Amazon Listing Automation
A hybrid pipeline for a Dublin Amazon seller — LLM listing copy plus a custom Photoshop script rendering pixel-perfect product images.
Overview
Built for Jungle Mug in Dublin, an Amazon seller with a catalogue too large to write by hand and too competitive to write badly. Amazon listings are a strange genre — half copywriting, half keyword engineering, half image production, entirely governed by rules about character limits and prohibited phrasing that the marketplace enforces silently.
The pipeline is hybrid on purpose: raw artwork and a target market go in, optimised listing copy comes out of the LLM, and a custom Photoshop script renders the product images pixel-perfect. It produced hundreds of listings and processed thousands of receipts, emails, and attachments on the operations side.
What it does
- Titles, bullets, and description generated per product from its attributes, written to Amazon's field limits rather than trimmed to fit afterwards.
- Backend keyword sets built from search-term research, deduplicated against what the title and bullets already cover — the space is limited, so repeating yourself wastes it.
- Catalogue-scale runs. The whole point is doing hundreds of SKUs in one pass, with consistent voice across the range instead of drift between whoever wrote which listing.
- Image rendering from artwork. A driven Photoshop script takes the raw artwork and composites the product shots to spec — no manual export step between design and listing.
- Compliance guardrails. Character caps and prohibited-claim rules are enforced in code, not left to the model's discretion.
Why it mattered
Listing quality is the difference between a product that converts and one that never gets seen. Doing it well by hand takes a copywriter who also understands Amazon's search ranking — an expensive combination, and a slow one. Automating the first draft moved the human effort from writing to reviewing, which is where the judgement actually adds value.