The most urgent question in paid media right now is not how to make more ads with AI. It is how to know which ads to make. At Smart Marketer, our team has been building an AI creative analysis system that answers that question at a depth and speed that simply was not possible before, and the early results are changing how we think about creative strategy entirely.
KEY TAKEAWAYS
- AI can now analyze the qualitative dimensions of an ad, including hook type, emotional motivator, persona, and story framework, and turn those dimensions into structured, searchable creative metadata.
- The real creative edge is not in generating ads. It is in knowing exactly what to generate and why, informed by what has already worked.
- Google Gemini is currently the only large language model capable of analyzing full video content, which makes it the foundation for any AI creative analysis system that includes video ads.
- A well-built creative analyst skill compounds over time. Each round of analysis makes the system smarter, and those gains do not have to be rebuilt from scratch.
Why Are Most People Using AI for Creative the Wrong Way?
Most brands are using AI to produce more creative. That is useful, but it misses the bigger opportunity. The real competitive advantage is using AI to understand what your existing creative is actually doing and why it’s working or failing, before you ever generate a single new ad.
The assumption driving most AI creative adoption is that volume creates winners. Make 500 ads, and by probability, a few will perform. That may technically be true. But testing 500 ads costs money, and most of those ads are not informed by any real intelligence about what your specific market responds to.
Our team has been building toward something different. The goal is a system that reads an ad account the way a great creative strategist would, with the same depth applied to qualitative creative elements that data pipelines bring to performance metrics. That means understanding not just which ads spent money, but why the ones that worked actually worked.
This is the part of AI creative that most people are missing. Tools like Thumbstop and other AI creative generators are genuinely useful for producing assets quickly. But the strategy layer, the system that tells you what to build and why, is where the real differentiation lives.
What Is Creative Metadata, and Why Does It Matter for Ad Performance?
Creative metadata is the structured, qualitative data that describes an ad above and beyond its performance numbers. It captures things like hook type, visual style, persona demographics, emotional motivator, and story framework, giving teams the ability to analyze creative the way Meta’s Andromeda system already does internally.
Meta’s advertising platform already knows, at a machine level, that a certain user responds to women her age with a southern accent and an urgency-based message. Meta knows what visual styles, tones, and personas resonate with every segment in its system. Until now, advertisers had no way to track that same layer of information on their own side of the equation.
Creative metadata bridges that gap. By indexing the same qualitative dimensions Meta is using to match ads to audiences, advertisers can start to see those patterns from the inside and make better decisions about what to build next.
The challenge has always been that this kind of analysis is extremely time-consuming to do manually. A skilled strategist could annotate every ad in an account with hook type, persona, style, and awareness level. But nobody would ever do that at scale. AI changes the equation completely.
How Does the Creative DNA Sheet Work?
The Creative DNA Sheet is an AI-generated intake document that describes every element of an ad the way a human strategist would experience it. A single one-minute video can produce three pages of structured creative metadata, covering everything from hook type and persona demographics to story framework and emotional motivator, with no analysis attached to the intake stage at all.
The first step in the system is what the team calls the intake layer. This is not analysis. It is pure description. The goal is to get the AI to see the ad the way a person sees it, understanding the performance, the emotional tone, the authenticity of the creator, the relevance of any trends used. Those are the things a human registers when watching an ad, and they are exactly the things that drive a response.
Getting that kind of qualitative output from an AI model requires significant prompt engineering. Our team has put over a month of concentrated development time into building the intake prompts that produce genuinely useful creative metadata. The short version: it is harder than it looks, and Google Gemini is currently the only major large language model capable of ingesting and analyzing video directly, which makes it the right tool for any creative library that includes video assets.
Here is a breakdown of the dimensions the Creative DNA Sheet captures:
| Dimension | What It Captures | Example |
| Hook | The literal opening words or image that initiates the ad | “Tired of scrubbing grout for hours?” |
| Hook Type | The strategic category the hook falls into | Problem/Solution, Novelty, Demonstration |
| Content Type | The creative format | Video, Static Image, Carousel |
| Awareness Level | The funnel stage the ad is designed for | Cold, Warm, Bottom of Funnel |
| Style | The visual and tonal aesthetic | UGC Authentic, Polished Brand, Educational |
| Persona | Demographics of the person shown or addressed | 35-year-old woman, Southern US, homeowner |
| Emotional Motivator | The underlying emotion being activated | Pride, Urgency, Curiosity |
| Story Framework | The narrative structure used (e.g., StoryBrand) | Guide/Hero, Problem-Solution-Resolution |
One of the most important structural decisions in this workflow is that the intake document is created once and stored. Future analysis can always reference it. The upfront work of generating the creative brief becomes a reusable asset that compounds in value over time. That separation of intake from analysis is something most people skip, and it is one of the reasons early-stage AI creative systems produce inconsistent results.
