The jobs to turn down in an AI studio
A shop that takes everything does its worst work at its thinnest margin. Disqualifiers across brief type, client sophistication, approvals and exposure.
The jobs that damage a studio are rarely the ones that lose money outright. They are the ones that take four times the planned effort, produce work you would not show anyone, occupy the calendar slot a good job wanted, and end with a client who is not quite satisfied and will not quite say so. You do your worst work at your thinnest margin, and it becomes the portfolio piece that attracts the next client like them.
Capacity is not the reason to be selective. Superside's Breakpoint research found 80% of creative teams at or beyond capacity and 70% of creative leaders reporting burnout despite AI adoption — generation stopped being the bottleneck, and filtering, governance, rights and taste became it. Adding output capacity does not fix an intake problem.
Here are the four categories worth disqualifying on, and what to say instead of yes.
Brief type
Some briefs are structurally bad fits regardless of the client or the budget. Three to watch.
The brief that hinges on one impossible shot. Not hard — impossible with current tools: a precise multi-step object interaction, legible paragraph-length legal copy generated into frame, an exact dieline matching a real pack. Attempt counts rise sharply on specific character action and complex motion, and the shot that scared you at quoting will eat the schedule. If the client will not accept a reshape, the job is not a fit. Feasibility questions to ask before you quote is the fifteen-minute version of this test.
The brief where the whole point is that it is not AI. Occasionally a client wants exactly what a traditional crew produces and comes to you because they think it will be cheaper. It will not be. The credible case studies show one pattern: AI used for exploration and speed, human craft applied on finish. Nike's Serena Williams work generated 130,000 virtual tennis matches as an exploration step and still put human craft on what shipped.
The brief with no direction in it. "Make us something viral" is not underspecified, it is undelegated. Direction is the expensive input now, and a brief that arrives without it means you are being asked to supply it for free, then be judged against a standard nobody wrote down. Either sell the strategy work explicitly or decline.
Client sophistication
Not about client size. Some of the least sophisticated buyers are large, and some one-person businesses brief beautifully. The signals to read:
| Signal | What it predicts |
|---|---|
| Cannot describe what "good" looks like | Unbounded rounds; no exit condition |
| Has never bought this category before | Every norm has to be negotiated from scratch |
| Talks about volume, never about performance | The relationship has no success metric |
| Opened with a price objection before seeing scope | The whole engagement will be a price conversation |
| Asks for "AI pricing" as a discount category | See below |
That last one deserves a note, because the fear around it is larger than the phenomenon. Reporting aggregated by TechBullion puts 73% of agencies as never having been asked to cut prices despite adopting AI, and of the 27% who were asked, only 13% actually lowered rates. Clutch's 2024 figure of 61% of agency clients raising AI at renewal is about the subject coming up, not the price coming down. The consensus response is to reframe value rather than drop price — what to tell clients about using AI in delivery covers the framing, and raising prices on existing content clients is the harder version of the same conversation.
A client who opens with the discount, before scope, is a different animal from one who raises it at renewal.
Approval structure
The most reliable predictor of a job going wrong, and the least discussed. Four disqualifiers.
No named approver. If nobody will put their name against the sign-off, the sign-off does not exist, and every decision stays reopenable. Hold hardest on this one: easy to fix at intake, impossible to fix in week three.
A final approver who is not in the process. The executive who sees it first at delivery reopens the concept, and every decision below it, at the moment reversing costs the most. Ask who else will see this before it ships.
Approval by committee with no tiebreaker. Contradictory notes from three stakeholders is not one round of feedback, it is three rounds arriving simultaneously and cancelling out. Consolidated feedback from one named role is worth insisting on — a revision policy that stops scope creep has the language.
Approvers who change mid-job. A decision signed by one person and reopened by their replacement is a change order, and if you cannot get that written down, price the job as if it will happen.
The connecting logic: what you sell is direction, and direction requires someone with the authority to close options. Scoping a job by decisions rather than deliverables makes this checkable at intake rather than discoverable at delivery.
