Cosmy — New Sellers Research

How to use this tool →

Past Runs

Read directly from disk (run results, logs, and the seller database), not from live tracking in this page or the server process — so this list survives a page refresh, a closed tab, or the backend restarting. Runs from before this feature existed won't appear here, but their data is still intact; ask if you need to look one up.

Step 1: Keyword Research

This tool queries Jungle Scout to find real, volume-backed Amazon search keywords for your niche, rather than guessing. We filter by monthly search volume (both exact and broad match), word count, and organic product count to surface promising phrases.

Comma-separated. Each one costs its own Jungle Scout request (see Max Requests below) — this isn't a single batched call.
Reverse lookup: finds keywords these known products already rank for. Batched automatically, 10 ASINs per request.
Tip: a good seed keyword is specific, not generic. Aim for ~200-2,000 competing listings and 3+ words — enough depth that Amazon's pages 3-8 actually exist and surface newly-listed sellers, not so few that there's nothing to find and not so many that established incumbents still dominate that deep. Bad: "baby formula" (100K+/mo searches, tens of thousands of listings — pages 3-8 are still Enfamil/Similac). Good: "goat milk formula for infants" (roughly 1-3K/mo searches, a few hundred listings, 5 words — real depth without the saturation). More keywords helps too, but a handful of well-targeted ones tends to beat a long list of generic ones. See "Advanced filters" below to check these numbers directly before running.
Required. Hard cap on Jungle Scout API calls for this run — shared quota with other production usage, so the run refuses to start if it would need more than this. Suggested: the number of seed keywords above (or ASIN batches of 10), plus a little headroom.
Advanced filters (optional — leave blank for no filtering)
Jungle Scout's own marketplace code (e.g. us, uk, de) — not the same convention as the Amazon domain suffix used below in Step 2.
Comma-separated, must match Jungle Scout's own category names for this marketplace exactly (e.g. "Baby", "Home & Kitchen" for US). Leave blank to search all categories.
On by default (Jungle Scout's own default). Can surface many more keyword results per ASIN.
How many searches match this exact phrase per month. Confirmed in live testing: for keyword-seeded (not ASIN-seeded) queries, this value can come back identical across every result — treat it as a coarse filter, not a reliable ranking signal, for that case.
Includes plurals/misspellings/synonyms/related terms. The more reliable signal for keyword-seeded queries — this is why it's the default sort below. Suggested floor: ~500 to skip near-dead phrases.
3+ words tends to surface niche/new-seller phrasing rather than head terms owned by incumbents.
Recommended band ~200-2000: enough listings for page 3-8 results to exist, not so many that established sellers still dominate that deep.
Broad is the default: exact-match volume has been observed to tie identically across all results for a keyword-seeded query, which makes it an unreliable sort key there.
How many top-ranked keywords to show and be able to save. Suggested: 10-20.

Step 2: Run Scrape

Run the actual Amazon seller-discovery sweep for your niche and produce a downloadable results spreadsheet. This costs real money against an Apify budget, so set a hard spending cap.

Must already exist in config/niches.json (e.g. saved from Step 1 above) unless you also fill in Keywords below to define a new one on the fly.
Required. Hard spend cap in USD — the run stops and leaves unfinished sellers for next time rather than exceed this. Suggested: start small (e.g. $2-5) for a first run on a new niche, since actual cost-per-keyword isn't known until you've run it once.
Advanced options (optional — every field below has a working default)
Override the niche's configured keyword list for just this run, without editing config/niches.json.
Cap how many of the niche's keywords to use this run — handy for a cheap test before a full sweep.
Amazon domain suffix (e.g. com, co.uk, de) — a different convention from Jungle Scout's marketplace code in Step 1. Only "com" (amazon.com) is validated end-to-end so far.
Amazon result pages to read, sorted Newest Arrivals — pages 1-2 are skipped by design, they stay dominated by incumbent sellers.
Fallback range if a keyword returns zero results on its primary page range.
Overall cap on sellers scraped per keyword — a cost lever, not a relevance filter.
Per-keyword cap passed to the discovery actor.
How many product listings the discovery actor scans per keyword while looking for sellers.
amazon.com rarely publishes seller email/phone directly (that's an EU-marketplace disclosure requirement, not a US one) — expect empty fields even with this on.
Cheap pre-filter: keep sellers with fewer than this many ratings in the last 12 months. Not a reliable "is this seller new" signal by itself — just narrows the pool before the expensive verification step.
A seller counts as "new" if their earliest-ever review is within this many months. This is your actual definition of "new".
Flags (doesn't exclude) a seller as "possibly reactivated" if there's a gap this long between two of their reviews — a note for your own judgment, not an automatic rejection.
How deep to page into a candidate's review history during verification — the main cost lever on the expensive step (long-tenured false positives are the pricey ones to rule out).
Leave off for normal runs — the dedup ledger is what makes repeat runs on the same niche cheap over time.