Somewhere in your business, right now, someone is using AI. Maybe it’s you, drafting emails. Maybe it’s a freelancer you didn’t ask. Maybe it’s an agency that “leverages cutting-edge tools” and is hoping very hard you don’t ask which ones, for what.
Here’s the uncomfortable truth: your customers already assume AI is involved in your marketing. The only open question is whether you look like you’re hiding it. Silence reads as concealment. A vague disclaimer reads as something a lawyer made you do. The move that actually builds trust is the one almost nobody makes: a short, plain-language policy that says exactly where the machines stop.
Here’s ours. Copy it. The rest of this post explains each section so you can adapt it honestly, rather than just swapping in your logo and calling it culture.
The policy
Our Honest AI Policy
What AI does in our work: transcription, image resizing and reformatting, first-pass research, drafting variations of human-approved originals, assembling reports, and organizing files. Mechanical work with a clear right answer.
What AI never does: talk to you or your audience, set strategy, handle sensitive communication, invent facts or statistics, or publish anything.
Our guarantee: a named person reviews and approves everything before it goes out. Every claim traces to a source we’ve seen. If we can’t verify it, we don’t say it.
Ask us: if you ever want to know how a specific piece was made, ask. You’ll get a straight answer.
Last reviewed: [date]. We update this as tools change. The guarantee doesn’t.
That’s the whole thing. One page. On purpose.
Why each section is there
“What AI does” comes first, and it’s specific. Vague policies (“we use AI responsibly”) protect nobody and convince nobody. Naming actual tasks does two jobs at once: it tells customers what to expect, and it forces you to actually know your own workflow, which is more embarrassing than it sounds. If you can’t fill this section in, that’s not a writing problem. It’s a sign you haven’t decided where your line actually is yet.
“What AI never does” is the trust engine. Nobody is lying awake worried about AI-resized images. People worry about talking to a bot that’s pretending to be a person, and being sold with made-up facts. This section speaks to the real fear, directly. Only list things you’ll genuinely uphold, because one caught exception costs more than the policy ever earned.
The guarantee names accountability, not process. “A named person approves everything” is checkable. “We maintain rigorous quality standards” is wallpaper. This line is the load-bearing one. Everything else is trim.
“Ask us” is the cheapest section and the most disarming. Almost nobody will ever actually ask. Offering anyway signals you have nothing to manage. You do have to mean it, though. That’s the catch with honesty generally.
The review date keeps it honest over time. Tools change fast, and a policy dated three years ago says the opposite of what it intends. Put a recurring reminder in your calendar. Actually re-read it when the reminder fires, instead of doing what we all do with recurring reminders.
The three ways this becomes theatre
Publishing aspirations instead of practice. If AI currently drafts your customer service replies, the policy either says so, or you change the workflow before publishing. Not after someone notices. An honest policy about an imperfect setup beats a beautiful policy about a fictional one, every time.
Making it long. Every paragraph past one page is a paragraph nobody reads and a promise nobody remembers making. Your lawyer can have a separate document. This one is for humans.
Writing it once and forgetting it. The policy is a living line, not a plaque in the lobby. When you adopt a new tool, the first question is which side of the line it lands on. If the line moves, the policy updates the same week.
What actually happens when you publish it
Let’s set expectations honestly, since that’s the theme: most customers will never read it. But the ones who do are disproportionately the skeptical, careful ones. Exactly the people a vague answer would have lost, and exactly the people worth keeping.
And something better happens internally. Your team stops making quiet, individual judgement calls about what’s acceptable, because the line is on the wall where everyone can see it. The published page is for customers. The clarity is for you.
The full reasoning behind the split lives in The Credible Middle, and our own filled-in version is public at What We Let AI Do. If you want this policy as a formatted, fill-in-the-blanks document, grab the template here. No charge, no catch, no seventeen-field form asking for your company size.
Draw the line. Write it down. Say it out loud.