavioral signal matters more than the view count. Someone reading a solution and then immediately raising a support ticket anyway or not following through on the intended product action is not a consumption success. And that’s a gap you need to know about. The goal shouldn’t be traffic to your content; it’s your content doing the job it was created to do.

And engagement… man, this one requires the most honesty. It should be a measure of necessity, not volume. In a well-functioning support community, engagement is what happens when the existing knowledge base doesn’t cover it. If you’re seeing high engagement, that’s not automatically a health signal. It might mean you have redundant flows, a lack of necessary knowledge, or friction that’s pushing people into creating new threads when the answer already exists three pages back. Engagement when necessary is a feature. Engagement as a substitute for findability is a bug.

None of this means you should torch your existing dashboards tomorrow. Traffic, membership, and engagement volume still tell you something useful for operational health checks. They’re just not the whole story, and we’ve spent a decade letting them stand in for one. The gap between what you’re currently measuring and what you actually care about is where your strategy is quietly leaking brain juice.

You probably won’t rebuild your entire measurement framework this quarter (unless you’re a bonafide data nerd or have one kicking around). But you could start with one honest audit: review each metric on your current dashboard and ask what decision it actually informs. If you can’t answer that cleanly, you already know what to do.

Go find out if your community is doing its job… not just whether people showed up.

The business promise was simple: if we show people exactly what they care about, they’ll feel more at home and therefore happier (read: more spendy, more loyal, etc.) customer.

And honestly? It works. Up to a point.

If you were to ask me what a community is, I’d say something like “It’s a group of individuals with shared values, behaviors, and artifacts.” Cool cool. (Also thanks to The Community Roundtable for drilling that definition into my head.)

But here’s the problem: personalization is also the fastest way to accidentally build a community where nobody shares the same experience. And where the thing that was supposed to make people feel seen quietly makes it impossible for them to find each other.

There’s a useful distinction hiding inside the word “belonging.” Belonging-as-recognition is the feeling that I am seen, known, welcomed. Belonging-as-connection is the feeling that we are in this together. Both matter. But they pull in different directions the moment a personalization algorithm gets involved.

When we personalize for recognition - think tailored onboarding, relevant topic suggestions, the right content at the right time - we’re often doing our best work. New members especially need to find their footing fast, and a well-timed “based on your interests, you might like this thread” can be the thing that converts a lurker into a contributor. That kind of personalization is generous. It reduces the tax of figuring out where you belong.

But personalization-as-connection is trickier, and most platforms haven’t cracked it in a scalable sustainable way. Because connection requires shared reference points. It requires the experience of the same thing happening to all of us at once.

Maybe it’s a spicy and rather aggressive take, but here is what I think: Personalization optimized purely for individual relevance will corrode the commons over time.

Think about what it felt like when your community had a moment. Maybe it was a thread everyone was reading, a debate that split the room, an announcement that landed differently depending on who you were. Those moments aren’t just engagement metrics. They’re the connective tissue. They’re what people refer back to six months later. They’re how inside jokes get born. Remove them, or fragment them by routing different content to different members, and you end up with a lot of people who feel personally served but have nothing to talk about together.

So where does that leave us, practically?

First: protect your shared spaces. Whatever your version of the town square is whether it be the main feed, the weekly digest, the all-community thread, be deliberate about what you do and don’t personalize there. If everything is filtered, there’s no shared ground. Even if some content feels less relevant to some members, relevance isn’t the only value. Shared experience is a value too, but it rarely shows up in your analytics.

Second: personalize the path in, not the experience of being there. Onboarding, topic discovery and first-week nudges are great candidates for personalization. The goal is to get people oriented so they can participate in the commons, not to build them a private commons of their own. The difference matters.

Third: watch for the fragmentation signal. It shows up when members start struggling to reference shared events. You’ll know it’s happening when someone says “wait, that was posted? I never saw that.” It shows up when subgroups stop being able to talk to each other because they’ve been optimized into parallel universes. It’s subtle at first. By the time it’s obvious, you’ve got a cohesion problem that’s hard to reverse.

Fourth: think about what your community needs to remember together. Memory is underrated as a community mechanic. Shared history - the thing that happened, the conversation that changed something, the moment the community showed up for itself - is what turns a platform into a place. Personalization can help individuals find their people, but it can’t manufacture collective memory. That has to be built in the open, where everyone can see it.

None of this means personalization is the enemy. It’s one of the best tools we have for reducing the friction that keeps people from ever finding their footing in the first place. I’d rather have a thoughtful, context-aware onboarding flow than throw a new member into the deep end and hope for the best.

And therefore, the question isn’t whether to personalize. It’s where and for what purpose.

Personalize to help people find the community. Don’t personalize the community itself out of existence.

Supposedly, at least. And if they do show up, they’ll do something (play baseball!?!). Ask questions, share wins, argue about the right way to onboard a new hire, post the occasional meme that somehow perfectly captures a Tuesday.

Participation is the whole darn point. It’s what separates a community from a mailing list. But what happens when participation becomes the input we automate rather than the output we cultivate?

Answer: We get communities that look busy and feel empty.

You’ve almost certainly seen it on LinkedIn. Maybe you’ve stopped posting there because of it (I flirt with the idea weekly). Our feeds have a particular slimy texture now - a dude in a tech bro uniform sharing a “hot take” that is aggressively tepid, followed by seventeen replies that all say some version of “this is so important, thanks for sharing.” Nobody is actually talking to each other. They’re performing engagement at each other. The replies don’t reference the post in any meaningful way. The post doesn’t respond to the replies. It’s participation as theater, automated or half-automated, optimized for the appearance of a thriving network all for the sake of reach and influence.

