In the current digital advertising landscape, the barrier to entry for high-quality creative has shifted from capital expenditure to intelligent curation. As Meta’s algorithms evolve—specifically following the industry-shifting Andromeda update, which rendered the strategy of running hundreds of micro-variations obsolete—marketers face a new mandate: produce high-volume, high-quality creative that resonates deeply with specific audiences.
For many, this creates a paradox. How can a small brand maintain the creative cadence required by modern algorithms without burning out their design teams or exhausting their production budgets? According to Fraser Cottrell, CEO of the direct-to-consumer agency Fraggell, the solution lies in a disciplined, three-step framework that leverages generative AI not as a shortcut, but as a sophisticated production engine.
The Misconceptions of AI-Driven Creative
Before implementing an AI-centric workflow, marketers must shed two persistent misconceptions that often stifle innovation. The first is the belief that using AI for creative work is inherently “lazy.” In practice, getting an AI model to produce output that is brand-consistent and strategically sound requires significant human input, iterative prompting, and expert oversight.
The second misconception concerns output quality. While early generative models were criticized for aesthetic inconsistencies, current iterations—particularly in image generation—produce assets that are nearly indistinguishable from professional photography. While video generation is still maturing, the static-image sector has reached a tipping point where quality is no longer the bottleneck. The real challenge is context: AI models are only as effective as the data and instructions they are fed.
Step 1: Building a Foundational Brand Knowledge Base
The most critical error marketers make is attempting to generate ads without first providing the AI with a "north star." Without a foundational understanding of the brand’s identity, target audience, and past performance, AI-generated creative will lack the nuance required to drive conversions.
The Deep Research Protocol
Fraser Cottrell advocates for a "deep research" methodology. Unlike a standard search engine query, deep research prompts utilize Large Language Models (LLMs) to scan the internet, synthesize information, and produce comprehensive analytical documents.
The goal is to create an external profile that answers fundamental questions: Who is buying this product? Why are they buying it? Crucially, why are people who encounter the brand choosing not to convert? By analyzing Reddit threads, customer complaint logs, and geographic market concentrations, marketers can surface the specific pain points and objections that make an ad copy resonate.
To execute this, practitioners can utilize voice dictation tools (like Whisper Flow) to instruct AI models such as Google Gemini—which is generally preferred for its speed and access to live web data—to compile these findings. Once the document is generated, it must be verified. A highly effective technique involves uploading the document to an alternative LLM, like Claude, and instructing it to conduct a "Socratic interrogation." By asking the human to confirm key facts and clarify discrepancies one by one, the AI ensures the final knowledge base is both accurate and comprehensive.
Blending External Insights with Internal Truth
The final step in this phase is the integration of proprietary knowledge. While the AI can scour the public web for sentiment, it cannot access internal CRM data, unique product nuances, or the qualitative insights gained from customer service calls. Manually injecting this "tribal knowledge" into the research document creates a hybrid foundation that captures both the market’s perception and the brand’s reality.
Step 2: Training a Dedicated Claude Project
Once the research is finalized, the data must be housed in a way that ensures consistency. Claude Projects offer a dedicated workspace where the AI retains a "persistent memory." This prevents the "bleeding" of information from unrelated chats and ensures that every creative generation is informed by the brand’s core documentation.
Essential Components of the Knowledge Base
A well-trained project should include:
- The Verified Deep Research Document: The core summary of market sentiment and competitive positioning.
- Voice-of-Customer Data: Exported reviews and testimonials. Language used by actual customers is the most potent input for ad copy, as it mirrors the vernacular of the target audience.
- Internal Brand Guidelines: A manifesto or "style guide" that defines the brand’s tone, mission, and the specific characteristics of what constitutes a "winning" ad.
- Performance Data and Visual Analysis: Using tools like Poppy—which can analyze visual pacing and screen action—marketers can upload their top-ten performing ads from the previous quarter. By feeding these insights into the project, the AI learns not just what was said, but the visual styles and cadences that have historically driven results.
Step 3: The Hybrid Workflow for Production
With a centralized knowledge base in place, the production process becomes a systematic, repeatable operation.
Static Image Creation
Rather than relying on AI to generate entire ad designs with embedded text, Cottrell recommends a hybrid approach. Generate the base imagery through AI models—such as Nano Banana 2 Pro—and handle the typography and layout manually. This modularity allows for rapid testing: marketers can iterate on headlines and copy variations against the same visual without the need for constant re-generation.
The prompt engineering process here is iterative. Start by asking the AI to draft headlines based on the project knowledge. If the results are suboptimal, provide specific feedback: “Use more of this tone, and less of that one.” Because the project workspace remembers previous interactions, the AI’s suggestions become progressively more refined over time.
Scripting for Video
While fully AI-generated video has yet to meet the high standards required for top-tier DTC advertising, AI is an indispensable tool for scripting and ideation. By describing the target persona and the intended message to the Claude project, marketers can generate high-quality, timestamped scripts in seconds.
While the output requires a human touch to ensure emotional resonance and conversational flow, it provides a "30% head start." This allows human copywriters to focus on the high-level narrative structure rather than the tedium of drafting from a blank page.
Implications for the Modern Ad Landscape
The implications of this shift are profound. The ability to generate high-quality creative at scale levels the playing field, allowing smaller brands to compete with larger competitors who previously held an advantage due to massive production budgets.
As Meta’s algorithm continues to prioritize the performance of individual creative assets, the ability to rapidly test, iterate, and deploy varied messaging is no longer a luxury—it is a survival requirement. By building a systematic, AI-informed production process, brands can ensure they remain responsive to market shifts, customer objections, and the evolving requirements of digital advertising platforms.
The future of ad creative is not "AI versus Human," but rather a synthesis: the AI provides the volume, data-backed research, and speed of ideation, while the human provides the strategic oversight, brand integrity, and the emotional intelligence required to turn a click into a loyal customer. In this new era, those who learn to train their tools will be the ones who dominate the feed.
