When the client is also using AI: a new kind of creative collaboration

Your kick-off call starts, and the client shares their screen. There’s a mood board, a positioning statement, three headline options, and a rough content structure. All of it was generated with AI before breakfast. They’re enthusiastic, organized, and ready to move fast. The only problem: none of it is theirs in any meaningful legal or creative sense, the brief has no success metrics, and you’re now being asked to build on a foundation that looks solid but isn’t.

This is what client-led AI use looks like in creative collaboration right now. And most projects aren’t set up to handle it well.

The brief looks complete. That’s the problem

AI-generated briefs create an illusion of clarity. They’re well-formatted, use the right vocabulary, and cover the expected sections. But polish isn’t the same as substance.

According to BetterBriefs research commissioned by the IPA, across 1,700 respondents in 70+ countries, 80% of marketers believe they write a good brief. Only 10% of agencies agree. An estimated 33% of marketing budgets are wasted as a result of poor briefs.

That gap existed long before AI. What AI does is accelerate the production of briefs that look complete but still lack the things that matter: clear success metrics, genuine strategic direction, and real insight about the audience or the problem. A well-structured document produced in ten minutes hasn’t replaced the thinking that a good brief actually requires.

When a client hands you an AI-generated brief, the first question isn’t “do we have enough to start?” It’s “what human judgment has actually been applied here, and where?”

The IP question nobody is asking

When a client provides AI-generated concepts, copy, or mood boards as part of a project handoff, those materials almost certainly have no copyright protection.

The U.S. Copyright Office clarified this in January 2025: prompts alone do not create copyright. AI-generated content without meaningful human editing or selection cannot be copyrighted. The protection, if any exists, attaches to the human choices layered on top of the AI output, not the output itself.

This matters practically. If a client provides you with an AI-generated tagline and asks you to develop a campaign around it, and you do the creative work, your contribution is what creates protectable intellectual property. Theirs may not be protected at all.

This reframes what the creator is actually being hired to do. You’re not building on the client’s asset. You’re bringing the craft, judgment, and original contribution that transforms a first-pass AI output into something that belongs to someone. That’s a different kind of work, and it should be scoped accordingly.

*This describes the U.S. Copyright Office’s publicly stated position and emerging industry practice, not legal advice. Consult a lawyer for guidance specific to your contracts and jurisdiction.*

If you’re thinking through what questions to ask before hiring a creator for this kind of work, What to ask before hiring an AI-assisted creator covers the hiring side of the same conversation.

The rules of the room have changed

Here’s what’s now common: both parties arrive at a project having already used AI. The client has generated concepts, references, maybe copy. The creator has used AI for research, ideation, or production support. Neither party’s tools have any context from the other. And neither party has necessarily disclosed how much of what they’ve produced is AI-assisted.

This creates a new kind of alignment problem that goes beyond the usual expectation gap. The Spark Report (2026) found that 75% of creative agencies haven’t updated their contracts to reflect AI usage at all. The rules for this scenario don’t exist yet, which means the conversation you have at the start of the project is the only thing standing between a productive collaboration and a scope dispute.

Prof. Maneesh Agrawala of Stanford HAI put it plainly in March 2026:

“While the models seem amazing, they are terrible collaborators.”

The models don’t negotiate. They don’t ask what matters most to you. They don’t surface the assumptions buried in a brief or flag when two people are working from incompatible premises. That work still falls to humans. And in a dual-AI project, there’s more of it, not less.

The Design Business Council’s 2025 research (680 clients, in-depth interviews) found that 80% of clients expect AI to reduce agency costs. At the same time, 94% say AI shouldn’t lead to brand sameness. Both of those things cannot be true without significant human orchestration. Someone has to do the work of making the output distinctively yours. That’s the creator’s job, and it’s exactly the kind of work that gets underscoped when everyone assumes the AI handles it.

A framework for AI creative collaboration

Before work begins on any project where the client has used AI to generate inputs, three things are worth establishing explicitly.

1. Audit the AI inputs. What did the client generate? Where did human judgment shape it? What reflects a real decision they made, and what is a first AI pass they haven’t critically evaluated? You don’t need to interrogate the client, but you do need to know what you’re building on.

2. Establish shared creative ground rules. Is the AI-generated material a constraint (this is fixed), a starting point (this is directional), or a signal (this is what we’re drawn to, but we’re open)? The answer changes the scope of the work significantly. Getting clarity here prevents the most common source of mid-project conflict: the client thinking you’re refining their vision, while you’re still trying to establish what their vision actually is.

3. Define scope for a different kind of work. Refining AI output is not the same as originating creative work. It requires a different set of skills, often more judgment-intensive because the raw material looks finished even when it isn’t. Scope, timeline, and pricing should reflect what you’re actually being asked to do.

For a related structure, The async brief offers a template for kicking off creative projects with clarity from the start, which pairs well with this framework.

Five questions to ask before work begins

You don’t need a lengthy intake process. These five questions, asked directly before scoping, will surface most of what you need to know.

• What did you generate with AI, and what did you add to it? You’re not asking this to audit their process. You’re asking because the answer tells you what’s genuinely theirs and what’s still up for interpretation.
• What are the success metrics for this project? AI briefs frequently omit these. If the client can’t answer, the brief isn’t done yet.
• If we move away from the AI-generated concept, what are you attached to? This distinguishes a constraint from a preference, and it’s the fastest way to find the real creative brief underneath the generated one.
• Who owns the AI-generated inputs you’ve provided? Not a legal challenge, a practical one. If the concepts came from a tool with terms that affect ownership or usage, that’s relevant to scope and deliverables.
• Are you expecting us to start from your AI outputs, or develop independently and then reconcile? These are different projects. Both are valid. But they should be scoped, priced, and timed differently.

What good looks like

Dual-AI collaboration works when both parties are explicit about what AI contributed and where human judgment was applied.

That means the client knows what they actually decided versus what the AI produced. The creator knows what they’re originating versus refining. And the scope reflects the real nature of the work rather than the polished appearance of the inputs.

The co-op model The Blue Mango uses is well-suited to this because it operates transparently. Creators aren’t defensive about which tools they use or how. Clients aren’t surprised by what they get. The conversation about AI happens at the start of the project, not as a disclaimer at the end.

That transparency doesn’t resolve every tension in a dual-AI project. But it does create the conditions for the conversation that needs to happen.

The best dual-AI projects we’ve seen start with one honest conversation before any tools are opened. That conversation is still the hardest part to get right.