Generative AI in brand production: what our 2026 projects show.
We reviewed every studio project that used generative tools this year. Speed went up everywhere. Quality only went up where a system came first.
Generative tools are now in almost every brand pipeline we see. The question we set out to answer was simple: when do they make the work better, and when do they only make it faster?
This report draws on our own delivery records for the first half of 2026: identity, campaign, content and interface projects where generative tools played a real part in production. We looked at time from brief to approval, rounds of revision, how often work was sent back for brand corrections and how clients rated the final result. It is a view from inside one studio, not a market survey, and we present it in that spirit.
Four findings
1. Speed improves almost everywhere
First drafts arrived much sooner on nearly every project. Concept boards, image explorations and copy variants that once took days were ready in hours. This is the finding everyone expects, and it held across sectors and project types.
2. Quality improves only where a system exists
On projects where the brand already had clear tokens, templates and a written grammar, generative output needed fewer corrections and clients rated it higher. Where the brand lived only in a PDF, the tools produced more work but not better work, and the time saved on drafts was often spent again on fixing drift.
3. Review becomes the bottleneck
When drafts become cheap, judgement becomes expensive. Several projects stalled not in production but in approval, because reviewers faced ten times the options with the same hours in the day. Teams that agreed selection criteria before generating anything moved far faster.
4. Provenance is now a client requirement
Clients in regulated and premium sectors increasingly ask where each asset came from, which model touched it and who approved it. Projects that logged this from the start answered in minutes. Those that did not faced an awkward reconstruction exercise.
Generative tools multiply whatever system they are given. Give them none and they multiply the mess.
Where the tools helped most
- Early exploration: wide visual territories in a day, so clients react to real options sooner.
- Variant production: resizing, reformatting and localising approved work across channels.
- Copy alternatives: headline and caption options within a defined tone of voice.
- Pre-visualisation: rough scenes for films and 3D work before expensive production begins.
Where they helped least
- Defining a new identity: the core idea still came from people arguing in a room.
- Typography: generated letterforms rarely held up at text sizes or across languages.
- Final hero imagery for premium brands, where small flaws undermine trust.
- Any task where the brief itself was unclear; the tools amplified the confusion.
What the best projects did differently
The projects with the strongest results shared a handful of habits. None of them is complicated, and all of them were in place before the first image was generated.
- They encoded the brand first: tokens, templates and prompt libraries agreed before production.
- They wrote selection criteria into the brief, so reviewers knew what good looked like.
- They kept a person responsible for every approval, by name, with the decision logged.
- They checked output automatically for contrast, logo use and layout before human review.
- They kept a record of provenance for every published asset.
What changed for the people doing the work
The shift was felt most by designers and art directors. Less of their week went on producing variants and more went on framing problems, setting constraints and judging output. Several told us the work had become more demanding, not less, because every decision now shaped hundreds of assets instead of one. The roles that grew in value were the ones that combine taste with the ability to explain it clearly enough for a system to follow.
What this means for brand leaders
The lesson is not that generative tools are overrated. They are already indispensable for volume work, and they will get better. The lesson is that their value depends almost entirely on what they are plugged into. A brand with a clear, machine-readable system gets faster and better at once. A brand without one gets faster at producing things it then has to fix.
For most organisations the highest-return investment this year is not another tool licence. It is the unglamorous work of writing the brand down in a form both people and machines can follow, deciding who approves what, and measuring whether the output is getting better as well as quicker.
A note on method
Our review covered projects delivered by the studio between January and June 2026 where generative tools contributed to production, not only to internal exploration. Figures were compared with similar projects from the previous year. Client ratings came from our standard close-out survey. We will repeat the review at the end of the year and publish what changes.
