AI Doesn't Have to Produce Average Ideas. Your Brief Might Be the Problem.
Ask an AI for ten creative campaign ideas and there's a decent chance you'll get ten ideas that sound like advertising.
A challenge. A community. A movement. Something about authenticity. Something about making everyday moments extraordinary. Add a hashtag and a brand film. Everybody can go home.
This is often presented as evidence that AI is incapable of originality. It may also be evidence that the brief was rubbish.
A human creative team asked to produce ten interesting ideas without a clear problem, audience or standard would have a difficult afternoon too. They might hide the uncertainty with experience and good taste. The uncertainty would still be there.
Creativity doesn't improve because you ask for more of it
A model can generate many plausible associations. That is useful. It can also reproduce familiar patterns, especially when the instruction leaves the destination almost entirely open.
The usual reaction is to ask for more ideas, stranger ideas or more disruptive ideas. The result is often more material to sift through rather than a better reason to choose any of it.
There is a different approach: make the problem more precise before making the request more extravagant.
What does the brand need to change? Whose behaviour matters? What does that person believe today? What evidence suggests the belief could shift? What can this brand credibly say that its competitors cannot? What would make an idea strategically wrong even if it sounded entertaining?
These are constraints. In creative work, constraints can be liberating because they give people something specific to solve.
The Doshi and Hauser short-story study (2024) found that AI suggestions could improve individual ratings while making stories collectively more similar. It studied short stories, not campaign ideas. The finding gives a reason to inspect variety as well as fluency; it does not establish that a tighter advertising brief guarantees originality.
The freedom of a tight brief
Good creative departments have known this for decades. A useful creative brief makes clear what the idea has to accomplish, while leaving room for creatives to find the answer.
The same principle applies when AI becomes part of the discussion.
Consider a fictional brief for a regional train operator. The marketing team wants more people to use trains for weekend trips. The initial request is simply: Give us ten creative ideas to encourage weekend train travel.
An AI assistant could produce perfectly serviceable suggestions: a social challenge, a travel inspiration series, a discount mechanic, a partnership with local attractions. None is necessarily wrong. None is yet anchored in a compelling strategic choice.
Now improve the brief. People already know trains exist, so an awareness objective leaves the obstacle unresolved. They assume a weekend trip will involve too much planning and too many decisions. The brand wants to make spontaneous day trips feel easy. The target audience is urban adults who enjoy getting out of town but frequently leave it too late to organise anything. The campaign must be credible without pretending trains are always cheaper or faster than driving.
The conversation changes. The assistant can explore ideas built around the tension between wanting to get away and not wanting to organise an escape. It can challenge whether spontaneity is the right behavioural lever. It can suggest ways to make the first step feel easier.
Then add a standard: reject routes that depend on unsupported claims, generic wanderlust imagery or a discount that could belong to any transport brand. Prefer concepts that can sustain a campaign platform across several executions.
This is an illustrative prompt progression. We have not run a controlled test, and no particular concept is being claimed as a proven winner. Each improvement gives the thinking a more useful problem to solve and a clearer basis for judgement; the result still needs to be evaluated.
A good idea isn't necessarily an original idea
There is a trap here. Tight constraints can produce more relevant ideas. They can also make work more conventional if the assumptions behind the brief are conventional.
Suppose the brief insists that younger audiences only respond to short-form social video. A model that follows that instruction perfectly may produce an excellent answer to a bad assumption. The output can be strategically tidy and creatively forgettable.
That's why a good AI-assisted creative process needs two kinds of challenge. First, challenge the definition of the problem. Then challenge the proposed solutions against it.
Interrogate the incoming brief and ask what evidence supports it. Ask which assumption would most change the answer if it proved false. Ask for competing interpretations of the same observation. Only after that should the team settle on the constraints that matter.
Originality and usefulness are related, but they aren't identical. A wildly novel idea can be wrong for the brand. A familiar mechanism can be effective when applied to a genuinely new strategic insight. Look for a strong idea that solves the right problem in a way worth noticing. Strangeness on its own says little about whether the idea belongs.
Strategy is creative work too
The neat distinction between rational strategy and imaginative creativity doesn't survive much contact with actual agency work.
Research can tell you what people say, do and believe. It rarely hands you the one connection that makes the problem look different. Someone still has to notice a pattern, combine observations from different places or ask whether everybody has been trying to solve the wrong problem.
That is a creative leap, even if it happens in a spreadsheet rather than an art studio.
AI can help because it is an unusually available thinking partner. It can compare interpretations, surface associations, challenge a first hypothesis and keep a conversation moving when a person is stuck. Sometimes the useful contribution is an answer. Sometimes it is a question that makes the strategist reconsider their own.
A thinking partner can still be missing information. The model does not know everything the agency knows or automatically understand which trade-offs the client is prepared to make. Those judgements still have to be made by people who own the consequences.
The danger isn't only average output
There's a more subtle risk than getting a mediocre list of ideas. A plausible answer can arrive so quickly that nobody notices how little thinking preceded it.
The team moves from the first brief to the first idea, then uses the remaining time to polish the presentation. Research becomes decoration. Strategy becomes a retrospective explanation for an idea somebody already likes.
That is the opposite of what AI should make possible.
The more useful application is to interrogate the brief, expose assumptions, explore competing directions, establish a standard for judging the strategy and then use the model to help stretch the thinking. Keep the right to reject everything it suggests.
An average prompt can produce average work. A better brief gives the work a better chance. The difficult part is still recognising what deserves to survive.
A sequence that improves model outputs makes those decisions easier to inspect. The same standard applies to a brief developed with Mirel: it should give the creative team a problem worth solving and room to find the answer.