Better brand assets come from better production inputs

This anonymised case study focuses on the production outcome of machine-readable brand guidance: whether AI creates better work when it receives structured rules and task-specific skills. We tested four common asset types—a video script, event plan, social carousel, and proposal—under controlled information conditions. The same model completed every task, and independent reviewers scored the work blind against an answer key built from the approved brand guidance.

The structured conditions produced a clear improvement. Without guidance, AI followed around a third of the relevant rules and no asset was ready for sign-off. With the structured brand map, rule compliance reached nearly nine in ten. Reviewers judged nearly nine in ten pieces ready for first-pass approval, invented facts disappeared, and unsupported specifics fell sharply.

The production problem

Generative AI can create polished-looking material quickly, but surface quality is not the same as brand quality. An asset can read fluently while using the wrong terminology, missing a contextual rule, introducing an unsupported claim, or making a creative decision that cannot be explained.

The brand team needed to know whether structured guidance would improve real deliverables rather than simply improve answers to policy questions.

Four realistic asset briefs

The evaluation used four different production tasks:

a video script requiring language, narrative, and video guidance;
an event plan combining experience, messaging, and operational detail;
a social carousel requiring concise, sequenced content; and
a proposal applying brand language within a commercial document.

Each brief was completed twice under four conditions: no guidance, raw website pages, structured brand data, and structured data delivered through task-specific skills.

Two independent reviewers scored each asset without knowing which condition had produced it. The answer key was derived from the original brand site, giving reviewers a consistent basis for testing rule application, factual discipline, and readiness for sign-off.

The evaluation also included deliberately planted mistakes. This tested whether the operating skill could identify known failures rather than merely producing plausible copy.

What changed with structured data

Unguided outputs followed only around a third of the relevant rules. Raw pages helped the model sound more plausible, but important decisions remained inconsistent and difficult to trace.

Structured data more than doubled rule compliance. Nearly nine in ten relevant rules were followed, and nearly nine in ten outputs were judged ready for first-pass sign-off. Only the structured conditions removed invented facts entirely. Unsupported specifics fell by more than half compared with raw website pages.

Why skills improved production further

The task-specific skills loaded only the guidance needed for each brief. This reduced irrelevant context and gave the model an operating method for applying the structured rules.

The skills condition achieved the highest sign-off rate and made none of the planted mistakes. The result showed that better asset production depends on both layers: structured data supplies dependable brand knowledge, while the skill determines how that knowledge is selected and applied during the task.

Traceability for creative review

Every structured decision could be linked to the relevant rule and source. Reviewers could see why language had been chosen, why a correction was recommended, or which condition governed an asset.

This changes the review conversation. Instead of debating whether an output feels right, the brand team can inspect whether the approved rule was applied correctly and decide whether an exception is justified.

Outcome

The evaluation demonstrated that machine-readable brand guidance can improve both production quality and review efficiency. More assets reached a first-pass sign-off standard, factual failures reduced, and decisions became easier to verify.

The work provides a controlled foundation for extending AI-assisted production into additional asset types. Future evaluations can reuse the same answer keys and measures as the brand data, skills, and models evolve.

What this case study demonstrates

AI does not produce better branded assets simply because it has seen more brand content. Quality improves when approved guidance is structured into usable data, delivered through the right task skill, and evaluated against explicit rules before the work reaches production.

Brand production: Higher-quality assets from structured guidance