Panel data tells you who won the category last month. It cannot tell you who is winning it this morning — which shops are actually shifting volume right now, whether the leaders are official stores or resellers, and where they are shipping from.
That is a live question, and it takes one job to answer.
What you get back
Submit a set of marketplace search URLs, sorted by sales, covering the keywords that define your category. What comes back is every listing the search surfaced, with its price, its monthly sold count, and the shop behind it.
Aggregate by shop and you have the category's current sellers ranked by observed volume — with listing counts, average realised price and location.
Three parsing details that decide whether your numbers are real
Units sold is the numeric field, not the text one. The text version is a bucket — "10k+" — which makes a listing selling 10,001 units look identical to one selling 90,000. Read the numeric monthly_sold_count or your ranking is noise.
Price is scaled by 100,000. Divide, or every figure is five orders of magnitude wrong.
Deduplicate by item id before summing anything. The same listing can appear across multiple search frames. Without dedupe your category total is inflated by an amount you cannot estimate.
Get any of those wrong and the output still looks plausible, which is what makes them dangerous.
Official store versus reseller
The split is usually visible in the shop name, and it is the most actionable cut in the whole analysis. A category where official brand stores hold most of the volume behaves completely differently from one where resellers do — on price discipline, on assortment, and on how much control any brand actually has over its own listings.
If resellers dominate your category and you had assumed otherwise, that is the finding, and it usually predates the pricing problem somebody is currently trying to solve.
The honest caveat
Keyword search reaches what the search results surfaced. That is a wide and useful view, but it is not complete coverage, and you should not present it as market share.
Say "share of observed volume across these keywords" and the number is defensible. Say "market share" and it is not — you would be dividing by a denominator you did not measure. If you need true share, that is the panel's job, computed across full-category coverage.
Report how many submitted URLs returned data and how many failed, every time. A total with three missing frames is a different number from a complete one, and only you know which you have.
Cost and gating
Six credits per URL — a fraction of a cent — up to 50 URLs per request. A thorough sweep across twenty keyword variants costs well under a dollar.
This endpoint is approval-gated and processes between 08:00 and 18:00 GMT+8. It is an analysis tool, not something to hang a nightly job on. If you need recurring unattended collection, product-page fetches have no such restriction.
Recipe 2.2 in the agent skill carries the prompt with the parsing rules built in.