Reddit has a different flavor of the same problem. There are entire subreddits now seeded with AI-generated “stories” - first-person confessionals, AITA posts, dramatic workplace sagas - written not because someone had an experience worth sharing, but because the content generates upvotes and therefore karma farming and therefore reach. The karma system, originally designed to surface quality, has become the thing being gamed. You end up with a front page full of structurally perfect content that nobody actually lived.

It almost makes me miss the good ol’ days of social media. (Of course one could argue the 2009-2010 Farmville craze on Facebook was also not actually lived, unless y’all are farmers and I didn’t know about it.)

So what is actually bothering us community minded folks so fiercely? Well, simply put: The failure mode isn’t fake content. It’s the erosion of legible signal.

What dies first isn’t authenticity in some abstract sense. It’s usefulness. When you can’t tell if the ten replies praising your question are written by people who’ve actually wrestled with the problem, you stop trusting the answers. When the top post in a community is there because it was optimized to be there rather than because it resonated, you lose the ability to read the room. The crowd stops telling you anything. Even worse is when YOU are throwing content into the ether… and don’t care about trust, usefulness, or reading the room.

And this matters for community practitioners because the behavior doesn’t stay on LinkedIn and Reddit. It migrates and normalizes like a bad highly contagious virus (this is a good example of when a behavior has RO > 1… and we don’t want it to).

But why? If we don’t want to see it ourselves, why the heck are we emulating it?

First, because the people who are members of your community are already living in these environments. They’re learning - unconsciously, habitually - that participation can be low-effort, performative, and consequence-free. They’re importing those norms. Your thoughtfully structured discussion forum starts getting replies that are the community equivalent of a thumbs-up emoji rendered as a paragraph.

Second, because automation tools are getting better and cheaper, and community platforms are not immune. AI writing assistants can help members “respond faster”. The gap between “a tool that helps me articulate my thought” and “a tool that posts on my behalf while I do something else” is smaller than we’d like to pretend. (I say this as someone who has absolutely let an AI tool smooth out a reply I was too tired to write well. I’m not outside this problem.)

Third - and this is the one I think about most - automated participation creates a specific kind of ghost town: one that still has lights on. Vanity metrics stay healthy. Post counts, reply counts, maybe even DAUs look fine. But the qualitative signal - you know, the sense of whether people are actually in it - quietly degrades. By the time the metrics catch up, the community has already hollowed out.

What does this mean for how you manage?

It means presence detection matters more than participation counting. Are people responding to each other, or just posting into the void? Threaded conversation depth, direct replies, members who appear in someone else’s post rather than only their own… these tell you more than total reply volume.

It means friction (in the right places) is protective. A small barrier to posting such as a format requirement, a prompt that asks for specificity, a norm that replies should reference what they’re replying to filters out the low-effort automated churn without suppressing genuine participation. Not all friction is bad friction.

And it means you should probably get more deliberate about modeling what good participation looks like. Not as a policy document. As behavior… yours, your team’s, your community champions’. Humans still imitate humans. For now, anyway.

LinkedIn and Reddit aren’t cautionary tales about them. They’re early readings on a trend that’s coming for every community that touches the internet, which is all of them.

The crowd hasn’t disappeared. But you’re going to need to get better at knowing when it’s actually there.

But there’s another kind of failure that’s a heck of a lot quieter, more expensive, and honestly more embarrassing: the community that was loved to death. Engineered and orchestrated within an inch of its life. So carefully constructed that the humans inside it never quite had room to breathe.

But we don’t talk about that one as much. Probably because it requires admitting that we - the architects, the builders, the strategists, the people who genuinely care - can also be the problem.

Think of it this way. Imagine a dinner party.

Under-design looks like: no host, no food, a group chat that just says “my place, 7ish.” Some people show up. Nobody’s sure where to sit. The conversation is either great or excruciating, and there’s no in-between. A few people have a transformative night (mostly because they found the host’s dog and have had some quality canine time); most quietly slip out before 9 (with the excuse of needing to feed their dog).

Over-design looks like: assigned seating, a printed agenda, a Slack channel for pre-dinner conversation organized by topic, and a facilitated icebreaker where everyone shares one professional win and one personal growth area. The food is excellent. By 8pm, everyone is behaving beautifully… but feeling vaguely suffocated and as if their dinner jacket is a straight jacket.

The failure mode isn’t too much or too little effort. It’s effort applied without understanding what the community actually needs to feel alive.

In the end, both parties fail spectacularly… just differently. The under-designed one can’t retain anyone. The over-designed one can’t keep anyone real.

So what does this mean in practice, for those of us who spend our days thinking about this stuff?

The under-designed community fails at the threshold. People arrive and can’t find a foothold. No clear reason to post, no sense of who else is here or why, no signal that their participation matters. The intervention needed is structural: better onboarding, clearer purpose, a visible core of active members who model what “being here” looks like. This is the kind of failure community folks are trained to fix, which is why it gets most of the airtime. In fact, it’s so talked about that this paragraph is literally all I’m going to say about it.

But the over-designed community?

The over-designed community fails further in, and that delay is what makes it dangerous. Engagement metrics often look fine (if not fantastic) for a while. Members are completing the onboarding flow. They’re posting in the right channels. They’re responding to the prompts. And then, gradually, a kind of managed listlessness sets in. The community becomes a place where people perform participation rather than actually participate. You’ll notice it when the posts start to feel like press releases. When no one says anything surprising. When you realize you haven’t seen a genuine argument or original thought in months. (A community with no constructive dissent is not a thriving community.)

This is where you have to resist the instinct to add more structure as the fix. More programming, more content calendars, more themed weeks. If over-design is the disease, more design is not the cure. It would actually be a death sentence.

The harder question (and admittedly I’m still working on perfecting this one myself) is how to create enough structure that people can find each other and orient themselves, while leaving enough genuine openness that something unexpected can happen. That unexpected thing is the whole point. It’s what makes a community different from a newsletter or a course or a very organized Slack workspace.

