Everyone negates wasted search terms. Most stop one signal too early The standar
By Twillms · October 8, 2026 · Curated by George's Blog
Everyone negates wasted search terms. Most stop one signal too early
The standard move is a 60-day bulk file: anything with 20+ clicks and 0 orders gets negated. It works, and it misses the hardest money.
Two problems with one fixed number.
First, 20 clicks is arbitrary. At 2% CVR you need ~100 clicks before "0 orders" means anything. At 10%, ~25. One threshold can't fit both products.
Second, it never sees the long tail. Thousands of auto / broad / phrase terms with 2 to 6 clicks each, zero orders, every one under the bar. Individually invisible. Together, most of the leftover waste.
How we run it instead, multi-signal.
Programmatic: dynamic thresholds from each product's CVR, multi-lookback to tell one bad week from chronically bad, context from SQP, rank and all-time ACOS, and parent / child ASIN performance.
AI on top: semantic relevance to the product and to the originating target.
And the part nobody else does: we score every search term against the ASIN with a confidence model, and negate on relevance even when the data is thin. We can finally do this because AMS v2 gives us the advertised ASIN at search-term level.
Across one account: 1,094 terms flagged, $652 of pure waste, 93% of it under $1 per term. None of it would ever trip a 20-click rule.
Negation was never a 20-click rule. It's a judgment, made on many signals.
Full breakdown in the carousel.
What's your approach on negating with AI?