GUIDEUPDATED NOVEMBER 20, 2026

Your whole Amazon keyword universe, and the searches worth paying for

Amazon PPC keyword discovery is two jobs: map every search shoppers use in your markets, then pay only for the clicks that leave a profit. The map is larger and moves faster than most accounts allow for. Across 25 US brands on our platform, the median store’s Amazon ads got clicks on about 4,000 different searches a month, two thirds of the non-branded ones had no click in the two months before, and the campaigns that find new searches (phrase, broad and automatic) brought half of non-branded orders.

Laurence10 min read

What is keyword discovery on Amazon?

It is finding every search a shopper might type to find a product like yours, your keyword universe, and then the searches in it worth paying for. The universe holds searches for your brand, the generic searches in each category you sell in, and searches for competitors. It runs to thousands of searches, most of them small: the median store on our platform got ad clicks on about 12,800 different searches in 90 days.

Two words keep it in order. A market is an Amazon category your products sell in, one node of Amazon’s browse tree. A niche is a cluster of searches inside a market that share one shopper intent, such as “gps dog collar” and “dog tracker collar”. Your keyword universe is the niches of all your markets, each with its searches, plus your branded and competitor searches in groups of their own.

How do I size my Amazon market before choosing keywords?

Size each market your products sell in by its monthly revenue, then split it into niches and size each niche by its searches. The markets together are your total addressable market (TAM) on Amazon; the niches show where in it your products can win and what a click there is worth.

  • List your markets. Each product line sells in one category; note it, and the categories shoppers compare it with.
  • Size each market. Its monthly revenue, and how that splits between the leading brands, shows what is at stake and how hard it is to take share.
  • Collect its searches. Search Query Performance and Top Search Terms in Brand Analytics, your search term report, and a keyword tool’s estimates for searches your ads have not reached yet (what each source gives).
  • Group them into niches by what the shopper wants: a kind of product, never a size, flavour, pack count, price word or brand. Searches that name a brand go in their own groups, yours apart from competitors’.
  • Size each niche by its searches’ monthly volume, read as a range.
  • Mark where you already stand. Your organic rank and ad rank on each niche’s main searches show where ads would add sales, and where they may pay for sales your listing already wins.

Read every volume as a range. Amazon’s own counts come only from Search Query Performance, which lists just each product’s own top searches; keyword tools put searches in nearly the right order but often get their size wrong. Use them to rank niches against each other, not to forecast orders.

How many keywords should an Amazon PPC account target?

Every relevant search that can be bid at a profit, which is far more than a list of top keywords. At the median store on our platform, the 50 best-selling non-branded searches brought only half of non-branded orders; the other half came from thousands of smaller searches.

NON-BRANDED ORDERS BY SEARCH, RANKED BY ORDERS, MEDIAN STORE
SEARCHESSHARE OF NON-BRANDED ORDERS
The top 5051%
The top 50077%
Every search after the 500th (about 10,600)23%

25 US brands, Sponsored Products, August 16 to November 13, 2026. A search’s rank is set by its orders over the 90 days. Each share is the median across stores.

  • Half from the top 50. The 50 searches with the most orders brought 51% of non-branded orders at the median store, between 41% and 64% for the middle half of stores, and less than half in 12 of 25.
  • About 4,000 searches a month. The median store’s ads got clicks on 3,942 different searches in the last 30 days, branded ones included, between 1,400 and 6,100 for the middle half of stores.
  • Too many to bid one by one. Most of these searches get too few clicks to judge on their own (are low-volume keywords worth it). Discovery campaigns reach them as a group; exact keywords price the ones that prove themselves.

Which Amazon campaigns find new keywords?

Phrase, broad and automatic targeting: they show your ads on searches you did not list, and the search term report shows which of them sold. In our accounts they brought half of non-branded orders, at a higher cost per order than exact keywords, which is the cost of finding the searches that exact keywords then bid on one by one.

