How to Use AI Ethically as a Creator: The 4 Things I Won't Do

📺 Watch the video · 📄 Get the one-page AI Use Policy — free, fill it in your browser, print it · 📓 The Buy-Back Playbook

I want to open with something that sounds like a contradiction.

I automate a huge amount of my business with AI. Research, first drafts, assets, repurposing, entire workflows. I am not about to tell you to slow down and do everything by hand.

And there are still four things I will not do with it.

Not because I think the technology is evil. Not because I'm precious about craft. Because I have watched all four of these cost real people their money, their clients, or their channel. And one of them almost got me.

So this is not an ethics lecture. Think of it as a risk list. Four places where creators and solo business owners are quietly exposed right now, and most of them have no idea. If you've been searching for how to use AI ethically without giving up the speed, this is the version that fits on one page.

First, a disclosure — and it's the whole point

The presenter in the video is not a camera recording of me. It's my digital avatar, built from my face and my voice, trained with my permission.

Let me be specific about how I use it, because a vague disclosure is just hiding with extra steps. The thinking is mine. The position, the argument, the examples — that starts with me every time. AI comes in after that: it tightens a sentence that ran long, it fixes an order that wasn't working. That's the ceiling. Ghostwriter at most. It does not decide what I think. If that ever changes, I'll tell you — and honestly, if the ideas ever stop being mine, you should stop watching me.

I'm putting this up front because I'm about to spend two thousand words on disclosure. If I didn't disclose the synthetic person delivering the message, none of the rest would mean anything.

The one test

I use a single test for all four of these, and it fits in one sentence:

Would I be fine if you saw exactly how this was made?

You're seeing it right now. That's the whole post.

And to be clear, this is not an anti-AI position. If you hand a sword to a protector, a criminal, or an idiot, you can't blame the sword for what happens next. I still believe that. But "it's just a tool" doesn't end the conversation — it starts it. The person holding the tool still decides what they're making, who they're telling, and whose face they're using.

The reason this is more urgent than it was a year ago: the consequences stopped being reputational. They're contractual now. They're regulatory. And on some platforms they're automated. You don't get a friendly warning email from a policy system. Your video gets claimed. Your ad gets rejected. Your client reads a clause you never put in the agreement.

Four pressure points. One at a time.

1. Consent — labeling is not permission

Whose face, whose voice, whose work is in this?

The easy case is my avatar: my face, and I signed off on it. Where it turns is when the same tooling gets pointed at somebody who never agreed. A client. A podcast guest whose audio is already sitting on your drive. A contractor. A real person's face on a thumbnail because you know it'll get clicks.

Here's the part most creators have backwards, and it's worth writing down:

Labeling something as AI is not the same as having permission to use it.

Those are two different systems. Platforms run likeness claims separately from copyright claims, which means you can disclose honestly, correctly, exactly the way you were told to — and still get the video taken down. The label was never the permission.

Then there's voice. Tennessee's ELVIS Act took effect in July 2024 and treats your voice as a property right, including simulations of it. California, Illinois, and New York have their own digital-replica and biometric rules. There's no single federal law here — it's a state-by-state patchwork, not one clean answer.

The fix is boring, and it works. Consent in writing, before you generate. This likeness, this use, this long, these platforms. Somebody saying "yeah, go ahead" on a call is not a consent file.

And flip it around: put a likeness clause in your own contracts, so a brand can't clone you off the back of one paid campaign.

2. Disclosure — your audience and your client are two different jobs

People conflate these. Different triggers, different consequences.

Your audience. The test platforms use is whether content is meaningfully altered or synthetically generated and looks realistic. Does it make a real person appear to say something they never said? Does it alter footage of a real event? Does it show a realistic scene that never happened?

What's not covered is just as useful: beauty filters, color correction, captions, outlines, using AI to help you think. Nobody needs to know spell check helped. They do need to know if the person on screen isn't real.

I have to be honest about the specifics, though. These policies move constantly. Where the label shows up, when the platform applies one for you, whether you can remove it later — all of that has changed more than once and will change again after I publish this. Read the current page yourself (linked below). Don't take my word for the mechanics.

And the trap underneath all of it: disclosing isn't supposed to cost you reach. But inauthenticity is a completely separate problem, judged at the channel level — mass-produced repetitive output, synthetic personas handing out financial or medical or legal advice. Disclosure does not buy you permission to farm.

Your client. If somebody is paying you for a deliverable, AI in your pipeline is material information. It belongs in the agreement up front, not in an awkward conversation later. I've built websites and charged real money for them, and here's what I've actually learned: most clients don't care that you used AI. They care that they found out from somebody other than you.

Here's the line I use. Adapt it with your own lawyer:

AI tools may be used in producing this work. No client material is entered into any tool that trains on it. A named human reviews and approves every deliverable.