Exposure
The last category is the one that can cost more than the fee. Four kinds.
Rights you cannot verify. Client-supplied reference material with unclear provenance, a likeness whose consent nobody can produce, music with an unclear chain. The problem is not that you will be sued; it is that you cannot warrant what you have not seen, and the client will assume you did. Legal and licensing basics for AI business content covers the questions, and the paperwork file every campaign needs covers what you keep afterwards.
A full-ownership warranty on generated material. The clause most likely to be signed carelessly. Platform terms from major providers assign contractual ownership of outputs to the user, but a contract cannot manufacture copyright that statute declines to grant, and the US Copyright Office position is that protection attaches to human contribution — substantial modification, human-created elements, and the creative selection and arrangement of generated material. Raw output may carry no copyright at all, and you cannot assign a right that does not exist.
The practitioner fix is to decline the clause rather than the job: replace "deliverables are original works" with a carve-out stating that the agency does not warrant that AI-generated portions qualify as copyrightable and that status varies by jurisdiction. A client who will not accept the carve-out is asking you to promise something nobody can promise.
Uncapped liability, or a cap unrelated to the fee. The standard minimum AI addendum ties the liability cap to the fee paid, alongside disclosure, the IP warranty carve-out and a training-data exclusion for client materials. A job with paid media behind it and an uncapped indemnity is one where a bad week costs more than a good year.
Audience-facing risk you would not carry yourself. The underrated one, and not a legal question. eMarketer reports 60% of US ad professionals citing accuracy and transparency as a top barrier to AI adoption, and a 37-point perception gap between ad executives and Gen Z and millennial consumers on AI ads, widened from 32 points in 2024. Coca-Cola's AI holiday campaign compressed a roughly year-long timeline to about a month and drew significant public backlash at the same time. The commercial risk sits with the audience, not the invoice.
Which means a job where an AI-generated execution will read badly to the specific audience being sold to can succeed on every internal metric and fail in public. Ask whether the client has thought about it, and whether they will label the work. If the answer to both is no, the exposure is partly yours. Writing an AI disclosure line nobody scrolls past and the brand safety checklist for AI-generated content belong in that conversation.
How to decline
Declining badly costs referrals. Three rules.
Decline the shape, not the client. "This particular brief isn't a fit for how we work" leaves the door open. "We don't take work like this" does not.
Name the reason honestly, once. Clients rarely argue with a specific reason and often argue with a vague one. "The single-take pour is the whole ad and I can't reliably deliver it — I'd be taking your money to find out" is a sentence that gets you recommended.
Offer the version you would take. Most disqualified jobs have an adjacent version that works: the reshaped brief, the smaller pilot, the paid feasibility sprint before the full quote. Put it on the table and let them choose. A surprising share come back in the shape you offered, and those are usually the good ones.
FAQ
Isn't this a luxury for studios that are already busy?
The opposite holds. A studio with a thin pipeline is exactly the one that cannot afford three months consumed by a job producing no portfolio piece and no referral. The disqualifiers above are mostly about structure rather than budget, so they apply at any size.
What if the budget is genuinely large?
Then price the disqualifier rather than ignoring it — a paid feasibility sprint first, a change-order structure rather than a fixed price, a liability cap you can live with. What you should not do is treat a large budget as evidence that the structural problem will resolve itself. A large budget with no named approver is a large problem.
How do I know if I'm being too picky?
Track the jobs you turned down and what happened. If the ones you declined were delivered well by someone comparable, your criteria are too tight. If they went nowhere, changed shape three times, or came back at a better price, they were correctly declined. It takes about two quarters to get a signal.
Should any of this be on my website?
Some of it, framed as what you do rather than what you refuse. A clear statement of the brief types you are strong on, the approval structure you work with, and the fact that you quote after a feasibility conversation will disqualify the wrong enquiries before they reach you. It is the cheapest filter available and it costs one page.