If you sell on Amazon, you already have access to one of the richest datasets Amazon gives any seller: the Search Query Performance report inside Brand Analytics. Most sellers never open it past the first export. That is not a knock on them the report is genuinely hard to use, and the reasons it is hard to use are also the reasons it is worth the effort.
This post covers how Smacient built a Claude skill to work the report properly, what it looks for, and what it tells you once it has looked.
Table of Contents
The Report Nobody Opens
The Search Query Performance export gives sellers the top 1,000 search terms for their category, refreshed monthly, ranked by how well the brand converts on each one, not by search volume.
That single design choice is why the report frustrates people. Open two consecutive months and 50 to 80 per cent of the rows will be different terms. Most sellers read that as noisy data and stop trusting the report. The churn actually comes from the ranking logic: a term drops off the list the moment a brand stops converting well on it, whether or not the term itself is still being searched. That makes it structural, not random. Once that’s understood, the report becomes something to build a monthly habit around instead of something to glance at and abandon.
Two Things to Check Before You Trust Your Export
Before analyzing anything, check how the data was pulled. Two quirks in the raw export catch almost everyone.
Seller Central hides 12 of 33 columns by default. If an export looks thinner than expected, that is why, not a smaller report than expected. Go into column settings and expand it before pulling anything. This tripped Smacient’s own skill during testing until the team caught it, so don’t assume the default export is the full report.
The column labelled “Clicks: Click Rate %” is clicks divided by search volume, not by impressions, so it isn’t a true click-through rate. Read it as CTR, and click performance will be misjudged by roughly twentyfold.
Why Smacient Built a New Skill Instead of Using an Existing One
Smacient didn’t start from scratch. A third-party SQP skill had been circulating, and the team evaluated it before deciding to rebuild rather than adopt it.
It fell short in four ways. It had no time-series capability, so it could only describe one export at a time, not a trend. Its thresholds were hardcoded in USD, which breaks the moment the brand being analyzed sells on Amazon India. It only produced a brand-level view, with no way to see which products were driving or dragging the number. And its thresholds were presented as fact, with no reasoning behind where they came from, which meant there was no way to know when they didn’t apply to a given brand’s situation.
None of these is small gaps. They are the difference between a tool that produces a plausible-looking chart and one that can actually be acted on.
Get the Skill
The skill is open source and available now. It lives at smacient/marketing-skills/skills/amazon-sqp-analysis on GitHub, alongside the rest of Smacient’s Claude skills for marketers, ready to drop into Claude Code, Claude Desktop, or Cowork and run against your own Search Query Performance export.
The Scoring: Turning Raw Rows Into a Funnel
Every month, the skill scores a brand on four numbers.
| Score | Funnel Stage | What It Measures |
|---|---|---|
| Tile Score | Impression to click | Are people clicking on the brand at all |
| Page Score | Click to cart | Does the listing convert once someone lands |
| Offer Score | Cart to purchase | Do people complete the purchase at the listed price |
| Market Share | Category-wide | Share of the defined market the brand actually holds |
Brands often treat “conversion is down” as a single problem with a single fix. It rarely is. A brand with a weak Tile Score has a discovery problem: nobody is clicking, so the fix is imagery, title, or placement. A brand with a strong Tile Score but a weak Offer Score has a completely different problem: people are interested enough to add to cart and then walk away at checkout, which points to price or offer, not creative. Splitting the funnel into these four numbers is what makes it possible to tell the two apart instead of guessing.
Two Mistakes That Quietly Wreck a Naive Analysis
Two classification errors show up in almost every raw dataset, and both distort the numbers that matter most.
The first is unlabelled competitors getting counted as generic demand. If a competitor brand name isn’t recognized, its search volume gets folded into the unbranded bucket, inflating how big the “true” market looks.
The second is brand misspellings getting counted as unbranded demand. A shopper searching a slightly misspelt version of a brand’s own name isn’t generic demand, she already knows the brand. Miscounting her search understates how much of the traffic is actually branded.
Both errors feed directly into the number most people use to judge whether growth is real: the branded versus unbranded split. Get the classification wrong, and that judgment comes out wrong, confidently.
Determining the Market You’re Competing In
The single most consequential decision in any SQP analysis is defining the market a brand is competing in. This has to come from the data itself, never assumed going in.