3. Confidentiality — the setting almost nobody has opened

This is the one I see committed daily by people who would never do the other three.

You're in a hurry, it's late, and you paste a client contract — or a customer list, or an unreleased launch plan — straight into a consumer chat window.

What most people have never checked: on the consumer tiers of the major assistants, your conversations may be used for training unless you go turn that off yourself. On business, workspace, and API tiers, that's generally off by default. And opting in on a consumer account can extend how long that data sticks around.

So this isn't a budget decision. If you handle client work under an NDA, the business tier is the difference between "their terms may allow this" and "our contract says they don't."

My never list is short:

Client names attached to deal terms
Unreleased plans
Signed contracts, and anything under an NDA
Customer personal data, health files, financial files

And the one everybody forgets: uploading a PDF for a quick summary sends the entire document to a third party. Same rule applies.

Go open the data controls in whatever you're using tonight and actually look at them. Most people have never opened that panel. I'm including myself in that for longer than I'd like to admit — and I'm not going to tell you what the default is, because it moves by plan, by region, and by month. That's exactly why you check it instead of trusting a blog post.

4. Accuracy — the one that almost got me

I asked a model for supporting examples on a project. It came back clean, confident, and specific. Company names. Numbers. The whole thing. And a chunk of it was invented. Not wrong. Invented.

It escalates fast. Ships in a video — I look sloppy. Ships in a client deliverable — that's a refund and a reputation. Shows up as a testimonial or a review — it stops being an editorial problem and becomes a regulatory one.

The FTC's rule on consumer reviews and testimonials (16 CFR Part 465) took effect in October 2024. It covers creating, buying, selling, or spreading fake reviews and testimonials, and AI-generated ones are included. "I didn't know" is not much of a defense when "should have known" is the standard.

Let me be precise, because I don't want to scare you off something that's completely legitimate. Virtual influencers aren't banned. Fabricated experience is. An avatar can present information. An avatar cannot tell people it used the product.

One more thing almost nobody warns you about: in the United States, purely AI-generated material is not copyrightable. Prompts by themselves are generally not enough authorship — although your selection, arrangement, and editing of the output can be. Ownership here is complicated, and I wouldn't assume that generating something gives you unlimited rights to it. If you're telling a client they own an asset outright, document what you actually contributed.

My rule fits in one sentence: AI drafts the claim. I own the claim. Anything with a number, a name, or a source in it gets checked by me before it ships.

"Isn't this just fear?"

No. I'd say the opposite.

Every one of these takes one afternoon to set up and then stops costing you anything. Writing a consent template is a one-time job. Adding a clause to your agreement is a one-time job. Opening your data settings takes about four minutes.

None of this slows me down. It's the reason I can move as fast as I do without lying awake about it.

Your next step fits on one page

Write your own AI use policy. Five fields. One sheet. Then you stop making these decisions under pressure at eleven at night.

Approved tools and tiers — which tools, on which plan, for which kind of work.
What never goes in — the data categories that don't touch an AI tool at any tier.
What you disclose to your audience — what triggers a label on published work, in your own words.
What you disclose to clients — the line that goes in your agreements, written once and reused.
Who signs off — the named human accountable for every claim that gets published. If you're a solo operator, that's you. Writing your own name in that box changes how it feels.

The one-page policy is free, and there's no email gate on it. It's a fill-in sheet: your answers stay in your browser, nothing is uploaded, and it prints to a single page.

→ Open the AI Use Policy

If you want the bigger picture on where AI actually buys back time in a solo business, the 15-page Buy-Back Playbook is free too — that one's an email trade.

→ Get the Buy-Back Playbook

If you want people to compare notes with

The policy is the part that actually protects you, so do that first either way.

If you also want the templates — the consent form, the client clause, the disclosure language — plus a room of people building with AI who'll tell you when you're about to make one of these four mistakes, that's what the Roundtable is. Private community, resource library, prompt database, live sessions with the full archive, channels sorted by discipline so you ask in the room that matches your craft. $29/month, or $249/year (normally $348). The rate you join at is the rate you keep, and you can cancel yourself from your Stripe receipt — no retention call.

Who should skip it: if what you want is somebody to hand you finished output so you never have to learn the thing, that's not what's in there.

→ See what's inside the Roundtable

Do the one page this week. Then go right back to automating everything else — because that's what this whole thing is actually for.

Sources

Named rather than quoted on purpose — these pages change, and a stale claim in a rendered video can't be quietly edited. The durable test is what the video states; the current mechanics live here.

Not legal advice. I'm not a lawyer and I'm not your compliance department. Rules on synthetic media, endorsements, and AI-generated work vary by jurisdiction and change often — verify against the primary sources above before relying on any of it.

How this post was made: the argument, the examples, and the position are mine, carried over from the video script. AI tightened sentences that ran long. It didn't decide what I think.