There’s also a power dynamic worth naming here. Over-designed communities often reflect the anxieties of their builders more than the needs of their members. We fill the space because we’re afraid of what the silence means. We add a feature because we want to demonstrate value to our stakeholders. We launch another initiative because last quarter’s numbers were flat. The community becomes a canvas for our own professional discomfort, dressed up as member benefit.

And I know that’s not a comfortable thing to sit with. But it’s worth sitting with.

The practical upshot is this: when a community is struggling, the first diagnosis should not be “we need more.” It should be a genuine inquiry into which kind of failure you’re looking at. Is the space too sparse to be habitable? Or is it so curated that real humans have stopped showing up, even if their login activity says otherwise?

You can usually tell by asking one question: When’s the last time something happened here that surprised you?

If the answer is “never… things are very predictable,” you’ve got one problem. If the answer is “I have no idea, I don’t think anyone’s in there,” you’ve got a different one.

Both are solvable. But not with the same solution.

So, go leave some space in your community this week. See what fills it.

Meanwhile, somewhere in the foundation, a very important and very load bearing support beam just started to snap.

Things are about to start getting… interesting.

Why? Well because governance, rules, guidelines, policies, process - whatever you want to call them or whatever “legislature” type you want to lean on - in mature communities are not for decoration.

Platform governance provides structural assumptions about how power flows, how safety feels, and how status is earned. They determine who gets heard, who gets downvoted into oblivion, who moderates for free, and who quietly leaves. By the time a rule feels “normal” it is already load-bearing.

And yet, every so often someone decides the beams are a little inconvenient. A little too restrictive for that open-concept kitchen.

For example, when Twitter shifted verification from a status signal tied to identity into a purchasable badge, that was framed as democratization. In practice, it detonated the platform’s signaling system. Visibility, credibility, and impersonation risk all got scrambled at once. It wasn’t a feature update. It was a power redistribution event disguised as a subscription model. And who knows, maybe that was intentional. The platform now known as X soon eliminated most other trust and safety elements further degrading any trust the majority of legitimate users with human emotions. It’s now a cess pool.

Similarly, when Reddit altered API access in ways that fundamentally changed how moderators and third-party developers operated, the narrative focused on business sustainability. The ecosystem, however, experienced something else entirely. Volunteer moderators who had built governance infrastructure for years suddenly realized the ground rules of their labor had shifted. That is not a minor tweak. That is rewriting the social contract while pretending it is a billing update.

Communities can survive unpopular decisions. They cannot survive incoherent ones.

I think a lot of times we equate trust in a community with an almost affectionate perspective. But it’s not that at all. It’s predictability. It’s the quiet agreement that if I behave within known norms, I can anticipate outcomes. I understand how visibility works. I understand how conflict resolves. I understand how effort translates into status or impact.

And the reality is that most of us aren’t looking at the rules because we’re worried about breaking them. We feel empowered by rules knowing that they protect us from an ecosystem and culture we don’t want to be involved in or sink time into.

When those mechanics change without clear alignment to actual user pain, people do not thoughtfully debate it. They adapt defensively. They reduce investment. They conserve energy. They start treating the platform as unstable terrain.

And unstable terrain does not attract long-term builders.

Now compare that with rule evolution that feels annoyingly reasonable. On Substack, most major changes have reinforced the same central promise: creators own their audience, control monetization, and operate with autonomy. You can argue with individual decisions. But the direction has been legible. The system’s underlying logic has remained intact. That coherence buys patience, loyalty from long time users, and willingness to try from new users (oh hello, that would be me).

In strong private communities, the same pattern shows up. A real problem is named. The change is explicitly tied to that problem. The consequences are acknowledged instead of minimized. Leadership does not pretend nothing structural is happening. They explain what is moving and why.

It is not flashy. It is not revolutionary.

It is boringly competent governance.

So, how do you avoid burning down the house?

Most catastrophic rule changes share a few predictable traits.

• They optimize for metrics without understanding the fragile equilibrium those metrics reflect.

• They redistribute power while insisting power is unchanged (although 99% of the time I’d argue leadership doesn’t even know who is being impacted)

• They ignore second-order effects because modeling social cascades is inconvenient and spreadsheets are comforting.

Like we’ve talked about before, communities are ecosystems, not dashboards. Pull one incentive and ten behaviors shift. Change one visibility rule and an entire informal hierarchy reorganizes itself overnight. Pretending otherwise is not bold agile leadership. It is magical unicorn and rainbows thinking with a press release.

If you want to change rules without destabilizing trust, you have to treat them like infrastructure. Ask who built identity around them. Ask what invisible labor depends on them. Ask which status systems collapse if they disappear. Ask which new elites or predators emerge if you move them.

And then, crucially, say that part out loud.

Communities rarely implode because a rule changed. They implode because the meaning of participation changed without consent. When people realize the game they have been playing has new scoring mechanics, new referees, and a new prize structure, they aren’t going to be all “yay change!”. They reassess whether they want to play at all.

Rule changes that align with lived user problems feel like maintenance. Rule changes that ignore those problems feel like regime change.

One preserves trust. The other trends for a week and destabilizes culture for years.

This was always a bit misguided and felt a bit… gross? But in the age of AI, it is technically incorrect.

Now please excuse me while I nerd out for a moment, mmkay? I just spent far too many hours spelunking through arXiv (iykyk) for this to just reside in my brain and never see the light of day.

Modern AI systems, especially large language models, don’t operate like databases searching for the perfect match (or even a fuzzy match, fellow Excel lovers). They aren’t retrieving a single correct answer from a neatly labeled library drawer. They are pattern-recognition systems trained to predict likely continuations based on enormous amounts of text. What they learn is not facts in isolation, but statistical relationships between concepts, language, context, and outcomes. In this system, repetition is not noise but signal strength.

