The question of whether to disclose AI use has become a brand question, not a compliance one. Disclose nothing and you risk being caught, which costs trust at the worst possible moment. Disclose everything and you bury the meaningful admissions under a pile of ritual ones, training your audience to ignore the label. Most organisations are improvising a position one project at a time. A working policy — written once, applied consistently — is cheaper than the improvisation and far less risky.

Why disclosure is a brand decision

Disclosure of AI use is usually framed as a regulatory or legal matter, which makes it someone else's department. But the audience does not experience disclosure as compliance; they experience it as a signal about the company's honesty. A disclosure that arrives voluntarily, in plain language, reads as confidence and transparency. A disclosure extracted after the fact, or buried in terms nobody reads, reads as something the company tried to get away with. The same fact — we used AI here — lands as trustworthy or evasive depending entirely on how the brand chose to handle it. That is a brand decision, and leaving it to be improvised per project guarantees inconsistency at the exact moment consistency builds trust. The regulatory landscape is tightening — disclosure expectations for AI-generated and AI-assisted material are becoming explicit in more jurisdictions, and the responsibility is increasingly placed on both the creator and the brand. But meeting the legal minimum is not the same as having a brand position. The law tells you what you must disclose; the brand decision is what you choose to disclose, how, and in what voice, so that the disclosure builds rather than erodes the relationship.

The two failure modes

There are two ways to get this wrong, and they pull in opposite directions. The first is under-disclosure: saying nothing, on the theory that what the audience does not know will not hurt them. This works until it does not, and the cost of being discovered is far higher than the cost of having disclosed. The second is over-disclosure: labelling everything, including the trivial, until the disclosures become noise. A label on every email signature and every stock-grade image trains the audience that the label means nothing, which destroys its value for the cases that actually matter. Performative over-disclosure is its own kind of dishonesty — it looks like transparency while functioning as camouflage. A working policy threads between these by distinguishing what is material from what is not. The goal is not maximum disclosure; it is meaningful disclosure, reserved for the cases where a reasonable member of the audience would want to know and would think differently if they did.

What to disclose, and what not to

The principle that does the work is materiality: disclose AI use where it would change how a reasonable person interprets or trusts the output. That principle resolves most cases.

• Disclose — AI-generated imagery presented as real, synthetic voices or likenesses, AI-written content presented as a named human's view, and any output where the AI's involvement bears on the audience's trust in its accuracy or authenticity.
• Usually disclose — substantial AI contribution to creative work a client is commissioning and paying for as human craft, because the client's expectation is the thing at stake.
• Need not disclose — AI as a routine production tool with human authorship and accountability intact, in the same way you would not disclose spell-check, a calculator, or a stock photo library.

The line is not "did a machine touch this" but "would the audience feel misled to learn how it was made". That test is stable across cases and defensible when questioned, which a per-project guess is not. An AI-generated model wearing a garment on an e-commerce page is the clearest example: if the shoot never happened and the model does not exist, that bears directly on what the customer thinks they are seeing, and it sits firmly on the disclose side of the line.

How to disclose well

The how matters as much as the what. A good disclosure is in the brand's own voice, proportionate to the thing being disclosed, and placed where the relevant audience will actually see it rather than in terms they will not read. It states what was done plainly — "this image was generated", "this voice is synthetic" — without either apology or euphemism. It does not hide behind passive constructions or technical language designed to be skimmed past. And it is consistent: the same kind of use is disclosed the same way every time, so the audience learns that the label is reliable and therefore worth reading. This connects directly to the broader question of governing AI in creative work. A disclosure policy is one component of an AI ethics framework for brand creative; it sits alongside decisions about what to generate, what to govern, and what to keep human. Disclosure without those upstream decisions is a label on an ungoverned process, which is honesty about a problem rather than a solution to it.

Where these policies break down

Three patterns recur. The blanket label — a company so anxious about disclosure that it labels everything, destroying the signal it was trying to send. The buried admission — a technically-present disclosure placed where no ordinary reader will find it, which satisfies a lawyer and fools no one when it surfaces. The shifting line — disclosure decided case by case by whoever is in the room, producing inconsistency that reads, correctly, as the absence of a real position. Each is solved by the same thing: a written policy, grounded in materiality, applied the same way every time.

A policy that stays alive

The danger with any disclosure policy is that it is written once, filed, and frozen, while the practice it governs moves underneath it. AI use in a creative team changes quickly — new tools, new capabilities, new places the technology enters the workflow — and a policy that does not move with it drifts out of relevance within a quarter. A working policy needs an owner and a review cadence, the same as any operational standard: someone responsible for asking, periodically, whether the materiality line is still in the right place given how the team is now working. Keeping the policy alive also means treating edge cases as inputs rather than annoyances. Each time a project raises a case the policy does not cleanly resolve, that is information about where the line is fuzzy, and the resolution should be folded back into the written rule so the next person does not have to improvise the same call. Over time this produces a policy that reflects the team's actual practice rather than a tidy abstraction written before the hard cases appeared. The aim is a document the team trusts because it has been tested against real work, not one they route around because it never anticipated what they actually do.

What This Looks Like in Practice

In our work with Fanblock, the practical need was not a philosophical stance on AI but a policy the team could apply under deadline without re-litigating it each time. We worked from materiality: a short, written rule that named the cases requiring disclosure, the cases that did not, and the plain-language form the disclosure should take in the brand's voice. The effect was to take the decision out of the individual moment — where it was being made inconsistently and anxiously — and settle it once, so the team could move quickly and the audience could trust that when the brand disclosed something, it meant it. The policy was deliberately short, because a disclosure policy nobody can remember is a disclosure policy nobody applies.

Closing

Disclosing AI use well is not about disclosing the most; it is about disclosing what is material, in your own voice, the same way every time, so the audience learns that your label can be trusted. Write the policy once, ground it in whether a reasonable person would feel misled, and apply it consistently — that is what turns disclosure from a risk into a trust signal. If you are working out where your brand's line should sit on AI disclosure and want help drafting a policy you can actually apply, we are happy to help you write it. This article discusses disclosure practice in general terms and is not legal advice; the specific disclosure obligations that apply to you depend on your jurisdiction and circumstances.