Two claims about AI in creative work are both wrong. The first is that it now produces finished campaigns. The second is that it produces nothing worth using. The practical reality sits between them: generative tools have become genuinely useful at the exploratory and repetitive ends of production, and largely unhelpful in the middle, where judgement lives.
Where it earns its place: exploration
Early-stage visual thinking is expensive in time and cheap in commitment. A key visual direction for a summit, a stage concept, a mood for an awards evening — these normally arrive as a handful of references and one or two sketched routes, because producing more is uneconomic.
Generative tools change that arithmetic. A team can put twenty considered directions in front of a client instead of three, and — more importantly — can test whether a direction survives contact with different formats before anyone builds anything. The value is not the image; it is the speed of arriving at a decision.
Where it earns its place: volume and adaptation
The other clear fit is downstream. A single campaign frequently needs dozens of legitimate variations: platform ratios, language versions, session-specific graphics, delegate assets, recap formats. This work is necessary, precise and repetitive.
- Resizing and reformatting an approved visual system across channels.
- Drafting first-pass copy variants for a human editor to sharpen.
- Building storyboards and pre-visualisation before a shoot is committed.
- Turning long-form event footage and transcripts into structured content packs.
Where it does not belong
Strategy, brand judgement and factual claims. A model can generate a confident line about market leadership; it cannot know whether the claim is true, defensible, or appropriate for the organisation making it. It can generate a crowd of delighted attendees; presenting that image as a photograph of a real event is a straightforward misrepresentation.
AI is part of the workflow — not a replacement for creative judgement.
Three rules keep this straightforward in practice. Generated imagery is used for concept and abstract work, and is labelled as such. Anything presented as documentation of a real event is photographed. Facts, names, numbers and quotes are verified by a person before publication.
The workflow that actually works
A useful sequence looks less like automation and more like a studio with a faster sketchbook:
- Brief: a human defines audience, objective and constraints.
- Explore: AI-assisted generation of directions, moodboards and variations.
- Select: a creative director chooses and edits — usually discarding most of it.
- Craft: the chosen direction is rebuilt properly, with real typography, real assets and real photography where required.
- Review: brand, legal and factual checks by people who are accountable for them.
What clients should ask
Anyone commissioning creative work in 2026 is entitled to know how it was made. Reasonable questions include: which parts of this were AI-assisted; is any imagery generated rather than photographed; who reviewed the claims; and what happens to our brand assets and confidential material in the tools being used.
Studios that answer those questions plainly tend to produce better work, because transparency forces discipline about where the tools help. Used that way, generative production does not dilute craft. It removes some of the friction that previously stood between a good idea and seeing whether it holds.