The AI Disclosure Decision Tree for Brand Social

What is the decision tree in one sentence?

Disclose whenever AI created or materially changed anything the buyer uses as evidence, meaning a person, a voice, a result, or a demonstration. Everything upstream of the evidence, like drafts, research, editing, and captions delivered by real humans, needs no disclosure because no reasonable buyer would feel deceived by it.

That single test resolves almost every case my clients bring me, and it is stricter than what the platforms require. Platform labeling rules for synthetic media keep shifting, and legal requirements vary by market, so I do not build client policy on either one. I build it on the question underneath both: if this viewer found out the truth after purchasing, would they feel tricked? If yes, disclose or do not publish. If no, ship it.

How does the tree actually run, question by question?

When a piece of AI-touched content is headed for a brand account, I run it through four questions in order. The first yes decides the outcome.

1. Did AI generate or alter a human being? A synthetic face, a cloned voice, a digitally inserted person, an avatar presenter. If yes: disclose, always, visibly, not buried in a hashtag. In trust-heavy categories I usually go further and say do not run it at all, because a synthetic person doing a real person’s trust job undermines the reason the format works.

2. Did AI generate or alter a result or demonstration? Before-and-afters, skin texture, body shape, a product performing. If yes: this is not a disclosure situation, it is a do-not-publish situation for any brand where results drive the purchase. A label under a fake result does not neutralize the deception, it documents it.

3. Did AI generate or alter the product or scene the buyer is evaluating? A rendered product shot, an AI backdrop, a staged environment. This is the genuinely gray zone. My line: if the alteration changes what the buyer thinks they are getting, disclose. An AI sky behind a real product is set dressing. An AI render that makes packaging or texture look different from what arrives in the mail is a false expectation with a return rate attached.

4. Did AI only touch the words or the workflow? Scripts a real human delivers, caption drafts, research, brief scaffolding, edit assists, translations. No disclosure needed. This is authorship help, not evidence manipulation. The risk here is flatness, not ethics, which is why everything still gets rewritten by a human before it ships, a discipline I detailed in what I actually use AI for.

Where do the categories change the answer?

The tree is universal but the stakes are not. The more intimate the purchase, the harsher the penalty for getting this wrong.

Content situation	Low-trust category (apps, gadgets)	Trust-heavy category (beauty, wellness, med spa)
AI avatar presenter	Disclose and test	Avoid
AI-altered results or before-and-after	Do not publish	Do not publish
AI product render	Disclose if it differs from reality	Disclose, and prefer real photography
AI backgrounds and set dressing	No disclosure needed	No disclosure needed
AI-written script, human delivery	No disclosure	No disclosure, edit hard for voice
AI-assisted editing and cutdowns	No disclosure	No disclosure

The pattern in the right column is simple: when the buyer’s decision depends on believing a real body got a real outcome, synthetic evidence is not a shortcut, it is a category error. Skincare, supplements, aesthetics, and anything that touches health run on proof, and buyers in these categories are already skeptical of polish, let alone synthesis.

Why write the rules down before you need them?

Because case-by-case decisions under deadline pressure always drift in one direction: toward not disclosing. The asset looks fine, the launch is tomorrow, nobody outside the team would ever know. Every step down that slope feels reasonable, and the accumulated result is a brand account quietly stuffed with synthetic evidence and no paper trail of who decided what.

I have watched this drift happen inside otherwise careful teams, and it never starts with a bad actor. It starts with an ambiguous asset, a launch deadline, and nobody whose job it is to say no.

So I have every client adopt a one-page AI content policy before the first synthetic asset gets tested. It states the four questions, the category-specific hard lines, where the disclosure label goes, and who signs off on gray-zone calls. It takes an afternoon to write and it converts a recurring ethical debate into a checklist item. It also protects the team: when a platform policy shifts or a competitor gets publicly burned, you are adjusting a document instead of improvising a defense.

One more reason to be early rather than legal-minimum on this: disclosure is becoming a positioning asset. Audiences are getting sharper at spotting synthesis and increasingly resentful of discovering it. Brands that are casually, confidently transparent about where AI sits in their process read as secure. Brands that get caught read as everything the audience already feared about advertising.

If you want a second set of eyes on where AI belongs in your content operation and where it is quietly costing you trust, apply for a content growth diagnostic.