An Amazon PPC audit checks whether your Sponsored Products, Sponsored Brands and Sponsored Display spend goes to the right searches, products and placements. Connect Claude to Amazon Ads read-only, pull structure, search terms, budgets and placement data, work through ten checks, then turn the findings into a dry-run change plan you approve line by line.
- Run it on a read-only connection (https://www.whathead.com/mcp/read) so nothing can change by accident.
- Pull search terms at least monthly: the Ads API keeps 65 days of Sponsored Products search-term data.
- Judge ACOS against your break-even ACOS (your margin), not someone else’s benchmark.
- Fix in batches, from a written plan, with a preview of every change.
The steps use Claude with Whathead’s Amazon Ads connection, but the checklist works with any Amazon Ads MCP server, or with exports from the Amazon Ads console. Amazon facts come from Amazon’s own docs, checked on 19 November 2026. Whathead makes one of the tools mentioned.
What is an Amazon PPC audit?
A PPC audit is a structured review of a pay-per-click account: where the money goes, what it returns and what to change. An Amazon PPC audit applies that to sponsored ads (Sponsored Products, Sponsored Brands and Sponsored Display) and to the keywords, product targets, negatives, bids, placements and budgets inside them.
It differs from a Google Ads audit in ways that change the checklist. Shoppers search with buying intent on one store, every ad leads to a product page, and Amazon splits its data by profile, ad product and report type, each with its own history window. Miss a window and the data is gone from the API.
Doing it with Claude means the assistant pulls the reports, joins them and does the arithmetic, and you make the decisions. If you haven’t connected Claude to Amazon Ads yet, start with the Amazon Ads MCP guide.
What do you need before you start?
Data for the audit and how far back it goes
| Where it comes from | History in the API | |
|---|---|---|
| Campaign, ad group, keyword and target settings | List calls for SP, SB and SD, or Amazon exports | Current state |
| Campaign performance | Campaign reports (spCampaigns, sbCampaigns, sdCampaigns) | v3: SP 95 days, SB 60, SD 65. Reporting API v1: 24 months of daily data |
| Search terms | Search term reports (spSearchTerm, sbSearchTerm) | v3: SP 65 days, SB 60 days |
| Placements | SP campaign report grouped by campaignPlacement | Same as the campaign report |
| Budget pressure | Budget usage and budget recommendations | Time in budget over the past 7 days |
| Total sales, for TACOS | Seller Central or Vendor Central business reports, not the Ads API | Your own export |
| Margins and target ACOS | Your finance data | Not in Amazon |
One trap: the API’s search term report only includes impressions that led to at least one click, while the console’s report includes every term with an impression, so the row counts won’t match (Amazon’s FAQ). Product-page placements with no search keyword show the search term as an asterisk.
Write down three numbers before you start: product margin (for break-even ACOS), target ACOS by product line and average order value. Every threshold below comes from them, not from an industry average.
How do you run the audit read-only through /mcp/read?
- Connect Amazon Ads in Whathead and activate the profiles you want to audit (setup guide).
- Add the read-only connector. In Claude, add a custom connector with https://www.whathead.com/mcp/read. It lists only read tools, including all 70 Amazon read actions and report requests, and refuses any write before it leaves Whathead.
- Pick one profile. Ask Claude to list your profiles, then audit one marketplace at a time.
- Request the reports in one go: campaign (daily), campaign by placement, targeting, search term and advertised product, all for the last 60 days. A v3 report covers at most 31 days, so Claude requests two 30-day windows. Reports run asynchronously and can take several minutes.
- Give Claude your numbers. Paste margin, target ACOS and average order value, then run the checks below in order. Ask for tables with entity IDs so findings can become changes later.
Sixty days is the safe window because it fits inside every report type above, including Sponsored Brands search terms.
1. Is the account and profile structure clean?
Start with what exists. Ask Claude to list portfolios, campaigns by ad product, targeting type (auto or manual) and state, and to count ad groups, ads and targets per campaign.
- Active campaigns sit in portfolios that match how you report. Amazon’s best-practice guide suggests organising them "by brand, category, season".
- Each campaign holds ASINs from one category, as the same guide recommends.
- Names follow one convention, so ad type, targeting and match type can be read from the name (naming conventions).
- No enabled campaign has zero enabled ads or zero enabled targets: those look live but can’t serve.
- Each profile’s marketplace and currency match what you think you are auditing.
2. Do your campaign types cover discovery and control?
Look at the spend split between automatic and manual Sponsored Products campaigns, and between SP, SB and SD. Automatic targeting finds new search terms; manual keyword and product targeting controls bids on what already works.
- All spend in automatic campaigns: you have little bid control over proven terms.
- No automatic campaigns at all: you may have stopped discovering new terms.
- Products with sales but no ads, or ads on products that are out of stock or unprofitable.
