Have you ever known the *feel* of something – a design, a brand's voice, a song – but struggled to explain it or copy it? It happens all the time. It's hard to turn that gut feeling into clear instructions, especially for an AI. But what if you could capture that feeling, that structure, in a way that's easy to understand and perfect for AI? **You can, using JSON.** ![Example JSON Structure](https://assets.shapediver.com/uploads/2019/02/JSONSample.jpg) ## Why JSON Works You probably know JSON (JavaScript Object Notation) from web development. But it's much more than that. Think of it as a **simple way to break down and describe almost anything.** Why is it so useful here? * **Clear Structure:** JSON uses key-value pairs, making it organized and easy for both people and AI to read. * **Flexible:** It works for simple lists or very complex information. You can describe a basic logo or a detailed strategy. * **AI-Friendly:** AI models are great at reading structured data like JSON. They can easily spot patterns and use the JSON as a clear blueprint to follow. ## The Practical Workflow So, how do you actually put this into practice? Here's a general workflow for asking an AI to assess data and generate a reusable JSON profile:

1. Assemble Your Source

Gather the data you want the AI to analyze. This could be text, code, strategic documents, or a collection of images in a specific art style.

2. Choose Your Tool

Select an AI model capable of processing your source material. For images or videos, you'll need an AI with vision capabilities. For analysis tasks like the website example later, I recommend trying [Perplexity](https://perplexity.ai/).

3. Instruct LLM #1 (Analyzer)

Use a detailed prompt (like the "Content DNA" prompt below) to tell the AI to analyze the source material and extract its key characteristics into a structured JSON format. The goal is a JSON profile capturing the *essence* of the source.

4. Instruct LLM #2 (Recreator)

Take the generated JSON profile. Provide it to a second LLM (like GPT-4o). Instruct it to create *new* content (or modify existing content) that adheres to the style and structure defined in the JSON profile.

5. Review and Refine

Examine the AI-generated output. How well did it match the JSON profile? You might refine the JSON profile or the prompts for either LLM for better results.

This process transforms subjective qualities into a concrete format that AI can reliably work with. Let's look at how you can use JSON to structure *anything*. ## Structure anything with JSON The real power emerges when you apply this structured approach to things not typically considered 'data'. You can create a detailed JSON profile – like a 'DNA' – for nearly anything: * **Visuals:** Logos, web designs, image styles. * **Text:** Writing styles, email structures, brand voice. * **Code:** App structures, common patterns. * **Concepts:** Business strategies, marketing plans. * **Media:** Music elements, video styles, ad copy. By breaking things down into JSON, you give an AI a detailed map. It can then understand, analyze, adapt, or even recreate the original thing, or make something new in the exact same style. ## Real-World Example: Analyzing a Website with Perplexity This process works well for various types of analysis. For instance, I used [Perplexity](https://perplexity.ai/) to analyze my own website, [sourcetms.com](https://sourcetms.com/). I prompted it to act as an expert analyst, examine the site from multiple perspectives (code, design, brand, text, etc.), and generate a JSON profile that I could potentially use as a template for another LLM to recreate a similar site. Here's a glimpse of the process: Screenshot of Perplexity analyzing sourcetms.com and generating a JSON profile

This generated JSON, shown below, provides a structured breakdown that's much more useful for subsequent AI tasks than a simple text description. ## Example: Website Analysis Prompt & Profile Here's a simplified version of the prompt used in the example above and the resulting JSON structure. You can adapt this for analyzing other websites or different types of content. *(Remember, AI-generated profiles like the one shown are illustrative and often need refinement; the key is customizing this structure to your needs.)*

  
> **Your Role:** Act as an expert analyst.
> 
> **Input Source Material:** https://sourcetms.com/
> 
> **Task:** Assess this website and create a JSON profile from it. Look at the code, design, brand, style, text, font, imaging, and other perspectives I did not consider. Then, create a JSON profile I can use to provide to another LLM to recreate a website based on the JSON profile of it as a template. Use the structure defined below.
> 
> **Required JSON Output Structure (Website Profile):**
> ```json
> {
>   "website_profile": {
>     "technical_analysis": {
>       "code": {
>         "frontend_stack": ["HTML5", "CSS3", "JavaScript"],
>         "responsive_breakpoints": ["Mobile: <768px", "Tablet: 768-1024px", "Desktop: >1024px"],
>         "performance": {
>           "load_time": "~1.2s (estimated)",
>           "optimizations": ["Minified assets", "Lazy loading", "Cache headers"]
>         },
>         "seo": {
>           "meta_tags": ["standard Dublin Core", "OpenGraph protocols"],
>           "schema_markup": ["Person", "Organization"]
>         }
>       }
>     },
>     "design_analysis": {
>       "layout": {
>         "structure": "Single-page application (SPA)",
>         "grid_system": "CSS Grid/Flexbox",
>         "whitespace_ratio": "40% content / 60% negative space"
>       },
>       "typography": {
>         "primary_font": "Sans-serif (System UI stack)",
>         "font_scale": {
>           "h1": "2.5rem",
>           "body": "1.1rem",
>           "secondary": "0.9rem"
>         }
>       },
>       "color_palette": {
>         "primary": ["#2c3e50", "#ffffff"],
>         "accent": ["#3498db", "#2980b9"],
>         "contrast_ratio": "4.5:1 (WCAG AA compliant)"
>       }
>     },
>     "content_strategy": {
>       "messaging": {
>         "value_proposition": "Executive leadership in fintech",
>         "tone": ["Authoritative", "Concise", "Achievement-focused"]
>       },
>       "key_content_blocks": [
>         {
>           "type": "professional_summary",
>           "elements": ["Co-founder status", "Sector expertise", "Leadership experience"]
>         }
>       ]
>     },
>     "brand_attributes": {
>       "visual_identity": {
>         "logo_type": "Wordmark (text-only)",
>         "imagery_style": ["Corporate headshots", "Abstract tech patterns"]
>       },
>       "differentiators": ["Finance-tech crossover", "Scalable solutions focus"]
>     },
>     "interaction_patterns": {
>       "navigation": {
>         "menu_type": "Anchor-linked SPA",
>         "scroll_behavior": "Smooth scrolling"
>       },
>       "cta_strategy": {
>         "primary_cta": "Contact overlay trigger",
>         "secondary_cta": "Scroll-based engagement prompts"
>       }
>     },
>     "compliance": {
>       "accessibility": ["Basic ARIA labels", "Alt text on images"],
>       "privacy": ["GDPR-compliant analytics setup"]
>     }
>   }
> }
> ```

## More Than Just Copying Using JSON isn't just about copying things. It's about **understanding** them better. When you have that structured 'DNA,' you can ask an AI to: * **Analyze:** Find strengths or weaknesses. * **Adapt:** Change the format (like turning this post into a tweet thread). * **Compare:** See how different styles or strategies are similar or different. Moving beyond simple text prompts to structured JSON like this Content DNA profile grants you significantly more control and unlocks deeper insights when collaborating with AI. So, the next time you need to distill the essence of a complex idea or style, remember JSON.

Thomas Schlossmacher

Occasional insights on AI & building

Writing