Negative keywords are the cheapest optimisation on Amazon.

What negative keywords do on Amazon, precisely

A negative keyword tells Amazon not to serve your ad when the shopper's search term matches the exclusion. negativeExact matches only that exact query string, so negating running shoes blocks running shoes and nothing else. negativePhrase matches any query that contains the phrase in that order, so negating for dogs blocks shoes for dogs, boots for dogs and everything else that carries it. There is no negative broad on Amazon Sponsored Products keyword targeting, which is a genuine difference from Google Ads and trips up people crossing over.

Exclusions apply at ad group or campaign level. Ad group is the surgical option and campaign is the blunt one, and most sellers under-use campaign level out of caution. If a term should never serve for this product under any circumstances, campaign level is correct and saves you replicating the same exclusion into six ad groups.

Finding the ones worth adding

The search term report is the source. You are looking for two distinct populations, and conflating them is the most common error in this whole exercise. The first is irrelevant queries: searches for a product you do not sell, a size you do not stock, a competitor brand, or the word free. These get negated permanently, usually as negativePhrase, and there is no bid at which they become good.

The second is relevant queries performing badly. A term that genuinely describes your product but has spent money without converting might be an ad rank problem, a price problem or a detail page problem. Negating it removes the symptom and the diagnosis at once. Cut the bid, check the placement split, and look at whether the listing is actually competitive on that query before you exclude it. The discipline that separates the two: ask whether you would be happy to appear for this query at a lower cost. If yes, it is a bid decision. If no, it is a negative.

The negatives most accounts are missing

Modifier words that signal the wrong intent: free, cheap, used, refurbished, wholesale, bulk, and for a lot of catalogues, the word replacement or refill when you sell the base product rather than the consumable. Competitor brand names, unless conquesting is a deliberate strategy with its own campaign and its own target. Sizes, colours, quantities and voltages you do not stock, which auto targeting finds relentlessly. And accessories: if you sell the device, queries for the case will click and will not convert.

Then the structural one almost everyone skips. Every keyword you harvest out of an auto campaign into a manual exact campaign has to be negated back in the auto campaign. Otherwise both campaigns enter the same auction, you pay the higher of your own two bids, and the auto campaign's numbers slowly rot while nobody can work out why.

How to add Amazon negative keywords, step by step

1. Pull the search term report for a meaningful period. Thirty days minimum on most accounts, longer on low-volume ASINs. A week of data will have you negating terms that were simply unlucky.
2. Set a click threshold before you look. Decide in advance how many clicks with zero orders makes a term guilty. Writing the rule down before seeing the data is what stops you rationalising each row.
3. Separate irrelevant from underperforming. A query describing a product you do not sell is a permanent negativePhrase. A relevant query that converts badly is a bid problem, not an exclusion.
4. Choose the match type deliberately. negativeExact for a single bad query you still want the near-variants of. negativePhrase for a word that poisons every query it appears in, like free, used, refill or a competitor brand.
5. Place them at the right level. Ad group level when only that ad group is affected. Campaign level when the term should never serve for this product at all.
6. Negate harvested terms in the source campaign. Every keyword you promote from auto into a manual campaign must be negated in the auto campaign, or you bid against yourself.
7. Re-run it on a schedule. Once is a cleanup. Monthly is a process. Auto campaigns keep finding new language, some of it new and bad.

The mistakes that cost the most

• Negating a term after four clicks. That is noise. Pick a click threshold before you open the report and hold to it.
• Using negativePhrase where negativeExact was meant. One careless phrase exclusion can silently block a large family of profitable queries. Check what else contains the phrase.
• Negating relevant queries that just need a lower bid. You removed the revenue along with the cost. Ask whether you would take that click at half the price.
• Forgetting to negate harvested keywords in the auto campaign. Two of your own campaigns, one auction. This is the most expensive omission on the list.
• Treating it as a one-time cleanup. Auto targeting is a discovery engine that never stops. The exclusion list has to be maintained on a schedule.

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.

• “Add these as negative keywords across the ad group.”
  Batch or single ad-group exclusions in negativeExact or negativePhrase. Created ENABLED, so the exclusion takes effect immediately rather than waiting to be switched on. Tools: create_amazon_negative_keywords, create_amazon_negative_keyword.
• “Show me the search terms worth blocking this month.”
  Recomputes recommendations for a date range and returns them with a row id each, split into winners to harvest and wasters to block. Tools: refresh_amazon_search_terms, list_amazon_search_terms.
• “Apply that one, dismiss the other two.”
  Apply writes the harvest or the negative to the live account. Dismiss clears the recommendation and changes nothing on Amazon. Tools: apply_amazon_search_term, dismiss_amazon_search_term.
• “Exclude these competitor ASINs as well as the keywords.”
  Batch ASIN and brand exclusions for product targeting. Also created ENABLED. Tools: create_amazon_negative_product_targets.

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 add Amazon negative keywords for me?

Yes. Muze exposes create_amazon_negative_keyword and create_amazon_negative_keywords over MCP, so you can say which terms to block in ChatGPT or Claude and the assistant writes them to the account. The list previews before it is applied, and negatives are created ENABLED so the exclusion is live the moment you confirm.

What match types do Amazon negative keywords support?

negativeExact and negativePhrase. There is no negative broad match on Amazon Sponsored Products keyword targeting, which is a real difference from Google Ads.

How many clicks before I negate a keyword?

Pick a threshold in advance based on your conversion rate and your price point, and apply it consistently. The specific number matters less than deciding it before you look at the report, because deciding afterwards is how people talk themselves into negating good terms.

Should negatives go at ad group or campaign level?

Ad group level when only that ad group is affected. Campaign level when the term should never serve for this product at all. Campaign level is under-used and saves duplicating the same exclusion across ad groups.

Do negative keywords apply to auto campaigns?

Yes, and that is where they earn the most. Auto targeting is a discovery engine, and negatives are the only thing stopping it from continuing to pay for the bad language it discovered.

Can Muze do this automatically on a schedule?

It can. update_amazon_autopilot_config and run_amazon_autopilot let Muze apply eligible search-term actions within guardrails you configure. Enabling autopilot is an explicit choice and the tool description says so plainly, because autonomous means the confirmation step is gone.

Keep reading

The rest of the Amazon cluster