Skip to main content
Nova Analytics
Try for free

How to use AI to find wasted Amazon PPC spend and negative keywords

To find wasted Amazon PPC spend with AI, the assistant needs live access to your search terms, negatives and product margins, and Nova's MCP connector gives Claude, ChatGPT and other assistants that access to Amazon Ads and Seller Central. Through Nova, the assistant lists the search terms that spent without a sale, checks spend against each product's break-even ACOS, skips what is already blocked or paused, and gives you the negative keywords to add in Amazon Ads.

Updated

Connect your AI Copy the playbook
Example with made-up figures
Claude

Where is my Amazon ad spend being wasted?

NUsed Nova

Over the last 60 days, $110.30 of your $2,400 ad spend went to four search terms with no sale. Negate "plastic water bottle" and "kids water bottle" ($64.60): neither fits your steel bottle. "Insulated water bottle 32 oz" sells, but at 53.4% ACOS against a 30% break-even, so lower its bid. Apply these in Amazon Ads.

Ask anything

The playbook

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 period or the threshold if you need to.

This needs an assistant connected to Nova. Start the 14-day free trial, then connect your assistant.

Use Nova to find wasted spend in my Amazon ads and list the negative keywords to add.

Period: the last 60 complete days. If I name other dates, use those.

1. Set the bar. Call Nova's query_advertising with groupBy "childAsin" and the metrics spend, sales, orders, acos, breakEvenAcos and profitAfterAds. For each advertised product, work out what one order earns before ads: sales / orders x breakEvenAcos. Use the amount of the product with the most ad spend as the spend threshold below, unless I name one. Say which amount you used.

2. Size the waste on targets. Call Nova's query_advertising with groupBy "keyword" and the metrics spend, clicks, sales, acos, breakEvenAcos, profitAfterAds and spendWithoutSales, sorted by spendWithoutSales. Set aside targets whose state or servingStatus shows they are paused, because that money has already stopped. Call it again sorted by profitAfterAds, sortOrder "asc", to find live targets selling above their break-even ACOS.

3. Find the queries that never sold. Call Nova's query_search_terms with groupBy "searchTerm", the metrics spend, clicks, sales and orders, sorted by spend, and metricFilters spend above the threshold and sales below 1. Run it once with adProducts ["sponsored_products"] and once with ["sponsored_brands", "sponsored_brands_video"]. Ignore rows with the search term "*": those are product page placements, not queries. A row that looks like an ASIN comes from a product target, so treat it as a negative product target. Never add sales across ad types. Then call it once more for Sponsored Products with only sales below 1 as the filter, and limit 200. Its summary.total is all the spend on no-sale queries, however small each one is. Look for words that repeat across them.

4. Remove what is too early. Repeat step 3 with the period ending 14 days earlier. A query on both lists has had time for its sales to land. Mark the others "too early", and any query under 10 clicks too.

5. Find the queries that sell at a loss. Call Nova's query_search_terms for Sponsored Products with metricFilters spend above the threshold, sales above 0, and acos above the break-even ACOS from step 1. A query carries no product margin, so say this comparison uses the product's break-even, not the query's.

6. Check what is already blocked. For the 5 biggest candidates, call Nova's list_negative_targets with q set to the query. q matches part of the text, so a rule counts only if its keywordText matches the query exactly. It covers Sponsored Products only, so say that Sponsored Brands negatives need checking in Amazon Ads.

7. Find where to act. For the 5 biggest candidates, call Nova's query_search_terms with searchTerms set to that one query and groupBy "adGroup". It returns the campaigns and ad groups the query spent in. If the query sold in one of them, say so before suggesting a negative.

8. Report, in three lists. Compare each query with the product names from step 1 to judge whether it is relevant.
   - Negate: query, spend, clicks, where it spent, negative exact or negative phrase, and why it does not fit the product
   - Lower the bid: relevant queries with no sales or an ACOS above break-even, with the figures
   - Leave alone: too early, already blocked, or the target is paused
   Then give the total spend in the Negate list as a share of ad spend for the same ad types, any word that repeats across no-sale queries, and the Negate list ready to paste.

Rules: you cannot change anything on my account, so I apply every change in Amazon Ads. Do not suggest a negative only because a relevant query has no orders. Say which figures are missing and why. Relay every notice Nova returns (P&L rules, marketplaces added up, COGS gaps, currency, ad history).

Raw markdown

How Nova customers use this

Real requests from Nova customers, names removed.

  • “Cross-check wasted search terms against existing negative targets to find negation candidates not yet blocked.”