3. Are match types and negatives doing their job?
Keyword targeting uses broad, phrase and exact match. Negative keywords use negative exact or negative phrase, and Amazon lets you set them at campaign or ad group level; a campaign-level negative applies to every ad group in it.
- Converting search terms from broad and auto campaigns that were never added as exact keywords.
- Exact keywords that also run in broad or phrase elsewhere with no negatives, so your own campaigns compete for the same search.
- Negatives that block terms which used to convert. Ask Claude to cross-check every negative against the last 60 days of orders.
- Duplicate keywords with different bids in different ad groups.
4. Are you harvesting search terms?
Search-term harvesting means moving terms that convert into exact-match keywords with their own bids, and blocking terms that only spend. Sponsored Products search-term data is kept for 65 days in the API, so harvest at least monthly or the evidence ages out.
Two rules keep it objective. A term that converts at or under your target ACOS with a few orders is a candidate for exact match. A term whose spend has passed your allowable cost per order (average order value multiplied by target ACOS) without an order is a candidate for a negative. Search terms that are ASINs (usually starting with B0) belong in product targeting, not keyword lists.
5. What do ACOS, TACOS and ROAS tell you?
Efficiency metrics
| Formula | Use it for | |
|---|---|---|
| ACOS | Ad spend ÷ ad sales × 100 | Campaign and keyword efficiency against your target |
| ROAS | Ad sales ÷ ad spend (the inverse of ACOS) | Comparing Amazon with other channels |
| Break-even ACOS | Your profit margin before ad spend | The line above which ads lose money |
| TACOS | Ad spend ÷ total sales × 100 | Whether ads grow the business or just buy sales you would have had |
The ACOS and ROAS formulas are Amazon’s own, and Amazon says that "to maintain a profit, your Amazon ACOS needs to be lower than your profit margin". TACOS isn’t an Amazon console metric: it needs total sales from Seller Central or Vendor Central. Ask Claude to report ACOS by campaign and by advertised ASIN, flag anything above break-even, and, if you paste total sales, chart TACOS by month.
Keep the attribution column consistent: v3 reports offer sales and purchases over 1, 7, 14 and 30 days, and mixing them makes campaigns look better or worse than they are.
6. Are budget caps limiting your best campaigns?
A campaign that runs out of budget at midday misses the afternoon. Amazon’s budget recommendations return a recommended daily budget, the share of time the campaign was in budget over the past seven days, and estimated missed impressions, clicks and sales. Amazon calls these estimates, not a guarantee, and suggests waiting until a campaign has run for nine days.
- Campaigns under your target ACOS that are out of budget part of the day: candidates for more budget.
- Campaigns over break-even that always stay in budget: candidates for less.
- Seasonal peaks with no plan: Amazon’s budget rules can raise budgets for dates or events.
Whathead’s budget pacing board and anomaly alerts cover Meta, Snapchat, TikTok, X and DV360, not Amazon, so on Amazon this check runs through the budget usage and recommendation reads above.
7. Are bids and placements set on purpose?
Amazon’s dynamic bidding guide names three Sponsored Products bidding strategies: fixed bids, dynamic bids down only, and dynamic bids up and down, which can raise or lower bids by up to 100%. On top of that you can add placement adjustments of up to 900% for top of search, rest of search and product pages (best practices).
- Pull the campaign report grouped by placement and compare ACOS for top of search, rest of search and product pages.
- Large top-of-search adjustments on campaigns where top of search has the worst ACOS.
- Up-and-down bidding plus a large placement adjustment on the same campaign: the two stack.
- Keyword bids far above or below Amazon’s bid recommendations, with no reason written down.
8. What should you check in Sponsored Brands and Sponsored Display?
- Sponsored Brands: keywords and product targets, negatives, the ad formats running (product collection, store spotlight, video), and the SB search term report, which keeps 60 days.
- Sponsored Brands placements: use the SB placement report to see where spend lands.
- Sponsored Display: the split between product, category and audience targeting, the creatives in use, and ACOS by targeting type. SD campaign data is kept for 65 days in v3.
- Both: new ads go through Amazon moderation, so look for ads that are not serving.
9. Is product targeting pulling its weight?
Product targeting puts your ads on specific ASINs or categories. Ask Claude for ACOS and orders by product target, and check:
- ASIN targets that spend without orders: candidates for lower bids or negative product targets.
- Category targets that are too broad for the product.
- Your own ASINs targeted in a way that bids against your own organic listings, unless that is a deliberate defensive tactic.
- Competitor ASINs that convert well in search-term reports but aren’t targeted directly yet.
10. Where is spend being wasted?
Common waste patterns
| How Claude finds it | Usual fix | |
|---|---|---|
| Search terms with spend and no orders | Search term report against your allowable cost per order | Negative exact or phrase |
| Self-competition | Same term targeted in several campaigns or match types | Pick one owner and add negatives |
| Over-budget losers | High ACOS campaigns that never run out of budget | Lower budget or bids |
| Starved winners | Low ACOS campaigns out of budget part of the day | Raise budget |
| Placement drag | Placement with much higher ACOS than the campaign average | Lower that placement adjustment |
| Dead weight | Enabled targets with no impressions in 60 days | Pause, review bids or relevance |
Ask Claude to put a spend figure on each finding, so the change plan starts with what costs the most.