When the same idea appears many times, across different contexts, phrased in different ways, attached to different problems, the model doesn’t see the duplication we see. It sees reinforcement (and gosh - it <3’s statistical reinforcement). Variation tells the model what is essential versus incidental. Disagreement tells it where boundaries and tradeoffs exist. Context tells it when an idea applies and when it does not. A single pristine answer provides almost none of that (other than maybe a hallucination).

This is why aggressively deduplicated communities produce brittle AI outputs. If knowledge only exists in one canonical thread, the model learns a thin representation. It knows that something is true, but not how, why, or when it stops being true. The result is confident, generic answers that collapse under edge cases, which is exactly the failure mode people keep blaming on AI instead of on the training signal.

So yes: I am somewhat blaming hallucinations on perfection. Ironic, huh?

Intentional redundancy creates a much richer learning environment. From a technical perspective, repeated explanations increase “token co-occurrence frequency” across diverse semantic neighborhoods. That matters. Models build internal representations by clustering related concepts based on how often they appear together and in what forms. When an idea shows up in a YouTube video, a troubleshooting comment, a half-baked conversation, and a debate thread, the model learns the shape of the idea, not just its label.

Disagreement is especially valuable. When two answers diverge slightly, the model does not see confusion. It sees conditionality. This is how models learn phrases like “it depends,” not as a shrug, but as a structured response tied to context. Communities that suppress debate in favor of a single “correct” answer remove that gradient entirely. You get certainty where nuance should exist.

Humans benefit from this for the same reason. People do not learn from optimal explanations. They learn from explanations that match their mental model at that moment. Redundancy allows knowledge to meet people where they are cognitively, emotionally, and situationally. AI systems trained on that same redundancy inherit that flexibility.

Head hurting yet? Honestly mine too.

Let’s kind of explain it this way. If I was to ask you what color a barn, a church, and a car are - there’s a solid chance you could actually answer the first two with ease (in the US at least) but the third you’d struggle with. Why? Because in passive observation, we’ve gathered enough relational data to have ascertained that barns are usually painted red (that’s a whole other story I can tell) and churches are white (note to self: Google the heck out of this later). But we’ve observed cars of nearly every color, and therefore can’t determine a singular answer because there is no ascertainable pattern.

Now imagine I took a survey of all the barns and churches in New Hampshire (where I live and the majority of our structures do follow the red and white schema) and entered them into a database with “structure: barn OR church” and “color: [observed color]” for each entry. I now searched the database for the first occurrence of barn to determine the color of barns… and it’s blue. Database says barns are blue.

This is the most basic difference between co-occurrence token frequency and a database.

And sorry. Your head probably hurts more now and you’re questioning barns and church colors.

Back to the nerding…

There is also a temporal dimension that matters technically. Models are trained on snapshots of the world. Knowledge that appears once and then goes quiet looks static. Knowledge that reappears over time, with small shifts in framing, reflects evolution. This helps downstream systems infer which ideas are stable and which are drifting. Over-curated communities erase that signal. They look frozen even when reality is changing.

From an information theory perspective, redundancy increases robustness. Single points of truth are single points of failure. If the “best” thread becomes outdated, biased, or incomplete, everything downstream inherits that flaw. Redundant representations create fault tolerance. Errors are easier to detect because inconsistencies become visible. This is true in distributed systems, biological systems, and yes… knowledge systems feeding AI.

None of this means communities should drown in sheer and utter chaos. The goal is not endless repetition with no structure. The goal is guided redundancy. Strong linking instead of forced merging. Surfacing related discussions without shutting new ones down. Treating repeated questions as high-confidence signals about where understanding is fragile or incomplete. Designing systems that allow ideas to echo while still making navigation humane. And introducing disagreement where destabilization needs to occur.

The uncomfortable truth is that the “single source of truth” obsession was never about learning. It was about control and cleanliness because it felt good and made us look neat and tidy. But AI exposes that tradeoff brutally and without mercy. Clean knowledge is easy to manage but hard to reason with. Messy knowledge is harder to moderate and far more useful to both humans and machines.

So, if communities want to remain relevant in an AI-mediated world (hint: they do!!!), they need to stop optimizing for elegance and start optimizing for chaotic resilience. Redundancy is not waste. It is how intelligence - human or artificial - actually forms. And treating it as infrastructure instead of clutter is no longer a philosophical stance. It is a technical requirement.

And this makes sense. These are visible signals. They are easy to screenshot and easy to explain. They fit neatly into updates and dashboards. They let us answer the question, “Did this work?” without too much philosophical trouble.

But this model breaks down almost immediately when the thing you are introducing is not content, but behavior.

A new norm.
A new way of asking questions.
A different posture toward disagreement.
A subtle shift in who speaks first, or who responds, or who feels allowed to.

At that point, treating adoption like feature usage starts to feel off. Because what you are really dealing with is not activation. It is transmission.

This is where the idea of R0 becomes useful. Admittedly R0 is wildly on my brain right now as I sit next to a pile of used tissues and keep trying to will my sinuses to stop terrorizing my face. I caught whatever this blasted thing is from my husband who caught it from someone at work. I’d really like to know from who… and how many other folks did they get sick. I did not volunteer to be part of this, but alas. sniffles

In epidemiology, R0 is the basic reproduction number. It describes how many additional people one infected person will pass a disease on to, on average assuming “normal” conditions. If the number is above one, the behavior spreads. If it is below one, it quietly disappears, no matter how bad the initial case was or what happened to the initially infected person.

The important thing is this: R0 is not about how severe the disease is. It is about how easily it moves between people under real conditions.

Community ideas and behavior work the same way.

You can introduce a beautifully articulated principle. You can model it perfectly yourself. You can even get a handful of people to engage with it directly. None of that guarantees it will replicate. It could be the most short lived virus ever… or the most viral thing you’ll ever do.

The community oriented question is basically the same: When one member encounters a new idea, are they likely to pass it on through their own behavior?

