Generative tooling has crossed the line where the question is no longer "should we use it" but "where in the brand system does it sit". The companies still treating that as a moratorium-or-free-for-all binary are losing time. The companies writing a usable policy are quietly compounding an advantage. Where the line sits is now a brand-strategy decision, not a tooling decision.

Why the binary framing has stopped being useful

Most internal AI policy on creative work in 2026 still reads as one of two postures. Either "we use generative tools where it makes sense", which translates in practice to "every individual decides", or "we do not use generative tools for client work", which translates in practice to "we use them anyway and quietly". Both postures fail at the brand layer because they are not policies — they are absences of policy. A working policy treats generative output as one of several types of asset, each with its own governance. Some types of output are ephemeral and low-stakes: a draft layout for an internal review, a moodboard exploration, a placeholder image in a spec. Some are foundational and high-stakes: a brand mark, a typeface, a brand photographic library, a piece of headline copy on the marketing site. The policy clarifies which assets sit where, and what governance applies in each tier. The brand-strategy implication is that generative tooling has produced a third asset class alongside human-made and licensed: human-directed-machine-made. That class needs its own treatment, its own QA, and its own attribution.

The three-tier policy

A policy that survives contact with both the brand team and the legal review usually sorts assets into three tiers. The names are not load-bearing; the boundaries are.

• Tier 1 — exploratory. Generative tools are used freely. Outputs are internal artefacts: ideation moodboards, draft compositions, working sketches, placeholder copy in design files. The output never leaves the company in this form. The governance is light: track what tool was used and don't paste outputs into client deliverables verbatim.
• Tier 2 — production-supporting. Generative tools are used as a step in a human-led process. A human designer generates options, selects, refines, and ships work that a stranger could not unambiguously identify as machine-generated. The governance is documentation: what tool, what prompt class, what human refinement steps. The asset is treated as human-directed, with disclosure where required by client contract or platform.
• Tier 3 — foundational and identity-bearing. Generative tools are not used as the source. Brand marks, primary typography, distinctive typographic licensing, primary photographic library, hero illustrations, brand voice guidelines themselves — the assets that compound across years and channels — are produced through human craft with traceable provenance. The reasoning is risk: assets that carry brand identity weight cannot rely on training-data lineage that may be challenged later.

The strength of the three-tier framing is that it does not ask the team to decide whether AI is "good" or "bad". It asks them to decide which tier each piece of work sits in. Those decisions are tractable.

Where the three lines actually fall

The boundaries between tiers are where most policies fail. A short list of decisions that need to be made explicit: Photography. Generated photography for editorial supporting roles can sit in Tier 2 if the brand permits stylised illustration; for documentary or claim-bearing imagery (people, locations, products), it stays out. The line is whether a viewer would interpret the image as evidence of something the brand is asserting. Copy. Generated copy for first drafts, alt text, internal documentation, and structural exploration sits in Tier 1 freely. Generated copy that ships under the brand voice — homepage hero, headline campaign, founder essay, anything that constitutes the brand speaking — is Tier 3. The middle ground (case-study writing, product copy, support content) is where the policy has to be specific. A reasonable line: human-written first draft, AI used for polishing and consistency checking, human signs off; not the reverse. Iconography and illustration. Generative outputs in this category are tempting and frequently unusable. The asset class needs to feel cohesive across surfaces; generative outputs vary enough between sessions that achieving that cohesion is harder than it looks. Most companies end up with this in Tier 2 with strict treatment guidelines, or in Tier 3 entirely. Brand marks and identity systems. Tier 3, without ambiguity. The reasons combine craft (a logo is a constraint problem in two-colour, two-weight, sub-pixel performance, and generative output is not optimised for any of those constraints) and risk (the provenance question is currently unresolved enough that the brand asset that defines the company should not be the test case).

Three patterns recur in policies that fail

The blanket ban. Treating all generative tooling as off-limits causes the work to happen anyway, undocumented and ungoverned. The policy has not prevented the use; it has prevented the conversation about it. The blanket permission. "Use these tools where they help" sounds collegial and is unfalsifiable in practice. It produces a brand presentation that drifts in tone and quality across surfaces, because the level of human direction varies between teams and between assets. The brand becomes the average of how individual people decided to use the tools that week. The audit gap. A policy is written but no audit is performed. The marketing team self-reports compliance; the brand team trusts that report; six months in, an audit reveals that "AI-assisted" has come to mean different things to different people, and several Tier 3 surfaces have generative outputs in them. The fix is not stricter policy — it is a quarterly audit of asset provenance, treated as a normal governance task.

What to disclose, what to govern, what to refuse

Disclosure, governance, and refusal are three different decisions that often get bundled. They should be unbundled.

• Disclosure — when the audience or the contract requires it. Some clients now require disclosure of AI use in deliverables. Some platforms require it. The policy should specify which situations trigger disclosure and what form it takes. A line in the colophon is usually sufficient; a banner across the work is rarely necessary.
• Governance — internal record-keeping for assets the brand will live with for years. What tool, what version, what prompt class, what human refinement, what review cycle. This is invisible to the audience and load-bearing for legal, audit, and brand-evolution purposes.
• Refusal — the explicit list of work the brand will not produce with generative tools, regardless of efficiency. The Tier 3 set above is a starting point. The list should be specific enough that nobody guesses; the worst version of this discipline is one where individuals interpret "where it makes sense".

What This Looks Like in Practice

When we worked with Antidote Africa, the question was not whether to use generative tools — the team had been experimenting for over a year — but where the brand would and would not accept their output. The brand work included an explicit policy: copy drafting and structural exploration in Tier 1, marketing imagery in Tier 2 with treatment guidelines, brand mark and primary brand photography in Tier 3 with no generative source. The policy was not the headline of the engagement; it was a small section of the brand book. Two years on, it is the section that has aged best, because the team can answer "is this allowed" by reference to a document, not by individual interpretation.

Closing

The line between governed and human-made is not a moral statement about generative tooling. It is a brand decision about what compounds. Some assets are better produced quickly and cheaply; some need the provenance and craft that only human direction provides. The brands that will be coherent five years from now are the ones that drew that line in writing in the year the question first became unavoidable. If you are writing or revising your own policy and want a candid second opinion on where the lines actually fall for your category, we are happy to walk through it.