Should you audit in an Amazon PPC tool or with Claude?
Amazon PPC software (rule engines, bid automation, dashboards) and an AI assistant on an MCP server do different jobs, and many teams use both.
Amazon PPC software vs Claude with an MCP server
| Amazon PPC software | Claude with an MCP server | |
|---|---|---|
| Always-on automation | Strong: rules and bid changes run every day | No: a person starts each session |
| Custom questions | Limited to the tool’s reports and filters | Strong: any join or check you can describe |
| Explaining why | Charts and alerts | Plain-language reasoning with the numbers |
| Other ad platforms | Usually Amazon-focused | Depends on the server; Whathead adds Meta, Google, TikTok and more |
| Changes | Applied by the tool’s rules | Previewed and confirmed one batch at a time |
"Amazon PPC automation" usually means the first column: rules that run without you. An MCP connection is better at the audit itself and at one-off fixes. Some PPC tools now offer MCP access for their own customers, so it doesn’t have to be either-or.
How do you turn findings into a dry-run change plan?
Ask Claude to write the plan before anything changes, one row per change:
Change plan template
| What goes in it | |
|---|---|
| Finding | The check and the evidence, with spend and dates |
| Change | Add negative, raise budget, lower bid, pause target |
| Entities | Profile ID, campaign ID, ad group ID, keyword or target ID |
| Old → new | Current value and proposed value |
| Risk | What could go wrong, and how you would undo it |
- Review the plan and delete the rows you disagree with.
- Switch to the write connection (https://www.whathead.com/mcp). Every Amazon write starts as a dry run that shows the exact request.
- Approve one batch at a time: negatives first, then bids, then budgets.
- Deletes and archives need an extra confirmation. Prefer pausing, because archived Amazon entities can’t be changed again.
- Re-run the same checks two weeks later on the same window to see what moved.
For batching, naming and rollback habits that apply on every platform, see bulk editing ads safely.
Prompts you can copy
- “List my Amazon Ads profiles. For profile [ID], count campaigns by ad product, targeting type and state, and list enabled campaigns with no enabled ads or targets.”
- “Request the Sponsored Products search term report for the last 60 days. My average order value is $[X] and target ACOS is [Y]%. List terms whose spend is above AOV × target ACOS with zero orders.”
- “Find search terms with 2 or more orders at or under [Y]% ACOS that aren’t exact keywords anywhere in this profile.”
- “Show ACOS by placement for my 15 biggest campaigns and flag top-of-search adjustments where top of search has the worst ACOS.”
- “For every campaign, show the recommended daily budget, percent time in budget and estimated missed sales. Sort by missed sales.”
- “My margin is [Z]%. Flag every campaign and ASIN with ACOS above that, with spend over the last 60 days.”
- “Turn these findings into a change plan table with entity IDs and old and new values. Don’t make any changes.”
Amazon PPC audit: common questions
How often should you run an Amazon PPC audit?
A light audit monthly, because the Ads API keeps only 65 days of Sponsored Products search-term data and 60 days for Sponsored Brands. A full audit of structure, bidding and placements each quarter, and before big events such as Prime Day.
What is a good ACOS on Amazon?
There is no universal number. Amazon’s own guidance is that ACOS must stay below your profit margin to make money, so start from your break-even ACOS and set targets per product line.
What is the difference between ACOS and TACOS?
ACOS divides ad spend by ad-attributed sales. TACOS divides ad spend by total sales, including organic, so it shows whether advertising grows the whole business. TACOS needs total sales from Seller Central or Vendor Central.
Can Claude fix the problems it finds?
Yes, on a write-capable connection. In Whathead every Amazon change is a dry-run preview first, and deletes need an extra confirmation. Keep the audit itself on the read-only connection.
Do I still need Amazon PPC software?
If you want rules that run every day without you, yes. Claude with an MCP server is strongest at audits, custom questions and reviewed one-off changes. Many teams use both.
Audit Amazon Ads from Claude, read-only
Connect Sponsored Products, Brands and Display to Claude and run every check without risking a change.
Sources
- Amazon Ads: Sponsored Products best practices, ACOS and ROAS, dynamic bidding
- Amazon Ads API: v3 campaign reports, v3 search term reports, v3 reporting FAQ, Reporting API v1 data retention, negative keywords, budget recommendations, budget rules, bulksheets error reports
- Whathead: Amazon Ads MCP, MCP server, Amazon Ads connection setup, read-only connection, pricing
Written by Ahmed Shahin, founder of Whathead. Last reviewed 19 November 2026.



