The skill
Install once. Then attach your product CSV and ask Claude to build the feed. It maps your columns to OpenAI's schema and rewrites titles and descriptions to spec.
What's inside
Three files, one job: a clean feed
Built for performance marketers, not engineers.
Install guide
Five steps with screenshots to add the skill in Claude, plus a no-install option for anyone who would rather just paste a prompt.
The prompt
The same conversion and QA logic as a plain prompt. Open any Claude chat, attach your CSV, paste, done. Nothing to install.
How it works
Three steps, about an evening of work instead of a developer ticket.
Step 1Export
Export your catalog as a CSV from Shopify, Google Merchant Center, or your Meta catalog. The one you already use for Shopping ads works.
Step 2Convert
Hand the CSV to Claude with the skill on. It maps your fields to OpenAI's schema, rewrites titles and descriptions to spec, and normalizes the enums.
Step 3Validate
Get a clean feed plus a QA table of every row that would fail. Fix the blockers, validate a 100-item sample, then upload the full catalog.
You bring the export. The skill does the conversion and the QA. There is no live connection to your store, so there is nothing to authorize and nothing to break.
How to build a ChatGPT Ads product feed
In ChatGPT Ads, the feed is the creative. The ad title and description are pulled straight from your product catalog, so a clean, factual, well-mapped feed is what decides whether you show up at all.
What goes into a ChatGPT Ads feed
Building one by hand means mapping your columns to OpenAI's product feed schema, rewriting every title and description to spec, normalizing the availability enums, and catching the rows that would be rejected before you upload. For a large catalog, that is hours of work. For the full breakdown of the field rules and the spec, read the guide: ChatGPT Ads Now Eats a 2-Million-Item Feed.
How the free skill builds it for you
Export what you already have from Shopify, Google Merchant Center, or your Meta catalog as a CSV, hand it to Claude, and get back a spec-compliant feed plus a QA report that flags every blocker by row: missing required field, bad enum, sale price above price, non-HTTPS URL, title over the limit. You fix the source data, validate a small sample, and upload.
Why build it now
ChatGPT Ads is a new channel and most catalogs are still messy. Right now the auction is thin, which means cheaper clicks for whoever shows up clean and early. That window closes as more brands move in. The work that decides whether you compete is not bidding, it is whether your feed is live and correct before your competitor's is.