Evaluating synonyms — turning 300 raw candidates into 30 working keywords.

By the end of Episode 13 the working sheet holds 200–400 raw candidates. Almost all are reasonable on first read — that's why they were harvested. Evaluation is the discipline of cutting reasonable to right. The output is a working list of 25–40 keywords structured by where they belong on the listing.

The four evaluation columns

1. Search volume. Monthly Amazon-native volume. Sourced from the AMALYZE Extension, AMALYZER, Brand Analytics or Sponsored Products impression data. The cutoff varies by category — in mass-market kitchen, 800/month might be the floor; in niche specialty, 80/month is meaningful.
2. Intent match. How well the keyword's implied intent matches what the product actually delivers. A high-volume keyword with mismatched intent kills conversion rate, which kills rank — net negative even though the volume looks good.
3. Category fit. Whether the keyword's dominant Amazon category matches your product's category. A keyword whose top 20 results sit in a different category is one Amazon's algorithm will deprioritise for your listing even if you index for it.
4. Source confidence. A keyword that appears in 3+ of the Episode 03–13 sources is high-confidence. A keyword that only appeared once is single-source signal and gets weighted down.

The de-duplication pass

• Stem-merge. "loaf pan", "loaf pans", "loafpan", "loaf-pan" all collapse to one entry. Sum their volumes for the merged entry's score.
• Word-order merge. "non-stick loaf pan" and "loaf pan non-stick" are the same indexed keyword on Amazon. Collapse to one entry, keep the more natural-sounding form for the visible title slot.
• Synonym-cluster collapse. "loaf pan", "bread pan", "loaf tin" are different words for the same thing. Keep all three in the working list — they index separately — but flag them as a cluster so the writer doesn't waste title characters on the lowest-volume variant.

Bucketing the survivors

The 25–40 survivors get placed into four buckets by where they belong on the listing:

• Title — first 80 characters. 2–4 keywords. The highest-volume, highest-intent, category-fit terms. These earn the most algorithmic weight and the most click-influencing real estate.
• Title — back half. 3–5 keywords. Useful but lower-priority terms.
• Bullets. 8–14 keywords. Including use-case suffixes, audience signals (from Episode 12) and longer descriptive variants that read naturally in benefit copy.
• Backend search terms. 10–15 keywords. Misspellings, regional spellings, low-volume variants, lateral cluster members that didn't earn a visible slot.

The reject list (and what to do with it)

• Brand-prefixed competitor queries — discard permanently.
• Editorial framing words ("best", "review", "top") — discard.
• Mismatched-intent high-volume — discard with a note for the Module 9 lead so PPC doesn't accidentally bid on them either.
• Low-volume single-source candidates — park in a "watch" list for the next refresh cycle. Some of them grow.

The deliverable

One spreadsheet, one tab per ASIN, four bucket columns. Each row: keyword, source(s), volume, intent score, category fit, position assignment. That sheet is the input for Episode 15 (the AMALYZE Extension workflow) and ultimately for Module 8 (writing). No further keyword decisions should need to happen downstream — if a writer is asking "should I use this keyword?", the evaluation pass left something undecided.

How long this takes

For a single ASIN: 2–4 hours of focused work, assuming clean data from the harvest. For a portfolio of 50+ ASINs: a tool-supported workflow becomes essential — manual evaluation at scale erodes consistency and the scoring drifts. Episodes 15 and 16 cover the AMALYZE-side workflow that mechanises the volume-attachment step so the evaluator's time is spent on intent and bucketing instead of data lookup.