🗞️ AI Mimicry: The JSON Method💥🚀

If you’re new here, welcome! If you’ve been reading for a while, thank you for sticking around as we navigate this wild AI shaped shift happening across eCommerce. Each week I share what I’m experimenting with, what’s actually moving the needle, and the trends that deserve your attention before they hit your competitors’ playbooks.

A quick heads up. I’ve organized all previous editions into one searchable hub. If you want the full journey, it’s all here.

The Code Behind the Canvas: Why JSON is the Secret to AI Consistency

Most people think "style" is about adjectives. They try to replicate a look by typing “make it moody, cinematic, dark, blue lighting.” The result? Usually a caricature. It captures the vibe but misses the soul.

I recently discovered a better way, and it changes everything about how we use AI for creative work. It’s not about finding better words; it’s about changing the language entirely. I started feeding AI images and asking it to reverse-engineer them into JSON code.

What in the world is JSON?

JSON (JavaScript Object Notation) is a code format that acts as a blueprint, locking in specific variables like lighting and texture so the AI doesn't have to guess. This is what JSON looks like.

Wait, what? Code to describe an image?

That’s right. It sounds counterintuitive: using code to describe beauty. But structured data is the native language of AI. It cuts through the ambiguity of English.

The "Forensic" Workflow

I started with a static image. A high-gloss cosmetic ad featuring jellyfish by Japanese beauty brand, Kate. Instead of describing it, I asked the AI to analyze it and output the style as a JSON object. This gave me a digital fingerprint of that aesthetic: specific parameters for lighting, color grading, and composition, stripped of the subject matter.

When I fed that JSON back into the LLM, the results were striking. I could produce dozens of images that looked exactly like they came from the same campaign. Just look at this…!

It worked so well that I built a custom Gemini Gem just to handle this "Style-to-JSON" workflow.

Why JSON Mimicry Works (The "Slot Machine" Theory)

Why does this work better than normal prompting?

When you write a paragraph of text, you are forcing the AI to guess the relationships between words. "Dark" might modify "lighting," or it might modify "mood." The AI has to interpret your grammar.

JSON removes the guesswork. It forces the AI to decouple the Subject from the Style. Think of it like a slot machine.

• Text Prompt: You pull the handle, and all the wheels spin together. If you change the subject, the lighting often changes with it accidentally.

• JSON Prompt: The "Lighting" and "Camera" slots are locked. You only spin the "Subject" wheel.

By speaking to the model in data rather than prose, we bypass the "hallucination" of style. We aren't asking the AI to imagine a look; we are giving it the blueprints to build it.

From Images to Video

I didn't stop at static images. Once I realized JSON acts as a "rendering engine" rather than a chatbot, I applied it to video.

Steve Simonson recently shared a JSON for an Apple video. I took that JSON code, plugged it in into VEO3, and the AI recreated that distinct, high-fidelity Apple "feel" instantly. No long paragraphs of description, just structured data.

Just look at this beauty…

Have you tried using JSON for AI image style mimicry? How did the results turn out for you?

Breaking the Glass: The "Bulk File Automator"

The biggest frustration with AI is amnesia. You build a perfect workflow in a chat, but when you close the window, it’s gone. If you want to analyze an Amazon Ads bulk file, you usually have to paste messy snippets of data and explain the column headers every single time.

There is a better way: Claude Code. You can access in the top left of your Claude app window in your browser (no need to go to the desktop version.)

This tool allows the AI to break out of the browser and access a specific folder on your actual desktop. It transforms Claude from a chatbot into an operator that can read, edit, and generate files locally on your laptop.

The Workflow

I created a folder on my desktop called Bulk File Automator. I activated Claude Code in this directory and gave it one instruction:

Create or update my CLAUDE.MD file with all the knowledge necessary to create and edit Amazon bulk files, documented below. The user will then use this project to work on various bulk files in the future.

In the directory is a TEMPLATE Amazon Bulk File with the right header rows for each ad type, and a SAMPLE of a file with data.

The documentation from Amazon is here: https://advertising.amazon.com/API/docs/en-us/bulksheets/2-0/overview-about-bulksheets

Now, that folder is intelligent!

It "knows" the Amazon schema permanently. I can drop my actual bulk file (CSV/XLSX) into that folder and execute three powerful use cases:

1. Chat with your Bulk File You don't need to explain the columns—it already knows them.

"Are there any campaigns with an ACoS > 50% that are set to Fixed Bids?"

2. Extract & Create Subsets

"Extract the top 10 performing keywords and save them as a new file called top_performers.csv."

3. Create New Files on the Fly You can ask it to generate complex upload files from scratch.

"Create a new bulk file for a campaign named 'Jellyfish_Launch'. Set the daily budget to $50, default bid to $1.50, change strategy to 'Down Only', and include these 5 keywords."

Because you "trained" the folder with the documentation URL, it generates a perfectly formatted file ready for upload—no manual formatting required.

You know that feeling when you upload a bulk file and it uploads error free :)

Amazon’s AI shopping assistant has launched Deep Research Mode, a capability that moves Rufus from simple Q&A to a comprehensive research agent. By running up to 30 searches and synthesizing data from external editorial sources, Rufus now builds personalized, "Wirecutter-style" buying guides in seconds.

Why this matters for brands: Rufus is no longer selecting products based solely on sales ranking. It now prioritizes role, context, and use-case fit. To win in this new environment, brands must shift focus toward "Reasoning Path Optimization" (RPO) and external authority building to ensure they aren't just ranked high, but are understood as the right solution for the shopper’s specific goal.

Read the full post by Andrew Bell on LinkedIn.

We hope you liked this edition of the AI for E-Commerce Newsletter! Hit reply and let us know what you think! Thank you for being a subscriber! Know anyone who might be interested to receive this newsletter? Share it with them and they will thank you for it! 😃 Ritu