In the modern digital advertising landscape, the pressure to produce high-volume, high-quality creative has never been more intense. With platforms like Meta shifting away from the strategy of "testing hundreds of minor variations" toward a more consolidated approach following the Andromeda update, marketers are at a crossroads. The choice is binary: either burn out your design team with relentless production demands or evolve your workflow through the intelligent application of Generative AI.
Fraser Cottrell, CEO of the direct-to-consumer ad agency Fraggell, argues that the fear of AI-generated content is misplaced. The real challenge, he suggests, isn’t the technology itself, but the lack of human-led context provided to these models. By integrating AI into a structured, research-heavy pipeline, brands can move from resource-heavy studio shoots to agile, data-informed production that costs cents on the dollar.
The Misconception of AI as a Shortcut
Many marketers shy away from AI in creative workflows due to two pervasive myths. The first is the notion that AI is "lazy"—a tool for cutting corners. Cottrell pushes back, noting that achieving high-end output requires significant "prompt engineering" and iteration. The second myth involves quality: that AI produces derivative or amateurish visuals. In truth, modern generative models are capable of producing static imagery that is virtually indistinguishable from professional photography. While video generation still faces hurdles regarding consistency, the barrier for static-image ads has effectively vanished.
For small to mid-sized e-commerce brands, this shift is revolutionary. What once required thousands of dollars in studio rentals, models, and photographers can now be executed through precise AI prompting, leveling the playing field between boutique brands and established market giants.
Step 1: Deep Research as the Foundation
The most common failure point for AI adoption in marketing is skipping the training phase. Before generating a single image, brands must build a comprehensive "Brand Knowledge Base."
The Deep Research Protocol
Cottrell advocates for using LLMs (Large Language Models) like Google Gemini to conduct deep, internet-wide research. Unlike a standard search, a "Deep Research" prompt instructs the AI to synthesize a comprehensive external profile of the brand.
The core objectives include:
- Customer Sentiment: What are the drivers for purchase versus the reasons for abandonment?
- Competitive Landscape: What are the most frequent complaints found on platforms like Reddit or industry-specific forums?
- Geographic Concentration: Where does the brand’s most vocal customer base reside?
To standardize this, Cottrell uses voice-dictation tools like Whisper Flow to feed complex instructions into models like Claude or Gemini. By asking the AI to act as an objective researcher, brands can identify the specific pain points and objections that make an ad resonate with a target audience.
Verification and Human Synthesis
Once the AI returns a comprehensive document, accuracy is paramount. A critical technique is to use a secondary AI—Claude—as a "verifier." By pasting the research document into a Claude session and asking it to quiz you, the human, on the accuracy of the facts, you can identify hallucinations or mischaracterizations.
Finally, you must inject the "proprietary human element." AI can scrape the internet, but it cannot know your internal data, the nuances of your product development, or the stories your customer service team hears on the phone. Blending this internal knowledge with AI-driven market intelligence creates a robust, unique brand DNA.
Step 2: Training a "Claude Project"
Once your research is verified, the next step is to load this data into a "Claude Project." Unlike a standard chat, a Project acts as a persistent workspace with its own memory. Every interaction within this environment is grounded in your uploaded knowledge base.
The Essential Data Inputs
To make your AI workspace truly effective, you must provide the following:
- The Deep Research Document: The foundation built in Step 1.
- Voice-of-Customer Data: Exporting customer reviews, testimonials, and support tickets into CSV files. This provides the AI with the literal language your customers use.
- Internal Brand Guidelines: A manifesto detailing what the brand stands for, its visual identity, and its definition of a "good ad."
- Performance Data: By feeding in the top-performing ads from the previous quarter, you provide a benchmark. Using tools like Poppy, you can have the AI "watch" video ads to analyze pacing, visual hooks, and on-screen elements, translating performance data into creative intelligence.
Step 3: Execution and Iteration
With a trained environment in place, you are ready to produce assets. Cottrell emphasizes a "hybrid approach" for image generation: generate the visual via AI, but layer the text manually. This ensures that you can test multiple headlines against a single visual without the need for constant regeneration.
Brainstorming and Visual Generation
When interacting with your Claude project, specificity is your greatest asset. Rather than asking for a generic ad, provide a brief that includes the target persona and the specific message.
For instance, a prompt for a sports brand might be: "I want an image targeting marathon runners with the message that they need to be properly hydrated before they run, so they should buy [product]."
When requesting an image, drag a photo of your actual product into the chat. This allows the AI to understand the product’s physical characteristics. A prompt such as, "Give me a prompt for [Product] to use for a professional studio product shot on a purple background, with soft lighting that looks real," yields far more usable results than a vague request.
The Role of AI in Video Ideation
While current AI video generation may not yet meet the high-fidelity standards required for premium commercial use, it is an unparalleled tool for scripting. Using the same Claude project, you can generate 30-second UGC (User Generated Content) scripts that are grounded in your brand’s voice.
While the AI draft is rarely a final product, it effectively moves a creative director 30% of the way to the finish line, allowing human writers to focus on refining tone and emotional resonance rather than staring at a blank page.
Implications for the Future of Ad Agency Work
The shift toward AI-integrated creative has profound implications for the industry. Agencies like Fraggell are pivoting to focus on strategy, research, and the "human edit," while offloading the heavy lifting of asset generation to machines.
Why This Matters Now
Meta’s algorithmic updates have fundamentally changed how ads are served. The platform no longer rewards the "spray and pray" method of launching hundreds of near-identical creative variations. Instead, the algorithm now demands high-quality, diverse creative that provides distinct value.
For the modern marketer, this means that speed to market and the ability to pivot based on real-time data are the new competitive advantages. By training an AI on deep, verified brand knowledge, marketers can produce high-quality variations at a scale that was previously restricted to firms with massive budgets.
Conclusion: The Human-AI Partnership
The future of ad creative is not an automated one—it is an augmented one. The most successful brands will be those that view AI not as a replacement for human creativity, but as a tireless, highly-informed research assistant and production aide. By building a knowledge base that is uniquely your own, you ensure that while your production speed increases, your brand voice remains consistent, authentic, and effective.
As the industry continues to evolve, the ability to synthesize vast amounts of customer data into actionable, visual, and written creative will define the next generation of marketing excellence. The tools are ready; the question is whether your brand is prepared to do the foundational research required to command them.
