I’ve been catching up on community-management content from earlier this year, and one post I keep coming back to is Tany Rios Castro‘s CMX Fellowship Panel review over at the CMX Hub. In June, Tany, a community manager with a focus on events and authentic connection, sat on a panel at TechSoup’s San Francisco office. The Community Managers Connect meetup was hosted by the fabulous Susan Tenby (who I’ve been lucky enough to work with over the years). The panel included Jenna Feldman from Adobe, Joanna Chen from Figma and was moderated by Marius Ciortea, Chief Community Officer at Higher Logic, (Higher Logic also sponsored the event). The panelists had recently attended the CMX Summit and shared what they’d heard and learned at the annual industry event.

In her recap, Rios Castro described how the discussion moved through everyone’s experiences integrating AI into their workflows: review agents, data pipelines, analytics. Then it turned to something more personal, notes with real community members. The DMs, the Slack threads, the guy asking why their favorite feature broke, again. The communications that make up a bulk of a Community Manager’s day. Tany relayed that the group was skeptical of using automated text for those types of messages, the ones meant to be interpersonal, not broadcast: “Real connection lives in the unpolished details—the lowercase style, the hyper-specific memes, the slight typos, and the playful language that signal a real human touch.”

That sentence has been stuck in the back of my head for over a week.

The typo isn’t the charm. It’s the receipt.

Think of a company newsletter, spotless, on-brand, not a comma out of place. You don’t read it carefully. You skim it, because you already know it was written once and sent to two hundred thousand people at the same time.

Now think of a text from a friend: “omw, 5 min, trafic is insaneee! ;)” Typos and all. You read that one immediately, because you know something a newsletter can’t fake: it was written just now, just for you, by someone thinking about you specifically while they were doing something else.

That’s the whole difference. A typo doesn’t build credibility by being cute. It earns trust by being proof of the thing that actually matters: someone’s limited attention was spent on you, right then, instead of being spent once and reused on everyone.

The problem underneath 

Here’s the part of that same panel that worries me more than the writing-style debate. The panelists described advocacy programs, the ones where people apply to represent a brand, were now being flooded with AI-generated admissions essays. Polished. Persuasive. Not real. Some teams have started running AI detection on the applications themselves, screening out fakes to protect the one thing an advocacy program is actually selling: a real person’s voice.

That’s not a hypothetical. That’s a team building a filter to catch a counterfeit of the exact thing the program exists to collect: real people who actually love what they’re applying to represent.

If it’s already happening to applications, it’s worth asking where else it’s happening. Somewhere, someone is waiting on a message from a person they trust to actually be there. Not a broadcast. Not a list. One message, meant for them. Here’s what I can’t answer: at what point does that message arrive polished and exactly on-topic enough that the person reading it can’t tell a human didn’t write it? Maybe it already has, to someone less AI-savvy, or someone with no reason to double-check. I don’t know. I don’t think the panelists did either. They just noticed something like it happening to their own applications.

A test, not a rule

I’m not sure how I really feel about AI, honestly. I have real concerns about water use, about what we lose as a species when the human touch goes out of things. I think creativity is one of the most human things we’ve got, not something a model can actually do, just something it can imitate. Usually poorly.

Here’s where I find myself landing as these tools show up in more and more places in my workflow, at least for now: AI earns its keep anywhere scale matters more than attention. Scanning ten thousand posts for a sentiment shift you’d never catch by hand. Summarizing a week of Discord threads before a stakeholder meeting. Reformatting a spreadsheet. Those feel like jobs for machines.

But a one-on-one reply isn’t an aggregate. It’s not a pattern across ten thousand people, it’s one person, asking you something specific. That’s where this test applies for community managers. Pull your last ten replies to actual members of your community. For each one, ask: what did I actually know about this person when I wrote it? Not “was I nice.” A real detail. That they’d asked the same question three months ago. That they mentioned a move, a new job, a kid’s first day of school. Something only they would recognize.

If you can’t point to that detail for most of them, AI’s not the problem. You’ve just stopped paying close enough attention, and no tool did that to you. The whole reason this job exists is that a person is on the other end of it, not a queue, not a ticket number, a person. That’s on you to protect.

I need to go check my own replies before I automate the wrong one.