Updated 2026-08-02

You ran Cerebro. Now what?

After a Cerebro export, do these five things in order: (1) collapse the thousands of phrasings into demand families, (2) read where each keyword came from (its provenance), (3) shortlist families where demand is high and Title Density is low, (4) look at the actual page 1 for each shortlisted keyword, and (5) decide on the segment level — never on a single keyword. Every tutorial stops at "export your keywords"; this guide is about what happens next.

Why is the raw export unreadable?

A real one-word seed export we work with contains 11,547 keywords. Sorted by volume, the top rows are generic monsters ("electronics", 878k volume) and the same demand counted many times: "car phone holder" appears in that corpus as 29 different phrasings — plurals, word-order variants, sizes — which together add up to 544,816 monthly searches. Read row by row, you will overweight whatever happens to be spelled consistently and miss demand that is fragmented across spellings.

Step 1 — Collapse phrasings into families

Group keywords that are the same demand written differently: normalize case and punctuation, ignore word order, merge plurals, fold sizes and dimensions into classes. On the 11,547-keyword corpus above, this collapse leaves 8,872 families — 23% fewer things to read; on a second corpus it removed 37% of the rows. The unit of work stops being the keyword and becomes the family. (Full details: how families work.)

Step 2 — Read the provenance columns

Three columns almost nobody reads — Smart Complete, Amazon Recommended, Organic — tell you how each keyword entered your export, and they are a free relevance filter. In our reference corpus, 34% of all keywords surfaced only through organic ranking data, and those are disproportionately noise from unrelated products. Triage order, not deletion: read Smart Complete first, treat organic-only with suspicion. (Full details: the provenance prior.)

Step 3 — Find demand-vs-titling outliers

Cross two numbers at the family level: combined search volume (demand exists) and the maximum Title Density across the family (how many page-1 titles contain the phrase). A family with real demand and a Title Density near zero is either an opportunity or an irrelevant phrasing — both are worth 60 seconds of your attention, and nothing else in the export is. In our corpus, "command hooks" carries 218,685 combined volume with a Title Density of 0. (Why max, and the trap to avoid: Title Density explained.)

Step 4 — Verify on the actual page

No spreadsheet number survives contact with page 1. For each shortlisted keyword, open the search results and ask: what products actually rank? Is the demand you measured served by one product type or five different ones? A "virgin territory" by the numbers is often a page full of products titled differently. The page is the proof; the export is only the map.

Step 5 — Decide at the segment level

A keyword is not a product opportunity; a segment of page 1 is. The decision unit is the concept — the group of products serving the same demand — and it deserves its own file: demand map, gaps, price range, proof status. That is the part that outgrows any spreadsheet, and it is what the full 10xH10 method is built around.