NON-BRANDED ORDERS BY THE TARGETING THAT MATCHED THE SEARCH, MEDIAN STORE
TARGETINGSHARE OF ORDERSSHARE OF SPENDCVRCOST PER ORDER
Exact keywords40.7%40.6%14.3%$13.25
Phrase and broad keywords23.2%24.8%11.1%$20.32
Automatic targeting22.0%16.9%8.9%$14.64
Product and category targeting4.4%5.4%13.0%$18.07
Phrase, broad and automatic together52.5%51.7%10.6%$16.48

25 US brands, Sponsored Products, August 16 to November 13, 2026: 320,000 non-branded searches with at least one click, 1.5M clicks, 268,000 orders and $3.1M of spend. Product page rows and the searches Amazon reports only as “(Other)” are left out. Each figure is the median across stores, so a column need not sum to 100%; the last row is read store by store, not added up from the rows above it.

  • Half the orders came from searches nobody listed. Phrase, broad and automatic targeting brought 52.5% of non-branded orders at the median store, between 28% and 71% for the middle half of stores, and 30% or more in 17 of 25.
  • At a higher cost per order. Their clicks converted at 10.6% against 14.3% for exact keywords, and an order cost $16.48 against $13.25. Exact keywords are usually searches that have already proven themselves; discovery campaigns pay to find them.
  • Automatic targeting found orders cheaply. It converted least, at 8.9%, but its clicks cost less, so an order cost $14.64, nearer exact keywords’ $13.25 than phrase and broad keywords’ $20.32.

Keep each kind in its own campaign, so a proven search is bought at its own bid and not at the bid of the broader campaign that found it (exact, phrase or broad).

How often do new search terms appear on Amazon?

Constantly. In our accounts, two thirds of the non-branded searches that got a click in a month had no click in the two months before, and those searches brought about a fifth of that month’s non-branded orders. Keyword discovery is never finished: keep discovery campaigns running after launch.

  • Two thirds new. Of the non-branded searches with a click in the last 30 days of the window, 66% had none in the 60 days before, between 64% and 69% for the middle half of stores.
  • A fifth of orders. Those searches brought 18.5% of those 30 days’ non-branded orders at the median store, between 15% and 24% for the middle half, and 10% or more in 19 of 23 stores.
  • Mostly small. They were two thirds of the searches and a fifth of the orders, so each one sells little; together they are too large to leave out.

How do I choose which discovered keywords to keep?

Keep a search when it fits the product and its clicks are worth more than they cost. A click is worth its CVR times the contribution margin of an order, so judge each search on enough clicks to read its CVR, lower its bid before you exclude it, and exclude at once only the searches that do not fit the product.

  1. 01
    Relevance firstWould a shopper typing this be satisfied by this product? If not, add it as a negative keyword now; that needs no data.
  2. 02
    Value per clickCVR × contribution margin per order, against the CPC. A search with no history borrows a CVR from similar searches for the same product (what a click is worth).
  3. 03
    Enough clicksDecide on clicks, not days, with the number set from your break-even CVR before the test starts (how many clicks before pausing).
  4. 04
    A price, not a verdictA relevant search that converts below target usually needs a lower bid, not a negative keyword: at a lower price it can still earn.
  5. 05
    Its own bid once provenGive a proven search its own exact keyword, and add it as a negative exact keyword where it was found, so the two campaigns do not bid against each other (moving search terms).

Why do my best Amazon keywords spend so much on product pages?

Because one keyword bid buys every placement, and product-page clicks convert at about half the rate of top-of-search clicks. On the 50 biggest non-branded searches of the median store on our platform, product pages took 31% of the spend while converting at 9.4% against 20.1% at the top of search. Bid each placement on its own CVR, with a lower base bid and a top-of-search adjustment (how to bid each placement).

