# How to analyze Amazon Search Query Performance with AI

A playbook for an AI assistant connected to Nova. It reads your Search Query Performance, finds the queries where you lose share, says which step of the funnel is weak, and checks your ads against them. It only reads your data.

Every step calls a Nova tool that reads your live Amazon data, so paste this into Claude, ChatGPT or any assistant connected to Nova's MCP server. Change the grain if you want a monthly read.

```text
Use Nova to analyze my Amazon Search Query Performance and tell me what to fix.

Period: the latest complete week. If I ask for a trend or a year-over-year read, use grain "month".

1. Get the period. Call Nova's query_sqp with grain "week", no dates, and read period.covers. Those are the dates for every other call. They run Sunday to Saturday. If dataFloor is present, say the plan withheld older data.

2. Read the market you are really in. Call Nova's query_sqp with groupBy "keyword", queryTypes ["core", "presence", "suggested"], and the metrics search_volume, search_volume_wow_pct, your_impression_share, your_click_rate, market_click_rate, your_conversion_rate, market_conversion_rate, your_median_purchase_price, market_median_purchase_price and your_purchases, sorted by search_volume, limit 30. Leave out too_broad and long_tail, because they are category noise. Use summary for any total: never add market figures across rows.

3. Name the weak step, per query. Impression share low: you are not being seen. Click rate below the market: main image, title or price. Conversion rate below the market: price, reviews or listing. Your median price above the market's: say so. Read attrs.reason and attrs.action for Nova's verdict, and say when you disagree with it.

4. Check your brand. Call Nova's query_sqp with queryTypes ["own_brand"], groupBy "keyword", sorted by search_volume, limit 10. Then the same with ["competitor"]. Say where rivals take purchases on your brand terms, and where you take theirs.

5. Check your ads. For the top 10 queries from step 2, take every spelling in attrs.variants. Call Nova's query_search_terms with searchTerms set to those spellings, and from and to set to period.covers, with spend, sales, clicks, orders and acos, groupBy "searchTerm". Say which big queries get no ad spend and which spend without selling.

6. Check the listing. For the products behind the weak queries, call Nova's query_sqp with groupBy "parentAsin" to name them, then query_listing_health for their title, bullets and backend keywords. Say whether the query's words appear in the listing.

7. Check the trend. For the top 5 queries, call Nova's query_sqp with grain "month", groupBy "period" and keywords set to one query at a time, with search_volume and your_impression_share, over the last 6 months. Say whether the gap is new.

8. Report, one row per query, largest first: search volume, your share, the weak step, the figures behind it, ad spend on it, and one action. Then the three actions worth the most.

Rules: this is every seller's organic behaviour on the query, not your ads. Say which period the data covers. Never add market figures across rows. Say which figures are missing and why, and relay every notice Nova returns. You cannot change anything on my account.
```

## How to read the result

| What you see | Where it points | What to do |
| --- | --- | --- |
| Volume up, your share flat or down | Demand you are not catching | Bid on it, and check the query's words are in the listing |
| Share fine, click rate below the market | Shoppers see you and pass | Main image, title, price |
| Clicks fine, conversion below the market | They come and do not buy | Price, reviews, listing content |
| Your median price above the market's | Possibly priced out | Check it against your margin first |
| Big query, no ad spend | A gap in your ads | Test a keyword target |
| Rivals buying on your brand terms | Brand defence | Bid on your own brand terms |
| Volume down, share flat | Demand, not you | Wait, and check the monthly trend |

## Limits

- Search Query Performance needs Brand Analytics, so it needs Brand Registry on Seller Central. Amazon does not serve it to vendors.
- Weekly data runs about 10 days behind, and monthly about 7 weeks.
- SQP weeks run Sunday to Saturday, so pass `period.covers` to any other tool.
- A query's figures repeat on every parent ASIN that ranks on it, so totals come from `summary`.
- Data is per parent ASIN, not per SKU.
- It shows the market on a query, not each rival's share.
- A trial may withhold older periods: the answer says so in `dataFloor`.