Sometimes the relevant boundary is an age band. Sometimes it’s a price tier. Sometimes it’s a specific product category nested inside a larger one. It has come out a different shape every time Smacient has run this.
Assume the wrong boundary and the output doesn’t look obviously broken. It produces a fluent, confident-sounding analysis of the wrong market, with nothing in the report itself revealing that the error happened. This is the step Smacient spends the most care on, because it’s the one place a mistake hides in plain sight.
The Four Insights It Surfaces
Once the scoring and the market boundary are right, the skill produces four kinds of insight that a manual read of the raw export wouldn’t surface.
A ranked to-do list. Every recommended action comes with an expected lift in orders per month, ranked by impact, so it’s clear what to fix next and what it’s worth.
Real growth versus borrowed growth. Revenue up 40 per cent feels like a win until it’s clear how much of that is the brand gaining share versus the whole category expanding and carrying it along for free.
The exact funnel stage where the sale is being lost. It can break at three points: nobody clicks, they click and abandon the cart, or they add to cart and abandon at checkout. Each is a different fix, and conflating them wastes effort on the wrong lever.
Growing search terms where the brand is nearly invisible. Terms doubling in size while a brand holds under 1 per cent market share look like nothing in a spreadsheet. They’re usually the highest-value row on the page.
Why Impact Is Measured in Orders, Never Currency
Every action point the skill produces is stated in additional orders per month, never in rupees, never in dollars.
Margin at the individual product level is rarely known with any precision, especially across a large catalogue. A profit figure looks precise, but it’s built on an assumption that can’t be verified. Orders are something that can be confirmed to have happened. A number that can be trusted completely is worth more than one that looks more impressive and is quietly built on a guess.
What You Get Out of It
Each run produces four documents.
| Whoever needs to defend or extend the analysis carries the full detail behind every number. | Built For |
|---|---|
| Growth Action Plan | Whoever owns execution, ranks actions by expected additional orders per month |
| Analyst Report | Whoever needs to defend or extend the analysis, carries the full detail behind every number |
| Leadership Note | Anyone who needs the headline without the methodology, a plain-language summary |
| Metric Definitions | Anyone downstream who needs to know exactly what Tile Score or Offer Score means |
What This Doesn’t Cover Yet
It has only been tested against Amazon. in. Currency, the event calendar, and price thresholds haven’t been validated on any other marketplace yet.
It handles monthly exports only. Weekly and quarterly reporting cycles aren’t supported today.
No large-catalogue brand has run through it yet. Every brand it’s been applied to so far has been small to mid-sized, so the logic for an already-dominant brand approaching saturation is barely exercised.
The tokeniser is Latin-script only, so non-Latin search queries fall into an unclassified bucket rather than being read properly. On Amazon. in, that gap is real, not theoretical.
FAQs
Because the report ranks terms by how well a brand converts on them, not by search volume. A term drops off the list the moment a brand stops converting well on it, even if the term is still being searched just as much. That makes the churn structural rather than a sign of unreliable data.
Seller Central hides 12 of the 33 available columns by default. Check column settings and expand the view before exporting. Otherwise, the analysis will be working from a partial report without realising it.
No. That column divides clicks by search volume, not by impressions, so it understates true click-through performance by roughly twentyfold if read as a standard CTR.
Because margin at the individual product level is rarely known precisely across a full catalogue, and a profit number that looks precise but rests on an unverifiable assumption is less useful than an order count that can be confirmed.
Not yet with full confidence. It has only been validated against Amazon. Currency, event calendars, and price thresholds haven’t been tested on other marketplaces.
It’s open source on GitHub at smacient/marketing-skills/skills/amazon-sqp-analysis, free to clone and run against your own Amazon Brand Analytics export.
Related Blogs
- Claude Skill for Amazon Market Research (No Helium10 or JungleScout subscription required)
- How to Extract Amazon Reviews Data Directly into Claude Using Smacient
- How to Extract Amazon Reviews Data into Google Sheets
- How to Extract Amazon Reviews: 5 Methods
- 12 Best Amazon Scrapers in 2025
- 10 Best Ecommerce Scraper APIs
- Top Claude Connectors to Automate Workflows and AI Tasks for Marketers
- Claude Cowork vs Claude Chat vs Claude Code: Which Should Marketers Use?