    What the assistant did

    A recurring ad check. It listed the wasted search terms, removed the ones already blocked, then found which campaign each remaining term spent in.

    Nova tools used (2)
    1. query_search_terms
    2. list_negative_targets

  • “Daily ads pass: keywords with spend and no sales, as bid cut or negative candidates.”

    What the assistant did

    It compared each SKU with the day before, read campaigns over 7 days, listed the candidates, and checked whether earlier bid trims had held.

    Nova tools used (2)
    1. query_metrics
    2. query_advertising

  • “Search terms with spend and no sales as negative keyword candidates, and the best-selling search terms as harvest candidates, for the monthly ads report.”

    What the assistant did

    A monthly report: a scorecard against last month, a split by ad product, top and bottom campaigns, ASINs with a break-even ACOS check, then negatives and harvest candidates.

    Nova tools used (5)
    1. list_accounts
    2. query_metrics
    3. query_advertising
    4. query_search_terms
    5. query_dsp

  • “30-day search-term report for the Broad and Auto discovery campaigns, to find negatives and harvest candidates.”

    What the assistant did

    A PPC optimisation run: a campaign baseline, keyword bids, whether earlier changes held, then negatives, with Search Query Performance on the core queries next to it.

    Nova tools used (4)
    1. list_accounts
    2. query_advertising
    3. query_search_terms
    4. query_sqp

How much can a search term spend before it counts as wasted?

ACOS is ad spend divided by ad sales, with no margin in it. Break-even ACOS is the ACOS at which profit after ads is zero, and it equals the margin before ads. query_advertising returns it per product from your own costs.

What one order earns before ads is ad sales per order times break-even ACOS: at $40 and 30%, that is $12. A search term that spent more than that without a sale could not have paid for its clicks with its first sale. So the threshold is different per product, and it replaces a flat rule such as a fixed number of clicks. More in the break-even ACOS guide.

Which search terms spent and never sold?

query_search_terms with metricFilters returns customer queries with spend above the threshold and sales below 1. It covers Sponsored Products, Sponsored Brands and Sponsored Brands Video, and only returns queries with at least one click. Rows with "*" are product page placements.

Nova's spendWithoutSales, shown as Wasted Spend, is spend on targets with no sales over the whole period. A target can sell while some of its queries never do, so the query list can find waste that spendWithoutSales misses. The playbook reads both. A term is only firm if it also qualifies on a period ending 14 days earlier, because sales can land after the click.

The playbook also reads every no-sale query, however small, and looks for words that repeat across them.

Is it already blocked, or already paused?

list_negative_targets lists your negative keywords and product targets with their level (campaign or ad group) and match type, for Sponsored Products only. The negatedLevel field on a search term is empty almost everywhere, and that does not mean nothing is blocked: a negative that works stops the query appearing at all.

Targets carry state and servingStatus, so the assistant can set aside waste on a target that is already paused. Otherwise the biggest line in the report can be money that already stopped.

query_search_terms grouped by adGroup with searchTerms set to one query shows which campaigns and ad groups it spent in, so you know where the negative goes, and whether the query sold elsewhere.

Should I negate it or lower the bid?

A negative is for a search that does not fit the product. A relevant search with no sales is usually a bid, price or listing question.

Spend above the threshold, no sales, query unrelated to the product

What it usually means

The ad showed for the wrong search

What to do in Amazon Ads

Add a negative exact in that campaign or ad group

The same word across several no-sale queries

What it usually means

A pattern, not one bad query

What to do in Amazon Ads

Add a negative phrase, after checking no selling query contains the word

Relevant query, no sales

What it usually means

The query fits, so price, listing or bid may be the problem

What to do in Amazon Ads

Lower the bid and check the listing. Do not negate yet

Sales, ACOS above break-even

What it usually means

Each sale loses money after ads

What to do in Amazon Ads

Lower the bid, or keep it as a chosen cost during a launch

On the 60-day list only

What it usually means

Its sales may not have landed

What to do in Amazon Ads

Wait and run it again

An ASIN row with spend and no sales

What it usually means

A product target that does not convert

What to do in Amazon Ads

Add a negative product target

Large spendWithoutSales on a paused target

What it usually means

The spend already stopped

What to do in Amazon Ads

Nothing to do

Example with made-up figures

A wasted-spend check, worked through

An insulated steel water bottle. One order brings in $40 and the break-even ACOS is 30%, so one order earns $12 before ads, and $12 is the threshold. Ad spend over the last 60 days is $2,400.

