AGI for Creative Teams

There is one piece of research every creative team should know, and most have not read it. It is causal, peer-reviewed, and it says two things at once that are uncomfortable together.

Better individually, more similar collectively

Doshi and Hauser, published in Science Advances, gave writers access to AI story ideas and had external judges rate the results. With maximum AI access, novelty rose 8.1% and usefulness 9.0%. Among the least creative writers the gains were larger: novelty up 10.7%, usefulness up 11.5%, and up to 26.6% better written.

And the collective cost: stories in the AI-assisted condition were about 10.7% more similar to each other.

AI raises the floor of individual creative output and lowers the ceiling of collective variety. Both effects are real and they are in tension.

One more finding from the same study deserves attention, because it undermines how most teams evaluate this: participants using AI did not rate their own work as more creative, even though external judges did. Self-assessment of AI-assisted creative work is unreliable in both directions.

What that means in a studio

If everyone in your category is prompting similar models with similar briefs, the homogenisation is not a hypothetical. It is the measured default. The output that clears the bar will be output that started somewhere the model could not have gone: a real constraint, a specific client truth, a visual decision made before anything was generated.

That reframes where the craft sits. Not in producing the asset, which is now cheap and converging, but in the decisions upstream of it and the judgement applied downstream.

Adoption is near-total, depth is not

The IAB found 83% of ad executives have deployed AI in creative processes, up from 60% in 2024. Surveys of creative professionals put daily use around half, higher among agency owners.

But McKinsey's finding across 1,719 organisations applies here as hard as anywhere: 80% of individuals report productivity gains while only 37% of organisations report any earnings impact, and 73% of the ones getting real results fundamentally redesigned their workflows against 25% of everyone else. Faster asset production inside an unchanged process produces more assets and the same outcome.

The disclosure problem nobody has solved

This one is a live commercial risk rather than a philosophical question. In one survey of creative professionals, 58% said they had used AI without disclosing it to a client, and only 31% always disclose. Nearly half said they did not see why every tool needs disclosing.

Meanwhile a meaningful minority of clients explicitly want human-created work, and the EU AI Act's transparency obligations came into application on 2 August 2026, requiring machine-readable marking of AI-generated content and disclosure at the moment of contact rather than buried in a footer. Penalties run to €15 million or 3% of worldwide turnover.

An undisclosed practice that 58% of a profession quietly shares is not a secret. It is an unpriced liability.

What consumers actually think

Worth knowing before anyone builds a campaign around AI as a selling point. The IAB found 82% of executives believe young consumers feel positive about AI-generated ads. Only 45% of those consumers do, a 37-point perception gap, widened from 32 points in 2024.

The halo is also fading. Consumers describing AI-using brands as innovative fell from 30% in 2024 to 23% in 2026, while advertiser belief in AI as an innovation signal rose from 40% to 49%. The two are moving in opposite directions.

The junior pipeline is the structural story

US Census and BLS data show workers aged 20 to 24 fell to 6.5% of advertising and PR jobs in 2024, down from 10.5% in 2019. Survey work reports 57% of agencies having slowed or paused entry-level hiring.

There is a split inside that worth noticing: in one survey 59% of performance and media-heavy agencies planned significant headcount reductions within three years, against only 21% of creative and branding-focused agencies. The automation is hitting execution volume, not concept work.

And the holding-company numbers refuse to tell a simple story. WPP cut 10,656 roles, nearly 10%, between December 2024 and June 2026. Over roughly the same period Publicis added 5,900 people and grew organic revenue 4.8% in Q2 2026. Same market, same tools, opposite direction.

Where this leaves a creative team

Three things follow from the evidence rather than from the discourse.

• Treat AI output as a starting point, never a deliverable. The homogenisation finding is causal. If the model's first answer ships, you are shipping the category average.
• Redesign the process, not just the tools. The 73%-versus-25% gap is the clearest predictor of whether any of this reaches the bottom line.
• Decide your disclosure position now and write it down. It is becoming a legal requirement in one major market and a trust variable in every market.

The uncomfortable summary is that AI makes competent work easier to produce and distinctive work harder to notice. Which of those you are selling determines whether the last two years helped you.