Updated 2026-08-02

A complete Amazon niche research method: from one seed word to a proof-backed decision

The method in one paragraph: start from a single seed word, export its full Cerebro keyword universe, collapse the thousands of phrasings into demand families, read the families through three lenses (dominant demands, provenance, demand-vs-titling outliers), then leave the spreadsheet: open page 1 for each candidate, segment the products actually ranking, and promote the promising segment into a tracked concept with explicit proof status. Numbers propose; the page decides.

Step 0 — Choose a seed word, not a product

The seed is a word ("holder", "organizer", "stand") — deliberately not a product idea. Its job is discovery: one export surfaces 5,000–15,000 phrasings, including demands you could not have imagined. Starting from a product idea only confirms what you already believe; starting from a word shows you the territory.

Step 1 — Import everything, clean nothing by hand

Export the seed's Cerebro universe and ingest it whole — with its floors, text-numbers and invisible characters handled by rules, not by eyeballing. Unknown stays unknown; a floor stays a floor.

Step 2 — Collapse into families

Deterministic rules — normalization, word-order insensitivity, plurals, size classes, under-merge bias — turn 11,547 keywords into 8,872 families. Every metric from here on is a family metric, with per-column aggregation (volume sums; Title Density takes the max; cost of entry takes the min).

Step 3 — Read through three lenses

1. Dominant demands — the top families by combined volume: what this territory is actually about.
2. Provenance — Smart Complete first, organic-only under suspicion: a third of a real corpus enters through the noisy door.
3. Demand × Title Density — the outliers where real volume meets near-zero titling: the only rows worth manual attention.

The output of this stage is not a ranking — it is a short list of readable, refutable claims: "family F: combined demand X, max TD Y, provenance mostly SC, unverified on the page."

Step 4 — The page is the proof

For each candidate, open the actual search results. Segment what ranks: the same page usually serves several distinct concepts at different prices. Ask the only questions that matter — is the measured demand served? by whom, at what price, with what reviews? — and let the page overturn the spreadsheet freely. A claim that survives page contact is evidence; a claim that never met the page is a guess.

Step 5 — Track concepts, not keywords

Promote the surviving segment into a concept file: its demand map (which keywords are core vs adjacent to the concept, not to the seed), its gaps, its price structure, its proof status — and keep it alive over time, because pages change. Products disappear from page 1 in weeks; a decision made on a stale snapshot is a guess with a timestamp.

The principles under the method

• Never a composite score. One axis, one unit, one readable claim — anything you cannot argue with, you cannot trust.
• Under-merge, under-claim. Doubt keeps things separate and keeps claims unverified.
• Unknown ≠ zero. Missing data never wins a comparison.
• The page decides. Every number is a hypothesis until page 1 confirms it.

The first three steps run in seconds in the free Cerebro Export Analyzer — entirely in your browser. Steps 4 and 5 are where the desktop app lives; start with You ran Cerebro. Now what?