Adobe Patents a System That Ranks Ad Creative Elements by What Actually Worked
Every marketing team argues about what makes an ad work. Adobe is filing a patent for a system that settles those arguments with numbers, ranking every creative choice by its actual effect on past campaign performance.
How Adobe scores which ad ingredients drive real results
You're staring at a blank campaign brief, and your team is debating whether the headline should be upbeat or urgent, whether to use a photo or an illustration, whether the call-to-action belongs at the top or the bottom. Everyone has an opinion. Nobody has proof.
Adobe's patent describes a system that digs through your historical ad content and the performance numbers attached to each piece, then figures out which specific creative choices, a particular writing tone, a color palette, a video pacing, actually moved the needle. It doesn't just average things together; it runs a head-to-head comparison for each element, pitting ads that used that choice against all the ads that didn't.
The output is a ranked list: here are the content features that mattered most, and here are the specific versions of each feature that performed best. That list can then feed directly into Adobe's generative AI tools, so the AI starts from evidence rather than guesswork when building new creative.
A ranked set of attribute values is determined by scoring each attribute value using a comparative statistical test to compare the historical performance metric of each content asset with the attribute value to the historical performance metric of each remaining content asset without the attribute value.
Translation: It scores specific design elements by comparing ads that use them against ads that do not.
How the ranking engine compares content assets statistically
The system works in several steps, each handled by a named software component.
Step one: extraction. A machine-learning model scans a library of past content, pulling out attribute values (the specific choices made in each asset, like "headline uses a question" or "image shows a person" or "video is under 15 seconds") from text, images, and video alike.
Step two: attribute ranking. The system scores each broad content attribute, say, "headline tone" or "image subject matter," using a weighted combination of three measures:
- An absolute gain score (how much did this attribute shift performance overall?)
- A Kullback-Leibler divergence score (a statistical tool that measures how differently the performance numbers are distributed when this attribute is present versus absent)
- A comparative statistical test score (a formal significance check to rule out coincidence)
Step three: value ranking. For each attribute, every specific version is scored by comparing the performance of assets that used it against the performance of everything that didn't, using a comparative statistical test (a method that checks whether the difference between two groups is real or just random noise).
The final ranked lists can be shown to a human creative team or passed directly into a generative AI model as a creative brief grounded in evidence.
What this means for marketers using Adobe's AI tools
The practical payoff for marketers is a shorter feedback loop. Right now, figuring out which creative decisions drive conversions typically means running A/B tests that take weeks and cost real budget. This system proposes to mine your existing content library instead, surfacing patterns you already paid to generate.
Adobe's push to weave AI deeper into marketing workflows puts tools like this at the center of products like Adobe GenStudio, which is aimed squarely at enterprise marketing teams managing large content libraries. If the system works as described, it gives AI-generated content a data-backed starting point rather than a generic one, which is exactly the complaint most marketers have about AI creative tools today.
Adobe's fifth AI recommendation filing we've tracked since May builds on earlier applications like one picking email templates and one resizing graphic designs.
The problem this patent attacks is real and expensive. Marketing teams spend enormous amounts of time and money trying to figure out what made a past campaign succeed, and most of that analysis is either done manually, skipped entirely, or handed to expensive consultants.
The statistical approach here is credible. Using a proper comparison test to separate "this element correlated with good performance" from "this element actually caused it" is a meaningful upgrade over simple averages. The KL divergence layer adds another check against misleading patterns in uneven data.
The honest limitation is that the system is only as good as the historical data fed into it. For smaller brands with thin content libraries, or for anyone entering a new market or channel, past performance is a shaky foundation. The patent does not address that constraint, and it is worth keeping in mind when evaluating how broadly useful this would be in practice.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Adobe patent news breaks.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
5 drawing sheets from US 2026/0260262 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Be the first to weigh in