What Does the Creative Analyst Claude Skill Actually Do?
The Creative Analyst is a Claude skill Ben created that combines Smart Marketer’s creative frameworks, including content from Train My Traffic Person and the Creative Mini Class, with the structured intake data from the DNA Sheet to generate insights about what is actually driving performance across an ad account.
Once the creative DNA is captured, the next step is running analysis. The Claude skill is built to think like a data scientist, not just a creative strategist. It is prompted to look for cross-variable patterns, surface correlations between qualitative creative dimensions and performance metrics, and flag findings that would contradict conventional wisdom.
The first outputs from this kind of analysis are often obvious. The ad that received 75 percent of account spend is performing well. That is not a revelation. The real value emerges as the system is pushed to look harder: to find correlations between story framework and click-through rate, to surface personas in underperforming ads that suggest a targeting gap, to identify what is not working that most marketers would assume should be working.
In a recent analysis of a client account, the system identified that scarcity-based hooks were consistently underperforming at the bottom of the funnel. That finding contradicts standard direct response logic. But it opens up a genuinely useful strategic question: is it that scarcity does not work for this audience? Or is it that this brand simply does not have strong scarcity creative? The system makes the question visible, which is more valuable than any answer the data alone could provide.
The version of the skill that produced that report had been through 11 rounds of iteration. Each round of feedback improved the depth and usefulness of the output. That work is done. Every future analysis for every account builds on it.
How Do You Start Building an AI Creative Analysis System?
If you are interested in building something like this for your own brand or agency, here is how the process breaks down. This is not a quick setup, but each piece of work you do is additive, and the system gets more valuable with every client it touches.
- Start with Claude Code. Not the standard Claude.ai interface. Claude Code is accessible, learnable in about 15 minutes of YouTube tutorials, and opens up a level of capability that a chat interface simply cannot match. It is the foundation of a system like this.
- Separate intake from analysis. Build two distinct layers. The intake layer describes what is in the ad. The analysis layer interprets the data. Keeping these separate makes both layers more reusable and the whole system more reliable.
- Define your taxonomy before you build. Nail down the variables you want to track before you start prompting. If you cannot decide between options on a dimension like style, build a closed list of choices and add an open field that allows free description. The combination of both gives you structure and flexibility.
- Use Gemini for video. If your creative library includes video, Google Gemini is currently the only major language model capable of analyzing full video content. That capability changes the scope of what this system can do, especially for DTC brands running UGC-heavy ad accounts.
- Train your analyst skill on your own frameworks. Feed in your creative principles, your brand documents, your historical strategy resources. The richer the context, the more useful the outputs. For teams with access to Smart Marketer’s courses and resources, those documents form the foundation of the Creative Analyst skill.
What Comes Next for AI Creative Systems?
The next milestone is closing the loop. That means moving from analysis and brief generation to finished creative output, produced by AI and informed by everything the system has already learned about what works for a specific brand and audience.
The current system already produces recommended ad briefs. It identifies patterns in the creative DNA data, surfaces the dimensions that correlate to strong performance, and generates direction for what to build next. The final step is connecting that brief output directly to the creative production pipeline, so that the intelligence gathered from historical ads feeds automatically into the generation of new ones.
That is fundamentally different from generating volume for volume’s sake. The appeal of AI-generated creative at scale has always been probability: make enough ads and a few will win. But that model requires significant testing budget and produces a lot of noise in the process. The goal of a well-built system is the opposite. Fewer ads, each one informed by genuine intelligence about the account, the audience, and what has actually produced results.
The system compounds. Each new analysis makes the creative analyst skill smarter. Each round of feedback adds to the accumulated creative intelligence. That is the long-term value, not any single insight or report, but the growing body of knowledge about what works for every account the system touches.
If you manage paid creative for a direct response brand, the question worth sitting with is not whether AI should be part of your creative workflow. That conversation is over.
The question is whether you are using it to do the thing that actually matters: understanding what your creative is doing and why, at a level of depth that simply was not possible before.
Our team at Smart Marketer Agency is implementing this system across client accounts now. If you want to see how it might work for your brand, visit smartmarketer.com/agency to learn more about working with us.
