Sorting shops into categories with AI
Categorising a catalog by hand does not scale. Six hundred shops imported from a network arrive with no categories at all, and the work is not only putting each shop somewhere — it is deciding what the categories should be in the first place.
This does both, and it never creates a category without you.
Shop catalog → Bulk tools → Categories
Run, review, approve
Run. Every shop is looked at once. If it fits a category you already have, it is put there immediately. If nothing fits, the AI proposes a name — and that proposal is recorded, not created.
Review. The proposals arrive as a list, ranked by how many shops asked for each one:
Health and Wellness
84
Fashion and Clothing
61
Pet Supplies
31
Approve. Tick the ones you want and the categories are created — and every shop that asked for one is put into it straight away. There is no second run.
The first run assigns nothing, and that is correct
On a site with no categories yet there is nothing to sort shops into, so the first run proposes and stops. What you get back is a category set derived from your own catalog rather than a generic list — which is the harder half of the job done.
Approve twenty of them, and the rest of the catalog can be sorted into those twenty.
Approve before you carry on
The button says Suggest categories while you have none, and Assign categories once you do. Both do the same thing; the name changes because the outcome does.
Run it in batches and approve between them. It matters more than it sounds: each shop is looked at independently, with no knowledge of the others, so shops that have not been looked at yet cannot know what earlier ones proposed. Approve first and they choose from real categories instead of proposing their own near-duplicates of the same names.
Two rows that mean the same thing
They will happen — Health & Wellness and Health and Wellness from different shops. Edit the name on either row to match the other and the two merge, taking their shops with them.
The same trick attaches a proposal to a category you already have: type the existing name and no new category is created, the shops just join it.
Dismissing is not the end of it
Dismiss forgets a proposal without creating anything. The shops go back into the pool, so a later run asks about them again — by which time your vocabulary is bigger and they will usually fit something you have approved.
That is the intended treatment for the narrow tail. Proposals like Precious Metals or Eyewear, held by one shop each, are usually better dismissed than created: run again once the broad categories exist and those shops get absorbed into them.
Shops it skips
A shop the AI cannot identify is skipped rather than guessed at. Almost always this means the shop has no domain — there is nothing to identify it by but a name, and a name alone produces a confident, wrong answer.
Fill in Domain under Shop Info, or switch on Web search under Settings → AI, and run again.
Category instructions
Settings → AI → Category instructions is applied to every run. Use it for how broad your categories should be, wording to follow, or category names to avoid.
It works the same way as Shop instructions, and it is worth writing before a large run rather than after.
What it will not propose
Anything describing cashback, coupons, discounts or rewards. Those describe your site, not the merchant — every shop you list offers cashback, so a category called Cashback and Rewards sorts nothing.
Related
Last updated