Not explain it.
Not praise it.
Not like it.

Reproduce it.

From a design perspective, this shifts where the risk actually lives which kind of tosses our usual mitigation strategies in with the used tissues. The danger is not that people disagree with the idea. The danger is that it requires too darn much effort, confidence, or social capital to repeat. If it only works when performed by moderators, leaders, or the original author, its R0 is effectively zero.

From a behavioral perspective, R0 lives in friction. How easy is it to do the thing once you have seen it done? Does it feel safe to imitate? Does it fit inside existing conversational habits, or does it demand a personality transplant?

From a measurement perspective, this is deeply annoying to us, data scientists, and anyone who loves data. Transmission is harder to see than participation. You are looking for second-order effects while steering clear of correlation without causation. We’re keenly looking for the language someone else adopts. The tone that shows up without prompting. The pattern that appears where you did not seed it.

This is often when practitioners feel uneasy. Because your influence becomes less legible at the same moment it becomes more real. You can no longer point to a single post or program and say, “That was the thing I did. It worked.” The work has moved into the background, where attribution gets fuzzy and credit gets scarce. We either sound like mad scientists or wizards trying to explain what has just happened and why it’s the best work we’ve done yet.

A community idea with a high R0 does not look dramatic. It looks boring in the best possible way. It blends in. It stops needing explanation. People do it without remembering where they learned it.

And once that happens, it is very hard to reverse. Not because you enforced it, but because your little virus learned how to survive on its own. Go virus, go!

Now back to sniffling.

You don’t yet know how chaotic it will get or how long it will last; only that community is about to feel it. Again. And somehow, more intensely than anywhere else.

When this happens, it is tempting to frame the problem as community’s fault. We attract negativity. We’re too lenient. Our code of conduct isn’t strong enough. We should have prepared better. Jeez, did we design this mess? It looks and feels like a behavior problem of our own creation!

However, it rarely is. You can stop the self-loathing for now. (Do CM’s actually know how to not self-loathe? Asking for a friend.)

If your community suddenly feels louder but less helpful, chances are you do not have a community problem. You probably have a stage. A highly visible one at that, conveniently positioned exactly where pressure naturally flows when other paths stop working.

Great, huh?

This shift almost never begins inside the community itself. It starts elsewhere in the organization, when people learn through experience that quieter channels are unreliable or too echo-y. Support feels slow or opaque. Feedback mechanisms feel ceremonial black holes. Escalation paths exist in theory but not in practice. Over time, people stop trusting process and start trusting visibility. They look for the surface most likely to acknowledge them. And community is often that surface (“stupid people skills and stupid empathy” mutters the burned out community professional before giving another chunk of their soul to a customer).

A truly healthy collaborative community rests on a simple belief: if I show up honestly, something useful will happen.

An unhealthy performative community operates on a different assumption: if I am visible enough, someone will have to respond.

At a glance, these two spaces can look remarkably similar. Both are active. Both involve posting, commenting, and engagement.

But, the difference lies in what those behaviors are optimized for and what they quietly replace.

One of the earliest signals of a ‘stage’ shows up in how questions are asked. Instead of opening space for exploration or learning, questions begin to arrive preloaded with conclusions.

“Why haven’t you released XYZ yet?”

“I already tried XYZ.”

“Don’t send me to XYZ.”

They are framed not to invite help, but to make a point. The question format becomes a delivery mechanism rather than a request for understanding. This is not bad faith. It is adaptation. When asking privately has not worked, people learn to ask publicly, and to ask in ways that travel.

Feedback changes shape in similar ways. Rather than being specific, directional, or grounded in concrete tradeoffs, it becomes abstract and broadly agreeable. The more generalized the feedback, the easier it is for others to rally around it. Agreement accumulates quickly, but resolution does not. Public validation begins to stand in for progress, even when nothing actually moves. If they can’t get help, they can at least get acknowledgement from peers.

Even help starts to feel different. Responses skew warm and affirming, but rarely go deeper. There are fewer clarifying questions, fewer attempts to iterate, and less appetite for productive friction. Helping becomes a visible signal of alignment rather than a collaborative effort to solve something together. Being seen responding matters more than doing the work. And trust me; we want to put out the fires, but there’s just too darn many.

Over time, the language of the space shifts. Phrases associated with escalation and accountability creep into peer conversations. Calls for ownership replace offers of contribution. The tone becomes more declarative and less exploratory. Activity may increase, but trust thins. The community feels busy, yet brittle, like it is holding more than it was designed to carry.

This is where teams often misread the situation. When a community becomes a stage, it is rarely the root problem. It is a symptom. It signals a broader system in which visibility has become safer or more effective than process. People are not performing because they want attention. They are performing because they believe it is the only reliable way to be heard. The community did not create this behavior. It inherited it.

Yayyy…. crawls under desk and starts sobbing.

A community can absorb that pressure for a while, but it cannot resolve it on its own. When it becomes the primary place where people seek acknowledgment, it gradually loses its capacity for collaboration. The work is not to quiet the stage or discipline the audience. The work is to understand why the stage exists in the first place.

In a healthy system, community is not where everything lands. It is where people come to think together, not because it is the loudest option available, but because it is the most constructive one.

Fan-freaking-tastic. What now? What if we don’t want to be the channel with the most engagement in ALL CAPS BECAUSE PEOPLE ARE REALLY ANGRY AND NOT AMUSED.

First up is realizing that most organizations do not intend to create performative behavior. In fact, many explicitly say they want thoughtful feedback, early signals, and honest dialogue. What they underestimate is how quickly people learn which behaviors actually get acknowledged and which quietly disappear.

Humans, man. Observant little things, eh?

Training does not happen through policy statements or values decks. It happens through response patterns. People watch what gets answered, how long it takes, and under what conditions. Over time (sometimes a very short time that lasts as long as a Google search), they adapt accordingly.

