How AI classification works

What the AI Import does: marketplace category matching, attribute extraction per schema, confidence flagging, and the product statuses it produces.

The AI Import journey

After your products are imported, Start AI Import opens the AI Import journey — three steps: AI Launchpad (choose how many products to process), Processing Job (watch the run), and Review Job (check the results and publish). What the AI does for each product:

• Categories matched by embedding similarity — your product is matched to the target marketplace's own category tree.
• Product attributes extracted per marketplace schema — the fields that category requires (brand, size, material…) are filled from your source data.
• Low-confidence matches flagged for review — nothing uncertain slips through to publishing.

What the statuses mean

• Unprocessed — imported, not yet sent to AI.
• Matched — the AI found a clean category match with nothing missing; one click in review confirms it.
• Needs Review — a required field is missing, or the match confidence was below the bar; you decide.
• Ready — confirmed and complete; will be included in the next publish.
• Published — live in your inventory and sent to the marketplace.
• Failed — the AI couldn't process the row; re-run it from the review queue.

Watching a run

The Processing Job step shows live progress — a percentage, a products counter, and a running log of what the AI is doing ("Mapped 'Blue ceramic mug' to 'Home & Kitchen › Mugs'", "Warning: Missing required 'brand' for SKU: …"). You can leave the page; the run continues in the background. A Stop Import button is available while the run is active — stopped items are marked failed so you can re-run them later.

When AI isn't needed at all

Products pulled from the same marketplace and country you're publishing to already have the right category — they skip AI entirely, cost no credits, and go straight to Ready (or Needs Review if something required is still missing). The Launchpad tells you how many products qualify before you run.

Next: what a run costs, and reviewing the results.