AI Prompt Design
Patagonia Social Content
I approach LLM prompt design as translation work—converting brand voice principles into instructions an AI can execute. In this self-guided experiment, I chose the outdoor apparel company Patagonia and explored how to maintain its distinctive voice across product, values, and educational content through tailored ChatGPT prompting.
The Brand Context
On social media, Patagonia’s voice balances approachability with authoritativeness. They speak boldly and directly, bringing an action-oriented energy that’s mainly pragmatic and slightly motivational. When discussing products, the language highlights materials and construction quality, utilizing technical lexicon that showcases their commitment to craft—often connecting items to outdoor experiences or environmental contexts. They address environmental issues matter-of-factly, avoiding preachiness while demonstrating firm conviction in their values.
Brand Voice Reference
Brand Voice Reference
Brand Voice Reference
Brand Voice Reference
My Content Challenge
Patagonia needs to generate Instagram captions at scale for a product launch, a behind-the-scenes spotlight, and an educational post. The challenge is maintaining their distinctively bold, action-oriented, and matter-of-fact voice while working with AI tools that tend toward clichéd, repetitive content outputs.
I developed three prompt templates for different content types: a product spotlight, a behind-the-scenes story, and an educational post. For each, I iterated on prompt structure to capture Patagonia's voice—moving from generic brand guidelines to specific constraints, example language, and structural guidance.
Example 1: Product Spotlight
Iteration 1: Basic Prompt
The initial prompt produced some decent content, particularly the Rugged + Sincere option. I noted some core Patagonia voice elements: the matter-of-factness and the action-oriented adventurous spirit. Still, these outputs were too generic and inauthentically upbeat, so I refined my approach.
Iteration 2: Adding Brand Context
Now we were moving in the right direction. Notably, there was more narrative imagery in this output. Some of Patagonia’s best captions create a vivid sense of place to hook the reader.
When outlining the brand context, I spelled out both which parts of Patagonia’s voice I wanted to see reflected and what language elements I didn’t want to see. This “include-exclude” approach narrowed the model’s focus. I then knew that to get even better outputs, I should build upon this approach further.
Iteration 3: Specific Constraints
I became even more specific in my directions, identifying an existing product that would be the caption’s subject. I explicitly outlined my desired output’s structural beats, giving me a clear heuristic to evaluate the output and determine how well the model completed the task.
I included a link to that product’s page so the model could have it as an appropriate reference for analysis (later on, I remixed this aspect of my approach).
With the constraints, link to existing Patagonia language, and structural guidance, the model produced authentically Patagonia-sounding content that I could refine for a final version in several minutes.
I adjusted the language in places for flow and syntax. I decided against the statistical jargon after seeing that it disrupted the caption’s pace.
Final Text:
Mornings on the trail. Coffee shops downtown. Chilly campsites. The Better Sweater Fleece Jacket is built for real use — made from 100% recycled polyester sweater-knit fleece that balances warmth with mobility.
It resists abrasion, holds it shape, and earns its place in your routine.
Check out the Better Sweater at the link in our bio.
Example 2: Behind-the-Scenes Story
The designed prompt delivered strong content outputs—leading with the workers and avoiding virtue signaling. I provided another prompt with all my refinement directions, ending with a strong final version.
For Example 1, with the rounds of prompt iteration, arriving at my final content took around an hour. Starting with the designed prompt, getting to a similar level of quality took only 15-20 minutes.
After learning from several rounds of iteration in the first example, I knew to design prompts with specific, structural constraints from the start.
In the prompt, the first line outlined the situation. Then, I told the model to keep what it’s retained about Patagonia’s brand voice in mind, before providing explicit guidance on what to do and what to avoid. I provided three or four directions structural guidance, laying out my desired caption’s informational arc.
Including an entire URL for the messaging examples had mixed results (the model didn’t always properly fetch the link), so I pulled existing language from a Patagonia webpage to show the model how the company has previously spoken on the subject. I opted for longer illustrative sentences since they were information rich, but shorter phrases were also effective.
Final Text:
Inside a Fair Trade Certified™ factory in Vietnam, hundreds of skilled workers cut, sew, and assemble the gear built to last through years of wear. These workers earn additional premiums and decide together how to use them—from better housing to education funds. The program helps ensure safety, voice, and respect are part of the job, not perks. It’s one way to prove better business starts with better conditions.
Hear from Maria, a Patagonia production manager, on how Fair Trade is shaping our supply chain—link in bio.
Example 3: Educational Post
I then adjusted the language for flow and removed overexplanatory details, letting Patagonia’s characteristic directness, pragmatism, and commitment to their values shine through.
Example 3 shares the approach of Example 2: designing a specific prompt laying out things to include, content structure, and illustrative messaging examples:
Final Text:
Each wash of synthetic clothing releases thousands of microplastic fibers into waterways and oceans. They don’t break down—they build up, moving through ecosystems and into the food chain.
Since learning of our role in microfiber pollution in 2015, Patagonia has worked with scientists and engineers to reduce shedding, improve filtration, and design longer-lasting materials.
Hear from Carlos, a Patagonia materials engineer, about how we’re partnering across the industry to cut microfiber pollution and what you can do to help—link in bio.
Key Learnings
Structure guides content: Specifying caption structure consistently created better, more easily evaluative outputs
Example language is critical: Providing the AI with actual brand phrases helped it understand tone faster than adjectival descriptions
Constraints clarify voice: Listing out what not to say draws clear lines for the AI
Iteration unlocks strategy: Rounds of experimentation—particularly adding layers of complexity over time—get prompts to a format that give the AI what it needs