AI assistance, human judgment.
Mercheagle covers 16,620 shirt designs. Writing accurate, useful descriptions for each one at that scale requires AI assistance, there's no honest way around it. We use AI as a tool, not a replacement for editorial judgment.
What AI does
For individual design pages, we use Google's Gemini model to generate a visual description of each design based on the actual product image, an audience framing, and SEO metadata (title, meta description, alt text, bullet points).
The model is given the product image and a structured prompt that requires it to describe what is literally visible, naming specific entities like breeds, species, landmarks, instruments, and sports, and frame the design for a real buyer.
What AI does not do
AI does not decide which designs get featured on pillar or hub pages, write pillar or hub page content (those are written by human editors), assign categories (taxonomy decisions are made by humans), or override our QA checks.
Our QA process
Every AI-generated design description passes through an automated gate that checks title tag length (48–65 characters), meta description length (130–200 characters), body word count (120–400 words), alt text length (80–220 characters), absence of forbidden marketing phrases, and a confidence score threshold.
Designs that fail QA are not published until the issue is resolved.
Human review
All designs that received low confidence scores from the AI model were manually reviewed and rewritten by a human editor who viewed each design image directly. Pillar and hub pages, the site's main editorial content, are written by human editors, not AI.
Transparency
We believe in being direct about this. AI-assisted content at scale can be done well or done badly. We've tried to do it well: structured prompts, real image inputs, strict QA, human oversight on anything flagged as uncertain, and full human authorship on the pages that carry the most editorial weight.
Questions about our process: hello@mercheagle.com
Last updated: April 2026