One of the most common training signals is opacity. When feedback enters systems that provide no visible acknowledgment or follow-up, people cannot tell whether it was read, considered, or ignored. In the absence of feedback about feedback, visibility becomes the proxy for impact. And if private channels feel like a void, public ones start to feel safer, even when they are noisier. It’s kind of the same dynamic that a lot of people feel safer in crowded places.

Speed reinforces this dynamic. When public posts receive faster responses than formal channels, people learn that urgency is manufactured through exposure. The lesson is subtle but powerful. If you want movement, make it visible. If you want accountability, attach an audience. Process begins to feel slow and optional, while performance feels efficient.

And, inconsistency makes the effect worse. When some issues raised through formal channels are addressed while others are not, people stop trusting the system as a whole. They look for patterns instead. Often, the only reliable pattern they can find is visibility. What is seen gets addressed. What is quiet waits indefinitely.

Organizations also unintentionally train performativity through abstraction. Broad feedback is easier to agree with publicly and harder to resolve privately. When nuanced, specific input leads to silence or delay, while generalized sentiment sparks discussion, people learn to flatten their feedback. Precision gives way to resonance. Agreement replaces progress.

Recognition systems play a role as well. If everything is smoldering, but that one thing looks extra on fire to community visitors and is gaining a lot of traction - chances are you’re going to want to put that out first. But are you rewarding undesired behavior with desired outcomes?

None of this requires bad intent. It is the predictable outcome of systems that prioritize manageability over legibility. When process is hard to see and outcomes are hard to trace, people reach for the only lever they can trust. They make noise.

Over time, this behavior spreads. New participants learn from observation, not instruction. They see what works and replicate it. Performance becomes the dominant participation mode, not because it is preferred, but because it is modeled and reinforced.

The irony is that organizations often respond by trying to quiet the noise. They ask for more structure, more discipline, more restraint. But performative behavior is not solved by suppression. It is solved by redesign.

People stop performing when they trust that quieter signals will be acknowledged, that specificity will be rewarded, and that contribution will lead somewhere predictable. Until then, visibility will continue to feel like the safest strategy available.

If organizations want collaborative communities, they must first build collaborative systems, embrace collaborative processes, and take accountability for where things land. Otherwise, the stage will keep reappearing, no matter where people are told to stand.

You log in (if needed). You poke around. You see what is new. Maybe you contribute. Maybe you do not. Whatever the case, there is a modicum of intentionally in the act. You chose to seek out that community. You chose to look around. You chose to contribute.

And in that way, I’d consider them a destination. Optional. Intentional. Easy to forget. Just yet another tab or bookmark.

But, infrastructure works very differently.

Have you ever woken up or gone about your day thinking:

• Today, I use electricity!

• I visited the local highway system.

• I choose to leverage the most excellent cell phone towers.

Probably not.

But have you ever woken up or gone about your day and thought:

• Why the feck is the power out!?

• They really need to do something about this highway traffic.

• The reception around here is terrible.

You notice infrastructure only when it breaks, usually while you are already annoyed about something else.

And y’all… communities are quietly heading in that direction. I’d actually argue a lot of communities are already there. But our strategies and collective knowledge? Not quite.

And it’s not because communities are getting bigger or trendier, but because people are running out of patience for places that require extra effort just to be useful. The more work becomes distributed and asynchronous, the less tolerance there is for spaces that demand attention instead of supporting it.

Destinations ask members to show up. Infrastructure shows up for members.

And as someone who has been on this community journey long enough to have run a few destination communities - and now solidly in the land of “I am an infrastructure architect”, this is an uncomfortable transition for many community teams. Destinations reward visibility. Infrastructure rewards reliability. Destinations feel successful when things are lively. Infrastructure feels successful when nothing remarkable happens at all.

If that sounds underwhelming, it is worth remembering that nobody writes glowing status updates about how well the plumbing performed today (ie, “That toilet flushes so spectacularly!”). And yet, here we are.

Note to self: Go leave a review somewhere on the plumbing just to challenge the system.

You can usually tell which direction a community is leaning by how it integrates into daily life.

• If it requires a dedicated visit, a mental context switch, and a small pep talk to open the tab, it is still a destination.

• If it quietly solves problems, surfaces knowledge at the right moment, or connects the right people without fanfare, it is starting to behave like infrastructure.

Needless to say (gosh, I hope; you knew we were at this point… right?) this shift changes how communities should be designed.

Destinations are optimized for novelty and energy and exploration. Infrastructure is optimized for consistency and load and efficiency. A community that acts as infrastructure has to work on slow days, busy days, and chaotic days without needing to be rebuilt each time. It has to handle spikes without drama and lulls without existential crisis. (And yes, that means the community staff has to be able to handle it without an existential crisis. I’m still working on that one myself.)

It also changes how success is measured.

You do not measure infrastructure by excitement. You measure it by dependency. If the community disappeared tomorrow, what would break? What knowledge would be lost? What relationships would suddenly become much harder to maintain?

And, that is not a dashboard most teams are used to. Preparing for this future means getting comfortable with a few darn humbling realities.

First, the community itself may become less visible even as it becomes more important. When something works reliably, people stop talking about it. This can feel unsettling if you are used to proving value through activity and people saying nice things about you. But obscurity is often the price of usefulness.

Second, events become reinforcements rather than lifelines. They matter, but they no longer carry the entire system on their backs. Think of them less as fireworks and more as routine maintenance.

Third, governance stops being a background concern. Infrastructure fails when no one is clearly responsible for it. Clear stewardship, transparent decisions, and boring but necessary rules suddenly become signs of maturity rather than bureaucracy. Suddenly things like uptime, moderation, escalation procedures, and release readiness are your world.