PLACEMENTS ON EACH STORE’S 50 BIGGEST NON-BRANDED SEARCHES, MEDIAN STORE
PLACEMENTSHARE OF SPENDCVR
Top of search44.7%20.1%
Rest of search20.7%12.6%
Product pages30.6%9.4%

24 of the 25 US brands: those with 200 or more clicks in each placement on their 50 biggest non-branded searches by spend, Sponsored Products, August 16 to November 13, 2026. Each figure is the median across stores.

  • Half the CVR on product pages. Top-of-search clicks on these searches converted 1.95 times as often as product-page clicks at the median store, and more often in 22 of 24 stores.
  • A dearer order. An order from a product-page click cost 1.36 times one from the top of search at the median store, and more in 20 of 24 stores.
  • Your biggest searches carry the most waste. They take the largest share of spend, so a placement priced above what its clicks are worth costs most there.

How Laurence finds and selects your keywords

Laurence maps your markets before it spends, launches a campaign for each niche with discovery campaigns around them, and prices every search it reaches by what its clicks are worth to you. More on how Laurence maps your markets.

Scoping your markets

  • Found from your catalog: Laurence reads the Amazon categories your products are listed in and keeps the ones that together bring nearly all of your sales, then sizes each by its monthly revenue and how that splits between brands.
  • Every search, grouped into niches: each category’s searches, from our market research, are grouped by what the shopper wants, with searches for your brand, for competitors and for product IDs kept apart.
  • Sized from Amazon’s own counts: on the platform’s Markets page, each niche shows its searches and purchases a month from Search Query Performance, and your organic rank and ad rank on its searches from our own check of page one.

Exploring for new searches

  • A launch built on what already sells: each product’s keyword set starts from the searches its own ads, and those of its sibling products, have already sold on, grouped by intent. Your Deployment Strategist reviews every set before it goes live.
  • A campaign per niche, plus discovery: each non-branded niche gets its own exact campaign, beside a phrase campaign, an automatic campaign and product targeting; branded searches and competitors’ names run in campaigns of their own. Broad match is off unless your strategist turns it on.
  • Negatives from one rule: the phrase campaign excludes the searches the exact campaigns bid on, and a brand fence keeps your branded searches out of the non-branded campaigns.
  • Proven searches promoted by a person: searches that sold through the discovery campaigns but have no exact keyword yet are listed for your strategist, ranked by orders against your profit target, and the proven ones become exact keywords. No keyword goes live without a person’s approval.

Paying only for profitable clicks

  • Every search priced every hour: each target’s bid is the CVR Laurence expects times what an order is worth to you, repriced every hour from Amazon’s hourly data (how Laurence sets bids).
  • Careful with thin data: a search with little history borrows from similar searches for the same product and is bid from the cautious end of its range until its own clicks build up.
  • No fixed click count to pause: each click without an order lowers the CVR Laurence expects from that search, and its bid follows.
  • Each placement on its own CVR: Sponsored Products placement adjustments are recalculated nightly from each campaign’s own results by placement, so what a keyword pays at the top of search and on product pages follows how each placement converts.

Laurence maps every search in your markets and bids each one, in each placement, at what its clicks are worth to you.

How we measured this. Sponsored Products search term reports (Amazon Ads API) for 25 US stores on the Laurence platform, August 16 to November 13, 2026, at Amazon’s default attribution window. Searches for a store’s own brand or its own products are branded; every other search, competitors’ brands included, is non-branded. Rows whose search term is a product page (an ASIN) or Amazon’s “(Other)” are left out. The targeting that matched a search is the report’s match type: exact, phrase, broad, automatic (close match, loose match, substitutes, complements), or product and category targeting. A search is “new” in the last 30 days when it had no click in the 60 days before. Each figure is computed per store first, then summarized as the median across stores, or its quartiles; no store is named.

More guides

Tomorrow's brandsgrow differently today.

Backed by

  • Y Combinator
  • Susa Ventures
  • Box Group
  • Jump Trading
  • SV Angel
  • & angel investors from

    OpenAIAmazon