plastic water bottle

Spend

$46.20

Clicks

38

Orders

0

Sales

$0

ACOS

none

Verdict

Negate: the bottle is steel

kids water bottle

Spend

$18.40

Clicks

15

Orders

0

Sales

$0

ACOS

none

Verdict

Negate: not a kids' product

water bottle with straw

Spend

$31.50

Clicks

29

Orders

0

Sales

$0

ACOS

none

Verdict

Lower the bid: the bottle has a straw lid, so check the listing

insulated water bottle 32 oz

Spend

$64.10

Clicks

52

Orders

3

Sales

$120

ACOS

53.4%

Verdict

Lower the bid: loses $28.10 at a 30% break-even

steel bottle for gym

Spend

$14.20

Clicks

9

Orders

0

Sales

$0

ACOS

none

Verdict

Leave alone: under 10 clicks

The four search terms with no sale spent $110.30. The Negate list is $64.60, or 2.7% of ad spend. The target "bottle" also shows $52 of spendWithoutSales, but it has been paused since September, so there is nothing to do.

$120 of sales at a 30% break-even earns $36, against $64.10 of spend, which is the $28.10 loss. These changes are made in Amazon Ads. Then check the same search terms next month.

Can I just paste my search term report into ChatGPT?

You can, but the file holds no product margins, no existing negatives and no paused status, so you type those in or the assistant goes without them, and it is a snapshot you export again next time. With Nova the assistant reads all of them from your account.

This playbook only reads, so you apply its changes in Amazon Ads. PPC management is in beta and opening to the public soon. You can already request access.

What this cannot tell you

  • This playbook only reads, so you apply its changes in Amazon Ads. PPC management is in beta and opening to the public soon. You can already request access.
  • query_search_terms returns queries with at least one click. A query blocked before the period does not appear at all.
  • Amazon's search term report has no ASIN, so Nova gives a query no margin of its own. Break-even ACOS is read per product or per target.
  • There are no impressions or click-through rate per query.
  • Sponsored Display has no search term report.
  • list_negative_targets covers Sponsored Products only.
  • Break-even ACOS and profit after ads cover Sponsored Products at every level and Sponsored Display per product. They are empty for Sponsored Brands and on Vendor Central.
  • Sales count at each ad type's own window: Sponsored Products 7 days on Seller Central and 14 on Vendor Central, Sponsored Brands 14. Compare spend across ad types, never sales.
  • Break-even ACOS uses the costs you entered. Units with no COGS overstate margin, and Nova returns a notice when that happens.
  • Ad history starts about 60 days before you connected Amazon Ads. historyStartsAtByProduct gives the exact date per ad type.

For the full account review, see the Amazon PPC audit. To compare tools, see the best Amazon PPC software. To see the same data in the app, see PPC analytics.

Questions sellers ask

Can ChatGPT or Claude find wasted PPC spend without connecting my account?

Only from a report you export and paste. The assistant then works from that file alone, without your margins, your negatives or which targets are paused. With Nova it reads your Amazon Ads and Seller Central data directly.

Does this add the negative keywords for me?

Not this playbook: every Nova tool it uses only reads, so you add negatives and change bids in Amazon Ads. PPC management is in beta and opening to the public soon. You can already request access. The connection uses OAuth, per account, and you can revoke it at any time.

How much spend should a search term have before I negate it?

Start with what one order earns before ads, which is ad sales per order times break-even ACOS, and at least 10 clicks. That is a starting rule, so tell the assistant if you want another amount. A relevant search with no sales usually needs a lower bid or a better listing, not a negative.

Why does a search term not have its own break-even ACOS?

Amazon's search term report has no ASIN, so a query is not tied to one product's margin in Nova. Nova reads break-even ACOS and profit after ads per product and per target, and the playbook compares queries with those.

Does this cover Sponsored Brands and Sponsored Display?

Search terms cover Sponsored Products, Sponsored Brands and Sponsored Brands Video. Sponsored Display has no search term report. Break-even ACOS and profit after ads cover Sponsored Products at every level and Sponsored Display per product. Negative keyword lists cover Sponsored Products only.

How recent is the data?

Search terms refresh three times a day. Sponsored Products and Sponsored Display spend refreshes within the hour. Sales can still land after a click, which is why the playbook checks a period ending 14 days earlier. Details on refresh times.

Does this work for Vendor Central?

Search terms and negative keywords do, with channels set to VENDOR. Break-even ACOS and profit after ads come back empty on Vendor Central, so read ACOS and sales instead.

How is this different from a PPC audit?

An audit reviews the whole account in eight checks, including placements, bids and hidden winners. This playbook does one job: it finds the spend that earned nothing and the negatives to add.

Ask your assistant where your ad spend goes

Connect your AI