The communities that endure will not feel like destinations you remember to visit. They will feel like systems you assume are there. Quietly holding shared artifacts. Carrying context forward. Making coordination possible without fanfare. When they work well, they will be easy to overlook and deeply missed the moment they are gone. That is the tradeoff of infrastructure. Less applause. More reliance. And if that feels like a loss, it is worth remembering that the most important structures in our lives are rarely the ones we admire. They are the ones we build everything else on top of.

Go thank your favorite cell phone tower!

And finally, when the lights dim and music or opening lines arrive, the sound does not bounce chaotically or fight for dominance. It lands exactly where it should. Every note has somewhere to go.

Now think about an overly loud restaurant. Hard surfaces everywhere. No sound dampening. Conversations overlap, voices compete, laughter ricochets off the walls just the same as an argument. Nothing is technically wrong with the people inside; in fact some may have just come from an astounding performance leaving them excited and ready to chat about what they listened to. But… the problem is that the space was never designed to hold that much noise.

Both rooms are full of sound, but only one was built to make it meaningful. And communities face the same choice.

Disagreement, critique, and tension are forms of noise. They are inevitable in any space where people care. The question is not whether noise will exist, but whether the environment knows how to carry it.

Constructive dissent behaves like music in an opera house. It is complex, sometimes loud, sometimes uncomfortable, but shaped by the architecture around it. There are cues for when to enter and when to pause. Shared expectations about what belongs on the stage and what belongs backstage. The result is intensity without unintentional chaos.

Corrosive conflict feels like the noisy restaurant. Everyone is talking, but no one is really being heard. The echo chamber amplifies the sharpest sounds and drowns out everything else. Conversations collapse into repetition. People leave hoarse and frustrated, not because they disagreed, but because the room exhausted them.

As for communities? Every community eventually encounters disagreement. It comes in many forms - members disagreeing with each other, members disagreeing with community staff, members disagreeing with company decisions, or even the company disagreeing with customer feedback.

The difference shows up early if you know what to listen for.

In spaces designed to handle noise, disagreement deepens the conversation. Ideas evolve as they move through the room. People build on one another, even when they oppose each other. There is a sense that friction is part of the experience, not a disruption to it.

In spaces without thoughtful sound design, dissent escalates instead of resolving. The same arguments circle endlessly because there is no structure to absorb them. Tone sharpens because volume becomes the only way to be noticed. What feels like hostility is often just amplification without containment.

But all too often, leaders often respond by asking people to lower their voices or for staff to remove the noisiest.

That rarely works. You cannot politely request acoustics into existence.

Opera houses do not rely on better behaved audiences. They rely on intentional design. Balconies, materials, sight lines, boundaries between performers, orchestra, and crowd. These elements do not reduce sound. They make it intelligible.

Communities that handle conflict well do the same. They create dedicated spaces for dissent (ideation spaces, feedback flows, etc.). They establish rhythms that tell people when critique is invited and how it will be engaged. They model how disagreement sounds when it is contributing rather than competing.

The goal is not to make communities quieter.

It is to design them so that when noise does appear - and. it. will - it becomes part of the composition rather than a reason to leave the room.

But alas, persuasion is often an uncomfortable word in public-facing work.

Persuasion is often lumped together with manipulation, tracking, subliminal messaging, and growth tactics that treat people as inputs to be optimized, controlled, and swayed. As a result, many community builders avoid the topic altogether. They insist they are not persuading or manipulating anyone. They are simply providing a space and letting people decide how to engage.

That posture sounds perfectly lovely and ethical and full of unicorns and rainbows, but it ignores a basic truth about design…

Every space - physical and digital - persuades.

The only question is whether it does so transparently and in service of the people inside it, or quietly and in service of someone else’s goals.

mic drop

Wait, wait. I still need that. picks up mic again

Belonging does not emerge because people are convinced to stay. It emerges when the environment makes participation feel legible, safe, and meaningful. The persuasion at work is subtle, structural, and largely invisible to the person experiencing it.

So how exactly does one use persuasion for good? By getting a solid grasp on the mechanics.

The first and most ethical persuasion mechanic is clarity.

People feel a sense of belonging when they understand what kind of space they are in and what is expected of them. Clear purpose, clear boundaries, and clear norms reduce anxiety. They answer the unspoken questions every newcomer carries: What is this place for? What do people do here? What happens if I get it wrong? Am I going to be shunned? Shamed? Helped?

Clarity persuades by lowering cognitive load. It invites participation without pressure. There is no trick involved. The design simply removes unnecessary friction that would otherwise signal exclusion.

Another powerful mechanic is modeling.

Belonging is rarely taught explicitly. It is learned through observation. When newcomers see how disagreement is handled, how curiosity is rewarded, and how generosity is acknowledged, they infer what kind of behavior will be met with acceptance. This is persuasion through example, not instruction.

When we see others modeling behavior that gets the outcome we desire for ourselves, we feel confident in exhibiting the behavior ourselves.

Consistency is equally important.

When responses are predictable and norms are applied evenly, people develop trust in the system rather than in individual personalities. They do not have to perform or guess which version of the rules applies today. This predictability persuades people that the space is stable enough to invest in emotionally knowing that you won’t get verbally skewered this one oddball time.

In contrast, inconsistent enforcement and shifting expectations create hypervigilance. People participate less not because they lack interest, but because they cannot assess the risk. And when we can’t assess risk, we get skittish. Is engaging worth the potential conflict, awkwardness, or hostility?

Reciprocity also plays a role, but only when it is unforced.

When contributions are acknowledged in proportionate and genuine ways, people feel seen rather than leveraged. The difference between ethical persuasion and manipulation often shows up here. Manipulation creates obligation. Ethical persuasion creates appreciation. One closes a loop by demanding something in return. The other closes the loop by recognizing what was already given.

Finally, there is agency.

The strongest sense of belonging comes from having meaningful choice. Spaces that allow people to shape their experience, influence norms, or contribute to the evolution of the community persuade through ownership. People are not convinced to belong; they just recognize themselves in the structure and that’s shockingly motivating.

