The search term report is the only file that tells you the truth.

Search terms are not keywords, and the distinction is the whole point

A keyword is what you bid on. A search term is what the shopper typed. In an exact-match campaign the two are nearly the same, which is why exact match is easy to manage and offers no discovery. In broad, phrase and auto campaigns, one keyword can match hundreds of distinct search terms, and the report is the only place the difference is visible.

This is why an account can look tidy in the campaign view and be a mess underneath. The campaign shows a keyword called wireless earbuds with an acceptable ACoS. The search term report shows that keyword matched wireless earbuds case, wireless earbuds for kids, and a competitor brand name, two of which never convert and one of which is carrying the whole average. You cannot see that from the keyword row, and you cannot fix it from there either.

How to read the file without fooling yourself

Sort by spend descending. Sorting by ACoS surfaces the rows with one click and no order, which are statistically meaningless and emotionally compelling, and you will spend an hour negating noise. Sorting by spend puts the rows that actually cost you money at the top, in the order they cost it.

Then classify each significant row honestly. Irrelevant means the query describes something you do not sell, in which case no bid makes it good and it becomes a permanent negative. Relevant but unprofitable means the query fits your product but is not converting, which is a bid, price or detail page question and should not be negated until you have looked at those. Converting means it earned its spend, and it deserves promotion into a manual campaign where you set the bid rather than leaving it to Amazon's matching. Most sellers do the negating and skip the harvesting, which is half a process.

The loop, and why it has to repeat

The full cycle is: auto campaign discovers language, report reveals which language earns, winners are harvested into manual exact, harvested terms are negated in auto, wasters are negated everywhere. Run once, it is a cleanup and the account improves. Run monthly, it compounds, because the auto campaign keeps working as a discovery engine while the manual campaigns hold the terms you have already proven.

The reason it degrades without maintenance is that Amazon's matching changes, seasonality changes what people type, and your own catalogue changes. A negative list built in March for a summer catalogue will be quietly wrong by October. Put the review on a calendar with a named owner. A process that depends on someone remembering is a process that runs twice.

How to work an Amazon search term report

1. Pull at least thirty days. Shorter windows on Sponsored Products give you rows with two clicks each, which is a report you cannot act on.
2. Sort by spend descending, not by ACoS. ACoS sorting puts your worst rows at the top and they are all one-click rows. Spend sorting puts the money at the top, which is what you are here about.
3. Split into three buckets. Converting queries to harvest, spending-nothing-back queries to negate, and relevant-but-weak queries that are a bid or listing problem.
4. Harvest winners into manual exact match. Create the term as an exact keyword in a manual campaign so you control the bid rather than leaving it to auto.
5. Negate every harvested term in the source campaign. This is the step people skip. Without it, your auto campaign and your new manual campaign bid against each other.
6. Add the irrelevant queries as negatives. negativePhrase where a word poisons everything containing it, negativeExact where only that one query is bad.
7. Book the next review before you close the file. Monthly is the usual cadence. Auto targeting keeps discovering new language and some of it is new and expensive.

The mistakes that cost the most

• Sorting by ACoS. It surfaces one-click rows and hides the spend. Sort by spend descending.
• Negating without harvesting. You removed the waste and left the wins in a campaign where you cannot control the bid. That is half the value.
• Pulling a seven-day window. Not enough data per row to classify anything. Thirty days minimum, longer for low-volume ASINs.
• Reading the report before the attribution window closes. Recent rows will understate sales, so you will negate terms that converted after you looked.
• Doing it once. Auto targeting never stops finding new queries. The report is a monthly job, not a project.

What you can ask Muze to do

Muze connects Amazon Ads to ChatGPT, Claude or any MCP client with 48 tools, 12 read and 36 write. These are the real tool names behind each request.

• “Recompute my Amazon search term recommendations for last month.”
  Recomputes recommendations across a date range you give in YYYY-MM-DD bounds. It generates recommendations only and applies nothing. Tools: refresh_amazon_search_terms, submit_amazon_reports.
• “Show me what is pending and what you would do with each.”
  Pending recommendations for the profile, each with a row id, split into winners to harvest as keywords and wasters to block as negatives. Tools: list_amazon_search_terms.
• “Apply rows one, four and nine. Dismiss the rest.”
  Apply writes the harvest or the negative to the live Amazon account. Dismiss clears the row locally and touches nothing on Amazon. Tools: apply_amazon_search_term, dismiss_amazon_search_term.
• “Pull all my keywords down so the recommendations are current.”
  Refreshes Muze's local copy of the account and shows which daily reports have been requested and ingested. Data refresh only, nothing changes on Amazon. Tools: sync_amazon_keywords, run_amazon_kickoff, list_amazon_reports.

Nothing spends without you

• Every write previews first and waits for explicit confirmation.
• New campaigns, ad groups and product ads are created paused. Negative keywords are the exception: they are created enabled, because an exclusion that is switched off is not an exclusion.
• Pause and archive are separate tools. Archiving is irreversible on Amazon, so it is never what happens when you asked to stop spend.
• OAuth only. No Amazon Advertising API application to file, no keys to manage, and access is revocable.
• Never a percentage of ad spend. You pay for the software. Your budget stays yours.

Frequently asked questions

Can an AI assistant read my Amazon search term report?

Yes. Muze pulls search-term recommendations through refresh_amazon_search_terms and list_amazon_search_terms, and can apply the ones you approve with apply_amazon_search_term, which harvests a term as a keyword or adds it as a negative depending on the recommendation. Applying mutates the live account, so it previews and waits for your confirmation.

What is the difference between a keyword and a search term?

A keyword is what you bid on. A search term is what the shopper actually typed. Broad, phrase and auto targeting mean one keyword can match hundreds of search terms, and the report is the only place you can see which ones.

How far back should the report go?

Thirty days as a minimum on most accounts, and longer on low-volume ASINs where a month produces rows with a handful of clicks each. Shorter windows give you a file you cannot classify.

What should I do with a query that converts well?

Harvest it into a manual campaign as an exact-match keyword so you set the bid directly, then negate it in the campaign it came from so the two do not compete in the same auction.

Do Sponsored Brands and Sponsored Display have search term reports?

Sponsored Brands reports on search terms for keyword-targeted campaigns. Sponsored Display is targeted by product, category or audience rather than keywords, so the equivalent diagnostic is the targeting and placement report rather than a search term report.

Can Muze run this loop on a schedule?

It can. update_amazon_autopilot_config and run_amazon_autopilot let Muze apply eligible search-term actions inside guardrails you set, and list_amazon_autopilot_actions gives you the audit log of everything it did. Autopilot is off unless you turn it on.

Keep reading

The rest of the Amazon cluster