The CMJ Group’s AI Principles and Working Guidelines
AI is changing how we work individually, as a team, and with our clients. For us, the goal is not to avoid AI use entirely or to swing in the opposite direction and deploy it everywhere without discernment. The goal is to use it thoughtfully and transparently, in ways that support the kind of work and relationships we want to build.
This document helps us create and share:
internal clarity about a values-aligned approach to how we use AI as a team
external transparency for clients, collaborators, and community members about how we utilize AI
Please feel free to use this guide as a template for your own guidelines and policies with attribution back to The CMJ Group.
Outline
Part 1: Our Core Principles Guiding AI Deployment
Part 2: What We Do NOT Use AI for and What We Do
Part 3: Data Security and Privacy
Our Core Principles Guiding AI Deployment
1. AI never replaces our thinking and cannot accurately replicate our unique viewpoint and bespoke approach, no matter how much data we throw at it. We use AI tools as a support while knowing their limits.
We can use AI to help us move faster, get unstuck, or improve what we create. But it should not replace our ideation process, first or final drafts, or the relational parts of our work. We know it is a statistical probability machine and its dataset is limited, opaque, and can be biased.
2. Transparency always.
We must be forthright about how we use AI, including in:
Internal collaboration: When we share how we use AI, we all learn from each other and benefit. No shame or judgment allowed.
Client-facing materials: We default to disclosing our use of AI in the creation of deliverables, tooling, and ideation. This will create transparency and trust; we are not masking our usage and, in fact, are amplifying our capabilities and demonstrating to clients when AI use is appropriate and when it is not.
Community and public-facing policy language: Outward-facing policies will be finalized and shared to support a culture of transparency in our work.
3. Values over everything.
Our use of AI should never compromise our values, especially:
People Are the Point: We protect human dignity and avoid harm, even (especially) when inconvenient.
What we do NOT use AI for
1. ❌ Personal communication
We should not use AI to talk to each other or clients instead of actually talking to each other. As a general rule, we do not want AI to stand in for human relationships, which require human labor.
This includes not using AI to create:
direct messages
community interactions (DMs or replies)
internal CMJ Slack communication, unless there is a specific values-aligned rationale, and that use is disclosed
2. ❌ Published writing
We won’t use AI in core brand content where originality and integrity matter most, like:
newsletters/blog
thought leadership
writing that represents CMJ’s voice or perspective in a meaningful way
AI may help behind the scenes in limited ways, but it should not be shaping the heart of this work.
3. ❌ First drafts
The output from LLMs on a first draft is not generally very good, and reviewing bad outputs from AI (whether your own or a colleague’s) takes more mental load than reviewing poor human-generated output (there’s also usually less of it).
Instead, use AI to ask questions to clarify your thinking. You can even speak your messy, drawn-out thoughts and ideas into your chosen LLM for help refining, but it shouldn't draft anything without substantial human thought put into it first. Human input can be:
A voice memo
A bulleted list
A messy first draft
You make the first draft with your brain and brilliance, then ask it to refine.
4. ❌ Final drafts
Never copy-paste AI output and send or publish it as-is. Every communication needs human review, editing, and ownership. If you haven’t read it, don’t send it.
5. ❌ Handling sensitive/private information
No uploading identifiable client data or call transcripts unless names and company details have been stripped out. See Data Security and Privacy below.
What we DO use AI for
1. ✅ Editing
AI can be helpful in reviewing work (like de-identified sales proposals, playbooks, facilitation plans, etc) for:
repetition
clarity
structure
readability
Example requests:
Prompting an LLM to “Be a ruthless editor of my writing for clarity and conciseness. Remove jargon that would not be understandable to a general audience. The audience for this communication is community managers working in-house for enterprise firms,” and then share your original draft.
2. ✅ Thought partnership
Think of LLMs like coaches who have little subject matter expertise. AI can be useful when:
Coaching through decision-making, especially regarding decisions that feel murky (e.g., “talking” to Claude about how to price a new project - take all output with a grain of salt)
Collecting examples (e.g., “What are some examples of brand communities in the beauty industry?”)
Organizing ideas that already exist, like blog posts or transcripts (e.g., “What themes do you see in this conversation?”
Identifying action items or decisions from call transcripts (in our private transcript app only; do not upload call transcripts to outside LLMs unless names and sensitive information have been stripped out)
3. ✅ Research
LLMs can be useful with:
Identifying themes in qualitative research
To analyze survey data (always with identifying information stripped out, and always double-checked by a human)
Example use cases:
Utilizing Gemini’s analysis features inside of a Google Sheet to understand early patterns in survey data
Importing de-identified interview transcripts into Google Notebook LM and conversing with the associated chatbot to learn more about general patterns (this does not replace human review of the transcripts but augments that review and speeds up understanding of surface-level patterns)
4. ✅ As a CMJ expert and assistant
CMJ Team Assistant (in Gemini) is trained on all things CMJ, ready to help search our library of knowledge, improve copy to sound like our brand voice, and make work faster. It is trained on items like:
Our brand book (including values, positioning, mission, and vision)
Carrie’s book manuscript and other thought leadership content
Our services kit and packages
Launch Your Community course sales page and emails
Custom CMJ workshops on topics like onboarding, the commitment curve, and other key concepts
Example use cases:
For help matching the brand voice in CMJC automated DMs or emails announcing Fellowship acceptance or rejection
5. ✅ Developing wireframes, tooling, calculators, and workflows.
We aren’t engineers, but AI has democratized engineering and design tools that allow us to create things we only could dream of before!
Example use cases:
Community ROI Calculator: With a few key data points, tools like Claude, GitHub, and Vercel make it possible to create customized ROI calculators for our clients.
Mockups of community webpages: Quickly generate a mockup in Claude and Lovable to show a client before a build.
Visualizing workflows: Using tools like Scribe, Claude, and Gamma, visualize complex processes that have slowed down teams for years and identify the bottlenecks.
Data security and privacy
We should assume AI tools are not neutral and are never entirely secure. Even when a platform has strong security, there are risks.
Risks to keep in mind
hallucinations or made-up information
privacy leaks
model training insufficiency and lack of transparency
future breaches or hacks
over-sharing sensitive information in tools we do not fully control
sycophantic orientation that will default to telling you you’re right vs. challenging your thinking
Baseline rules
Do not put the following into AI tools unless we have explicitly agreed that a tool is safe for that use:
Client information: Community plans or content shared by or created for clients, like a playbook, research findings, content calendar, any documents shared from the client, etc.
Proprietary information: CMJ terms like “Community Compass” without a trademark (IP laws have not caught up to LLMs)
Sensitive internal business information: Client names, emails or contact information; team addresses or other personal identifying information
Raw transcripts or survey results with names, contact info, or confidential details, unless anonymized first
When in doubt, strip out names and details or do not use AI for that task. Always assume the tool isn’t smarter than you. Think critically about what data informed the output (and what was left out), and act accordingly.
These guidelines are current as of Summer 2026 and will be reviewed quarterly.