Manipulative systems remove agency while offering the illusion of choice. Ethical systems make agency visible and consequential.

The common thread across all of these mechanics is that none of them hide their intent. They do not rely on scarcity, urgency, or emotional exploitation. They work by aligning the structure of the environment with the lived experience of the people inside it.

So… Are you designing a choice or a trap?

You pour people in, add a few rules, maybe decorate the edges with features, and hope something tasty emerges. Maybe a cake? When it does not or becomes a festering petri dish, we blame the people. They are disengaged. They are toxic. They are lurking instead of contributing.

But containers are passive, non (intentionally) living. Ecosystems are not passive and very (intentionally) living.

An ecosystem is a living arrangement of relationships, feedback loops, and constraints that quietly shape what can survive there. Nothing thrives by accident. Stability is not enforced. It emerges from balance.

If we designed communities like ecosystems, the first shift would be this. We would stop asking how to drive activity and start asking what conditions make certain behaviors inevitable.

Think about a healthy forest. There is no single dominant species shouting instructions (except us; but alas…). Stability comes from diversity of roles that counterbalance one another. Predators prevent overpopulation. Decomposers recycle what would otherwise rot. Keystone species shape the terrain in ways that benefit others without direct coordination.

Communities work the same way, even if we rarely name it.

Every stable community has a few quiet keystone roles. Not the loudest voices or the most visible experts, but the people who translate norms into action. The ones who welcome newcomers by modeling how to behave rather than explaining it. The ones who ask the second question after the obvious one has already been answered. Remove them, and the space does not collapse immediately. It slowly loses coherence before inevitable decay and collapse.

Rituals play the role of seasonal cycles. They give people a shared sense of timing and rhythm. Weekly prompts, recurring showcases, average time to response, predictable moments of reflection. Not because people need novelty, but because repetition creates safety. In nature, seasons signal what is possible now. In communities, rituals do the same. Without them, participation becomes sporadic and exhausting; surprise consumes creative energy. Everything feels optional, which eventually means nothing feels worth showing up for.

Constraints are where most community designers get nervous, but they function like natural boundaries. Riverbanks don’t restrict water. They give it direction. Clear scopes, visible norms, intentional limits on what belongs where. These reduce chaos by reducing decision fatigue. When boundaries are absent, people bring everything. Opinions, grievances, off topic promotions. The system becomes noisy, not because the members are unruly, but because the environment never told them what kind of life it was designed to support.

Chaos enters ecosystems when one species is allowed to dominate unchecked. In communities, this often looks like over rewarding a single behavior that is sharp juxtaposition to what the community needs. Only posting is valued, so people post without listening. Only experts are elevated, so curiosity disappears. Only growth is celebrated, so trust erodes. The platform did not create bad actors. It created incentives that crowded out balance.

Moderation alone can’t fix this, just as culling a single species does not heal a damaged habitat. If the conditions remain the same, the imbalance returns over and over.

This is why so many community teams feel like they are constantly firefighting. They are trying to correct outcomes instead of redesigning the environment that produces them. You cannot ask for generosity in a system that rewards self promotion. You cannot demand civility in a space that amplifies outrage. You cannot cultivate belonging if the architecture only recognizes visibility.

When you see communities as ecosystems, leadership shifts from enforcement to stewardship. The work becomes less about control and more about calibration. Who is thriving here and why. Who cannot survive here and what does that tell us about the design. What behaviors are naturally reinforced by the layout of the space.

The most powerful question stops being how do we get people to act differently. It becomes what kind of environment would make the behavior we want the most natural thing to do.

What gets far less attention are the systems underneath - behavioral, structural, and cultural systems quietly shaping how people participate, disagree, collaborate, and eventually decide whether to stay.

This Substack exists to examine those systems more closely.

In 2026, I’m running a simple experiment: one question per week, for fifty-two weeks to see if it helps me become a better systems thinker who can more easily articulate the complex concepts ricocheting around my nerd brain.

Each post will begin with a question that has shown up repeatedly in real communities I’ve worked with. These are not optimization questions or growth tactics. They are questions about how communities actually function once real humans are involved.

Some of these questions will sound somewhat familiar. Others may feel uncomfortable. For example, when does friction improve a community rather than harm it? How can dissent strengthen a culture instead of destabilizing it? What happens when incentives quietly reward behaviors that leaders claim they want to discourage? And why do some communities feel like well-designed opera houses able to hold complexity, disagreement, and noise while others feel like overly loud restaurants, where sound bounces endlessly and nothing resolves?

This project is not intended to be a playbook or a collection of best practices. It is not a list of tips, templates, or “proven strategies.” If you’re looking for a quick win or a silver bullet, you are not there.

Instead, it is an ongoing examination of community and physical architecture, how rules, incentives, norms, and tools interact with human behavior in ways that are often invisible until they fail. It draws from behavioral science, systems thinking, metrics, and lived experience building and observing communities at scale.

I’m choosing to focus on questions rather than answers because answers tend to age poorly. Contexts change, tools evolve, and yesterday’s solution often becomes today’s constraint. Well-formed questions on the other hand, continue to do useful work. They help surface hidden assumptions, stress-test norms, and create space for more intentional design rather than imitation.

Each week, I’ll take one question and explore what it reveals about behavior, where community design commonly goes wrong, and how leaders, builders, and members might think differently about the systems they are shaping… whether deliberately or by default.

This Substack is for people who find themselves thinking deeply about community, even when they don’t mean to. It’s for those who have wrestled with a rule because they sensed it would change behavior in unexpected ways, or who have felt that a space was misaligned long before the metrics caught up. It’s for anyone who wants better language for understanding why a community feels coherent, resilient, or endlessly fragile.

If that sounds familiar, you’re in the right place. Welcome to The Community Nerd.

Subscribe now