The cliff in your search-term trend chart is probably Amazon's cap, not the mark
By Youval Peltier · October 10, 2026 · Curated by George's Blog
The cliff in your search-term trend chart is probably Amazon's cap, not the market.
Search Query Performance gives each brand its top 1,000 search terms every week. Volume and share for those. Everything else is a rank with no number attached.
We built a model to turn rank into estimated volume so brands could see trends across their whole term universe. In aggregate, it was fine. Then someone looked at a single term over time and saw a cliff. Volume falling off a table one week, back the next.
The market hadn't moved. The term had dropped out of the brand's top 1,000, our model had filled the gap, and the fill created a variance that looked like an anomaly. The chart was lying in exactly the view users cared about most.
Here's what fixed it, at the level I can share:
1. Treat data consistency as an attribute of the term. A term with steady history is trustworthy. One that drifts in and out of the cap isn't, and the chart should say so.
2. Use the brand's own movement as the control. When a term goes dark, the most reliable estimate of what happened is how the brand's other terms moved, not a universal curve.
3. Build the term-level view before re-surfacing category volume, so "category up, your sales down" arrives with an explanation attached.
The error rate dropped a lot. More importantly, analysts stopped hand-checking spikes before every client call.
The lesson for anyone building on Amazon's data: the platform's reporting limits leak into your charts as false signals. Knowing where the cap is matters more than the cleverness of the model that fills it.