In the modern digital advertising landscape, the pressure to maintain high-volume, high-quality output has never been greater. For small-to-medium enterprises, the traditional agency model—relying on expensive studio shoots and endless freelancer hours—is becoming increasingly unsustainable. This shift is driven largely by algorithmic evolution, specifically Meta’s "Andromeda" update, which has fundamentally changed how ad platforms evaluate performance. The old "spray and pray" tactic of running hundreds of near-identical creative variations no longer yields results; the platform now aggregates those variations, demanding genuine, distinct creative diversity to remain competitive.
For Fraser Cottrell, CEO of the direct-to-consumer agency Fraggell, the solution lies in a disciplined, three-step integration of generative AI. Far from being a "lazy" shortcut, Cottrell argues that leveraging AI for ad creative is a rigorous process that, when executed correctly, levels the playing field, allowing smaller brands to compete with industry giants for mere pennies on the dollar.
The Strategic Shift: Dispelling AI Myths
The industry is currently plagued by two primary misconceptions that prevent marketers from unlocking the potential of AI:
- The "Laziness" Fallacy: Critics often suggest that AI-generated creative is an attempt to cut corners. Cottrell pushes back, noting that coaxing high-quality, brand-aligned output from models like Claude or Gemini requires significant, intentional effort. It is not about automation; it is about iterative augmentation.
- The Quality Barrier: While early iterations of generative models produced inconsistent results, current technology has reached an inflection point. For static imagery, the gap between AI generation and professional photography is virtually non-existent. While video generation remains an area of active development, its current utility in scripting and concept ideation is already unparalleled.
Step 1: Building a Brand Knowledge Base Through Deep Research
Before an AI can generate a single headline or pixel of an image, it must understand the "soul" of the brand. Cottrell’s foundational step is to conduct a "Deep Research" session—a process that goes beyond simple keyword generation.
The Methodology of Deep Research
Utilizing LLMs like Google Gemini, marketers should prompt the AI to act as a market researcher. The goal is to build a comprehensive external profile of the brand. A robust research prompt must instruct the AI to:
- Analyze the customer base: Who is buying, and more importantly, why?
- Evaluate the "anti-customer": Why did those who encountered the product choose not to purchase?
- Synthesize sentiment: Scour platforms like Reddit to identify recurring complaints, common objections, and geographical clusters of demand.
To ensure the AI doesn’t provide a superficial summary, the prompt must explicitly demand that the model browse the internet thoroughly to build a multi-page, evidence-based document. Cottrell suggests using voice-to-text tools like Whisper Flow to dictate these complex requirements to an AI assistant, which can then draft a professional, multi-faceted research prompt.
Verification and Proprietary Integration
AI-generated research is a starting point, not a finished product. To verify the accuracy of the output, copy the document into Claude and instruct it to act as an interviewer. By having the AI ask you questions one by one, you can confirm key facts and immediately correct characterizations that deviate from reality.
Crucially, the "final" research document must be augmented with your proprietary, non-public knowledge. Your internal data—what your best customers say on sales calls, the technical nuances of your product, and your internal brand voice—must be manually injected into the AI’s document to create a source of truth that the open web cannot provide.
Step 2: Training a Dedicated Claude Project
Once the knowledge base is verified, it should be loaded into a "Claude Project." This environment acts as a persistent workspace that retains specific context, isolated from your other interactions.
The "Foundation" Contents
Every project should be loaded with the following pillars of data:
- The Verified Deep Research Document: The cornerstone of the project’s intelligence.
- Voice-of-Customer Data: Exported reviews and testimonials are gold mines. This data allows the AI to mirror the specific language, anxieties, and desires of your actual buyers.
- Internal Brand Guidelines: Every company should have a foundational document outlining its values, mission, and the specific definition of what constitutes a "good" ad.
- Performance Analytics with Visual Context: This is the most technical component. By taking your top-performing ads from the previous quarter and analyzing them—using tools like Poppy to assess visual pacing, on-screen action, and hooks—you can feed the AI the "why" behind your past successes.
By feeding these data points into a Claude Project, you create a "brand-aware" agent capable of evaluating new creative against the historical performance benchmarks of your specific audience.
Step 3: Executing the Creative Workflow
With the project trained, you can move into production. For static images, the most effective workflow is a hybrid approach: generate the visual with AI, but layer the copy manually. This ensures that you can test multiple headlines against a single visual without the need for time-consuming image regeneration.
Iterative Copywriting
Ask the Claude project to generate headlines based on the uploaded customer reviews. If the model provides four options, select two you like and two you dislike, providing explicit feedback on why. Because the project environment learns through conversation, this feedback loop progressively refines the model’s understanding of your brand’s specific tone and effectiveness.
Image Generation and Prompt Engineering
When generating images, specificity is the primary variable for success. If you are targeting marathon runners, provide the context: the target persona, the specific message, and the desired emotional outcome.
For product shots, simply uploading a photo of your product into the chat allows the AI to understand the object it is working with. Using natural language, you can then request a professional studio shot—specifying background colors, lighting, and textures—which the AI will translate into a precise prompt for image generation tools like Nano Banana 2 Pro.
AI in Video Ideation
While fully AI-generated video is still maturing, the AI’s ability to script and storyboard is world-class. By describing a scenario (e.g., a UGC creator running a marathon while discussing the product), the AI can produce a detailed, timestamped script in seconds.
While the AI draft is rarely a final product, it typically provides 30% of the heavy lifting. A human copywriter can then step in, refining the tone and ensuring the conversational flow matches human nuance, drastically reducing the time spent in the ideation phase.
Implications for the Future of Ad Agency Models
The integration of AI into ad creative represents a fundamental shift in the economics of digital marketing. The "leveling of the playing field" is not merely a metaphor; it is an operational reality. Small brands that previously struggled to produce enough creative to satisfy Meta’s algorithmic requirements can now generate high-quality, data-backed assets at a fraction of the cost.
This approach requires a move away from viewing AI as a "content engine" and toward viewing it as a "strategic partner." By investing time in the front-end—researching, training, and iterating—marketers move away from the frantic, manual production of ads and toward a scalable, intelligent creative engine. As the advertising ecosystem continues to reward genuine creative diversity, those who master these workflows will not only survive the transition but will thrive in an environment that finally favors strategy over raw budget.
