Here is the thread you do not want. A member posts a piece of work in your feedback channel. Fourteen comments in, someone says it looks generated. The original poster says they used a tool for part of it. Two people say that is fine and normal now. One person says they did not pay to have a machine's work critiqued. Somebody quotes a rule you never wrote. And then everyone turns to look at you.
That thread is coming to your community. It has probably already brushed past it. And the worst possible time to decide your position is in the middle of it, in public, with people you like on both sides.
So let us write the policy now, while it is calm and boring and nobody is upset. This takes about ninety minutes, it will save you a bad week, and I promise it is more interesting than it sounds.
Start by refusing to write a policy about AI
Sounds unhelpful. Stay with me.
The mistake nearly everyone makes is writing rules about a technology. Technology rules go stale in months and generate endless edge case arguments. Does a grammar checker count. What about the auto fill in your camera. What about a translation tool. What about the noise removal on a podcast. You will spend your life adjudicating tool taxonomy, and you will lose.
Write rules about the harms you actually care about instead. Underneath every AI argument in a creator community are three separate concerns wearing the same coat, and once you separate them, the policy writes itself.
The first is deception. Someone presents work as something it is not, and other members make decisions based on that impression. Buying, hiring, offering feedback, feeling inadequate.
The second is effort mismatch. In spaces built on peer critique or mutual accountability, people are trading their time. Giving forty minutes of careful feedback on something generated in forty seconds feels like a bad trade, and that resentment is legitimate regardless of anyone's philosophical position on the technology.
The third is volume. Generated content is cheap to produce, so it can flood a space, drown out slower human contributions, and turn a conversation into a stream.
Deception, effort mismatch, volume. Write for those and your policy will still make sense in three years no matter what tools exist.
Decide your community's actual stance, and be honest about it
There are three coherent positions, and all three are defensible. What is not defensible is pretending to be relaxed while quietly disapproving, because your members will feel your disapproval and nobody will know where they stand.
The open position: AI tools are welcome as part of the work, disclosure is expected in specific contexts, and the conversation is about craft and outcomes. This suits communities of marketers, builders, small business operators, and anyone whose members are paid for results.
The restricted position: AI tools are welcome for process work (research, outlines, admin, transcripts) but the work presented for critique or sale in this community must be substantially human made, and we will say what substantially means. This suits craft communities, writing groups, illustration communities, anywhere the point is skill development.
The excluded position: this space is for human made work only, and generated work is off topic here. Perfectly legitimate. It requires you to accept that you will be enforcing something hard to verify, and to be gracious rather than forensic about it.
Pick one. Write it in a single sentence at the top. Members can disagree with a clear position far more comfortably than they can navigate an unclear one.
Write disclosure rules by context, not by percentage
Do not write "you must disclose if more than thirty percent of the work is AI generated." Nobody can measure that and everybody will argue about it.
Write disclosure into the specific moments where it changes a member's decision. There are usually four.
Feedback and critique: say what you used and what you want feedback on. That is it. One line. "Draft was generated, I rewrote the middle section, I want feedback on the structure." Now the person spending forty minutes gets to choose how to spend them, which is the whole point.
Anything for sale: if a member is selling work to other members, or promoting a service, the description should reflect how it was made. This is where deception does actual financial damage.
Portfolio and showcase posts: if the norm in your community is "look what I made," people are calibrating their own progress against what they see. Undisclosed generated work in a showcase channel makes other members feel bad about their own skill for no reason. That is a real harm and it is worth one line of policy.
Advice on serious matters: if someone asks a machine for an answer about money, law, health, or safety and pastes it into your community as an answer, that needs disclosure and probably a caution. Not because the answer is necessarily wrong, but because confident, fluent wrongness is uniquely dangerous in those areas, and a member may act on it.
Notice what is not on that list. Casual conversation. Admin. Someone's rough notes. Do not make people file paperwork to talk in the kitchen.
Handle volume with a limit, not a judgment
If your worry is flooding, the cleanest fix has nothing to do with AI at all. Put a posting norm in place that applies to everybody.
Something like: showcase posts are limited to two a week per member. Or: the feedback channel runs on reciprocity, give one substantive critique for every piece you post. Or: long form posts go in threads rather than as new items.
These rules are fair, easy to enforce, and completely tool agnostic. A person posting eleven generated pieces a day and a person posting eleven human pieces a day are the same problem for the room, and solving it as a volume issue avoids turning your moderation into a witch hunt.
Never accuse, and never use a detector
This is the part I feel strongly about, so let me be plain.
Do not run members' work through detection tools and act on the output. Those tools produce false positives, and a false accusation of dishonesty in a community someone paid to join is a wound that does not heal. You will lose that member, several of their friends, and your reputation as a fair-minded host, and you will deserve to.
If you suspect something, your move is a private, curious, non accusatory message. "Hey, loved this. For the showcase channel we ask people to note their process, would you mind adding a line about how you made it?" Ninety nine times out of a hundred that ends the matter, either with a cheerful disclosure or with a slightly baffled explanation that they drew it by hand over two weeks.
If someone repeatedly and knowingly misrepresents work, that is a dishonesty problem, and you handle it the way you handle any other dishonesty problem: privately, in stages, with a written record of what you said and when. Which means you need your general escalation approach in place, and this policy simply plugs into it rather than inventing a parallel justice system.
Write yours in seven lines
Long policies do not get read. Here is the shape that works, and you can draft it in one sitting.
One line of stance. What this community's position is, plainly.
One line of principle. "We care about honesty, not about which tools you own."
Four lines of context specific disclosure. Critique, selling, showcase, serious advice.
One line of volume norm, applying to everyone.
One line on enforcement tone. "We will ask before we assume, always privately, always first."
That is it. Put it in your guidelines document, pin it once, mention it in onboarding, and then get on with your life. Resist the urge to expand it every time an edge case appears. Answer the edge case in the moment and only change the document if the same case turns up three times.
Say your own position out loud too
One last thing, and it is the one founders skip. Tell your members what you do.
If you use these tools for your outlines, your show notes, your transcript cleanup, your admin, say so. If you do not use them at all in the work you sell, say that. Your members are already wondering, and a founder who quietly does the thing they have implicitly discouraged is a trust problem waiting to detonate.
Being straightforward here also does something lovely: it turns a tense subject into a normal one. Communities where the founder has said "here is exactly what I do and why" tend to have calm, curious, practical conversations about tools. Communities where nobody knows the founder's position tend to have the fourteen comment thread.
Do this this week
Write the seven lines. Do not research first, do not survey your members first, just write what you actually believe and give yourself permission to revise it later.
Post it once, in a normal voice, and add a genuine question at the end: what did I miss. You will get useful answers, and more importantly you will have made the subject discussable before it became a fight.
Then add one line to your onboarding so every new member meets the policy on day one. Ninety minutes of work today, and the thread you are dreading turns into a two reply conversation that ends with somebody saying thanks for clarifying.