Published May 16, 2026 · By CampaignsLive · Insights
Half a million campaigns is enough to make patterns stop being noise. When you study a dataset that large at once — not one campaign at a time, the way most of us see them in the wild, but the whole archive in cross-section — the work that performs starts to look different from the work that does not. Not in the ways the industry usually talks about, either. The patterns are quieter than that.
Here is what the archive shows.
Hero compositions cluster around a small number of structures
Brand campaign creative is more disciplined than it looks. The visual language is wide, but the compositional grammar underneath is narrow. Hero frames cluster around a handful of structures — the centered subject, the rule-of-thirds talent shot, the negative-space wide, the off-axis environment, the product-on-hand close-up. Once you can see these structures as a finite set, the question of why most campaign creative reads as professional and most generic AI imagery does not stops being mysterious. Professional creative is composed inside a known grammar. Generic AI imagery is composed inside a much larger one — closer to the open internet than to advertising.
The work in the dataset that performs almost never invents a new composition. It re-uses a structure with discipline and lets the brand and the moment do the differentiation.
Color is more disciplined than people think
A common belief inside agencies is that brand color guidelines get violated all the time in the wild. The archive does not support that. Across categories, geographies, and channels, palette discipline is high. The biggest brands hold to two to four core hues across every placement type, and the variance you do see is mostly tonal — the same brand red, but warmer in summer creative and cooler in fall.
What changes across years is the role of color, not the colors themselves. A 2019 social tile is more likely to lead with full-saturation brand color; a 2024 one is more likely to lead with a documentary palette and let brand color sit in supporting elements. That shift, watched at scale, is one of the most visible markers of how visual language has changed under the pressure of feed contexts.
Typography is the most under-managed lever
If color is the area where brand teams over-index on discipline, typography is the area where they under-index. The archive contains tens of thousands of campaigns where the visual identity is unmistakable in image and ambiguous in type. Many brands run the same headline treatment across formats that have very different reading distances. Many run a single weight across hero and supporting copy. Many use no display face at all, leaning on the sans-serif their CMS happens to ship with.
The campaigns that perform tend to treat type as a part of the composition rather than a layer on top of it. Type that is sized for its viewing distance, weighted to its role, and integrated into the image — instead of pasted on it — is consistently associated with better-performing work.
Cross-format consistency is the silent quality marker
Almost any campaign looks competent inside its own format. The harder thing — and the thing the archive shows the strongest brands all do — is hold consistency across formats. The 48-sheet, the bus shelter, the press execution, the digital companion, and the social tile of the same campaign should read as one campaign, not five executions of the same brief.
When you study sets at scale, the brands that do this well share a working assumption: the same parent composition travels, with only the elements that have to change actually changing. Talent direction, palette, framing logic — those hold. Aspect ratio, peripheral elements, type sizing — those adapt. The work that breaks consistency tends to do so by reinterpreting the parent at each placement, which is what most agencies are structurally set up to do.
The briefs that produce the best work are the most specific
A reasonable counter-intuition is that loose briefs unlock the most creative work. The archive disagrees. The briefs that produce the strongest output tend to be unusually specific — about audience, about the moment, about the channel, about the visual register the brand is committing to in this particular execution.
The mechanism is not mysterious. Specificity is a forcing function. A brief that says “tell the story of the new model in a documentary register, aimed at thirty-five-year-old urban drivers who already own a competing car, in formats that will run primarily on TikTok and on transit panels in three European capitals” leaves very little room for the safe-default version of the work. A brief that says “make it premium” leaves nothing but room for the safe default.
What this means for what we build
A generation system that learns from an archive like this — instead of from the open internet — does not produce more creative. It produces creative inside a tighter visual grammar. The compositional choices come from a known set; color sits in the disciplined range the archive teaches; typography earns its place in the frame instead of floating on top; cross-format work travels because that is what the parent material rewards.
That is not the only difference between brand-grade AI creative and the AI imagery flooding the rest of the internet, but it is the most important one. The output is a function of the training. The patterns the system has seen become the work it produces. Half a million proven campaigns is a starting point that the open internet, however large, simply cannot match.