Amazon Keyword Research for Sponsored Products: Match Types, Harvesting, and Negatives

CH.01What match types actually do in 2026

Sponsored Products manual targeting offers three match types, broad, phrase, and exact, and all three already account for plurals, misspellings, and translations. The thing that breaks the old pyramid is what broad now does: a shopping query can contain the keyword terms in any order and may include singulars, plurals, variations, synonyms, and related terms, and the keyword itself may not even appear in the customer's query.

Amazon's own example makes the looseness concrete. The broad keyword sneakers can match queries like basketball shoes, cleats, trainers, or foam runners. None of those contain the word sneakers. So broad is no longer a literal-string promise, it is the loosest discovery dial. Phrase is tighter, requiring the close variations in sequence, and exact is the precise query plus its close variations. Read the three together as a looseness gradient, not a literal-match guarantee.

Automatic targeting uses groups, not match types

Automatic campaigns do not use broad, phrase, or exact at all. They use four targeting groups: close match, loose match, substitutes (similar products from other brands viewed on detail pages), and complements (products viewed alongside yours). For a product like Doppler 400-count Cotton Sheets, close match catches "400-count sheets," loose match catches "bed sheets," and complements show your ad on "queen quilt" and "feather pillows" detail pages. Those four groups are the cheapest discovery surface Amazon gives you.

CH.02Map the keyword universe before you spend

Before you assign a single match type, build the seed universe from sources that do not cost a click. The point is to enter the auction already knowing the words a buyer types when they are ready to purchase, described the way the buyer describes the product rather than the way your catalog names it. In our experience, indexing those terms in your listing copy and backend search terms first tends to help the campaigns that follow. In our experience, indexing those terms in your listing copy and backend search terms first tends to help the campaigns that follow.

• Your own listing copy and backend search terms, since these are the words your detail page is already built around.
• Amazon autocomplete, typed directly into the search box for your main product term.
• Competitor ASINs and titles, noting the recurring nouns and modifiers you can target with product targeting later.
• Brand Analytics Search Query Performance, the highest-signal zero-cost seed source Amazon publishes.

Bucket every seed by intent before you assign match types: clear purchase-intent terms, broader research terms, and brand terms. Brand Analytics deserves its own pass because it shows real query-level demand and conversion share. We cover that data layer in depth in our guide to Amazon Search Query Performance in 2026, and it is worth noting that AI-driven discovery is reshaping which queries convert, which we unpack in how Rufus is changing ASIN visibility.

CH.03Launch auto plus broad as your discovery layer

The cheapest way to find converting queries is to let Amazon find them for you. Launch an automatic campaign so its four targeting groups can surface queries and ASINs you would never have guessed, and pair it with a broad-match manual campaign on your strongest seed roots to widen the net further. Amazon explicitly recommends running automatic and manual campaigns together for exactly this reason.

1. Launch the automatic campaign firstLet close, loose, substitutes, and complements run as your research engine. Set a conservative discovery bid and dynamic bids set to down only so you learn without overpaying.
2. Add a broad-match manual campaignSeed it with your intent-bucketed roots. Broad now pulls synonyms and related terms, so it discovers adjacent demand the auto campaign may miss.
3. Give it a real run-inAmazon suggests letting an auto campaign run about two weeks before you build the manual campaign off its data. Resist the urge to harvest on day three.

CH.04Pull and read the search term report

The search term report is the single document that makes the rest of this work. Because broad and the auto groups buy queries that never literally match your keyword, the report is the only place that shows what you actually paid for. Pull it on a regular cadence: it is available for only a limited rolling window, so download it routinely rather than assuming history will be there later.

The columns that matter, and what they tell you

Sort the report into two piles. The first is converting queries with orders at or below your target ACOS, which become harvesting candidates. The second is spend-only queries that have accumulated clicks and cost without orders, which become negative candidates. Almost everything you do next is moving items from one pile to the right campaign.

CH.05Harvest winners up the match-type ladder

Harvesting is the canonical loop Amazon itself describes: use the automatic campaign's search term report to identify the best keywords and products, then add the top performers to manual campaigns where you can bid more competitively. The discipline is to move proven converting search terms up the looseness gradient, from auto or broad into phrase, then into exact, tightening control as the evidence strengthens.

1. Identify the proven queryA customer search term with orders and an ACOS at or below target, ideally with enough clicks behind it to trust the conversion rate.
2. Graduate it into exactCreate the winning term as an exact-match keyword in a tight performance ad group and bid up, since intent here is proven and you want to win the placement.
3. Use phrase for controlled expansionWhen you want close variations of a proven root without going fully wide, phrase sits between broad discovery and exact precision.
4. Mind the ceilingYou can hold up to 1,000 keywords per ad group for manual targeting, so structure tightly enough that one ad group's data stays readable rather than cramming everything into one.

CH.06Negate the leaks and seal cannibalization

Negatives do two jobs. They cut spend on queries that will not convert, and they stop your discovery campaigns from bidding against the exact campaign that already owns a proven query. Both jobs are governed by hard limits worth committing to memory.

Negatives can only be phrase or exact, and you cannot change the match type of an existing negative. Negative phrase allows a maximum of four words and 80 characters; negative exact allows a maximum of ten words and 80 characters. The four-word negative-phrase cap is a real-world snag: a long junk query can exceed it, so you sometimes need negative exact or a shorter phrase fragment to block it.

1. Wait for the evidenceAmazon recommends evaluating a keyword after it has received at least 20 clicks before adding it as a negative, then letting negatives run two weeks or more before changing strategy.
2. Negate spend-only leaksQueries with clicks and cost but no orders, past the 20-click threshold, get added as negative exact or negative phrase in the campaign that bought them.
3. Seal the cannibalizationWhen you graduate a query into exact, add that same term as a negative exact in its source auto or broad campaign so the cheaper campaign stops competing for a query you now own at a known price.
4. Respect the bufferNegative keywords take up to 72 hours to take effect and negative products take up to 96 hours, so a leak does not stop the instant you add the negative. Cut early.

For tactics that extend past Amazon's mechanics, including reusable negative lists, scripts, and automation, see our companion deep dive on negative keyword strategy for 2026.

CH.07Build the closed loop, cadence and structure

Everything above becomes durable performance only when it runs on a rhythm against a deliberate campaign architecture. The goal of the structure is simple: every query is bought once, in the campaign that owns it, at its cheapest proven price.

• Discovery auto: the four targeting groups, conservative bids, permanent research engine.
• Discovery broad: proven roots in broad match, widening the net with synonyms and related terms.
• Performance phrase: close variations of proven roots, moderate bids, controlled expansion.
• Performance exact: graduated winners, highest bids, protected by negatives in the discovery campaigns.
• Negatives wire it together: every graduated term is negated out of its source so the layers never cannibalize each other.

Run the harvest-and-negate cycle weekly or biweekly. A weekly rhythm beats a monthly deep clean because winning terms get their proper bid sooner and wasted spend gets cut faster, and the 72-hour negative buffer means delays compound if you wait. This is the same full-funnel discipline we apply across paid channels, including keeping Amazon and Google from competing for the same shopper, which we cover in running Google Shopping without Amazon stealing your clicks.

Discover with auto and broad, prove in the report, graduate into exact, and negate the source. Run that loop on a cadence and the account compounds instead of leaking. Capconvert Amazon Ads playbook