Your automatic campaign looked like a winner for three weeks. Then the spend line separated from the sales line, and now you're scrolling a search-term report full of queries you'd never have bid on, unsure which clicks were buyers and which were browsers.
That report is not a list of keywords waiting to be copied into a manual campaign. It's evidence, and it's the most useful thing your ad account produces, because it records the exact words people typed right before they decided whether to buy.
Amazon PPC keyword research is the practice of turning that evidence into bids, negatives, and campaign structure, then running the loop again next week. Run it once at launch and you'll spend the rest of the year paying for traffic that was never going to convert.
Research Stages and the Data Behind Each
Amazon's Sponsored Products targeting guide presents targeting as a process of testing and refinement. Broad match allows wider discovery, while phrase and exact match give you progressively tighter control over relevance and spend. Automatic targeting can surface shopper queries; manual targeting lets you choose which keywords to fund directly.
Use each report for a different decision. Review shopper searches that generated clicks or sales, promote relevant terms into manual campaigns, and add negatives for queries that spend without producing useful results. The search-term report shows only recent history, so export it regularly if you need to compare results over time.

| Stage | Data source | Output | Decision it drives |
|---|---|---|---|
| Seed generation | Search autosuggest, competitor ASIN listings, your own and competitor reviews | Starter list grouped by product attribute | Which terms enter the first campaign |
| Launch | Automatic targeting campaigns | Shopper queries that matched your products | Which queries to promote to manual targeting |
| Review | Search term report | Queries with at least one click, plus orders and spend | Promote winning queries and add negatives for waste |
| Trend check | Targeting report | Keyword and match-type performance over time | Whether a match type still earns its spend |
| Demand gap | Brand Analytics search query performance for brand-registered sellers | Search demand and click share for queries | High-demand terms you are not yet targeting |
One gap remains: searches that never earn an ad click do not appear in the search-term report. Brand-registered sellers can use Amazon Brand Analytics search query data to examine demand and click share, including queries they are not targeting. That evidence can reveal opportunities, but it does not establish that a query will be profitable for your product.
If you also sell through Shopify, the same habit of recording evidence and making decisions from it can support your SEO work. This guide explains how to build a measurable Shopify SEO plan.
Export the reports for your latest data window before changing bids. Tag each keyword by its source and stage, so you can distinguish terms you selected from queries Amazon surfaced and judge each on the evidence available.
How to Harvest Seed Keywords From Listings and Autosuggest
A seed list is not a keyword list. It's the raw vocabulary you test, and its job is to be broad enough to cover how buyers describe the problem and specific enough that each term can be traced back to something real on your listing or a competitor's.

Work these six steps in order and you'll have a defensible seed sheet in an afternoon.
Name the core noun plus three attributes. Take material, use case, and buyer, and put them in one phrase: "insulated stainless water bottle 32 oz for hiking." Keep phrasing tight because every variation you create is a term you eventually have to fund and judge, and a bloated seed list produces too little data per term to conclude anything.
Type each seed into Amazon's search bar and record every completion. The bar returns real query strings ranked by frequency, not modeled estimates, and the modifiers it volunteers do the useful work: "for", "with", "without", "kids", "gift". Those little words are intent markers, so copy them exactly as they appear.
Open the top ten competitor listings in your category and harvest title phrases, bullet wording, and repeated Q&A questions. Competitor titles are already ranked for something, and the questions shoppers ask reveal the attributes they check before buying. Pair this with a structured competitor keyword analysis so overlap between several listings gets treated as a stronger signal than one seller's phrasing.
Read 30 to 50 reviews across your product and two or three competitors. Note the words buyers use for the problem they're solving, not the words they use for the product. Someone buying a bottle for a hospital shift writes about ice lasting through a twelve-hour rotation, and that phrasing never appears in a spec sheet.
Check your backend search terms last, and only for gaps. The field is a place to catch synonyms and misspellings that read awkwardly in visible copy, not a dumping ground for every variation you generated. Repeating a term you already use in the title wastes scarce characters and gains nothing.
Sort everything into one sheet with four columns: term, product attribute, source, and type. Mark type as generic, brand, or competitor. This single column prevents the most expensive mistake in the list, which is letting competitor brand names ride along inside your main campaign budget.
That last point deserves its own test group. Competitor brand queries convert at a different rate and a different cost than generic ones, and blending them into the same ad group makes both sets of numbers unreadable.
Sign of a healthy seed sheet: four to six product groups, no empty buckets, and you can name the source of every single term without checking.
A thin sheet shows up as three groups where one term carries all the weight, or a group built entirely from autosuggest with no review language behind it. Fix that before you spend, not after.
Map Search Terms to Intent Before You Bid
Long-tail keywords are supposed to be the cheap, conversion-ready end of the market. Account-level evidence says otherwise. A 2026 analysis of 58 Amazon ad accounts across 44 brands and nine marketplaces found median conversion rates of 8.4% for one- to two-word searches, 8.0% for three- to four-word searches, and 7.3% for searches of five or more words, with the longest terms also carrying the highest median ACoS at 37.4%, according to Olifant Digital's long-tail keyword analysis.
The same analysis found the split isn't clean: among 44 accounts with enough data in both groups, longer terms converted worse in 23 and did not convert worse in 21.
What separates the winners isn't word count. It's whether the query names a product attribute, a use case, a compatibility requirement, a brand preference, or a price constraint your listing satisfies.

Sort every search term into one of six classes and the bidding decision mostly makes itself.
| Intent class | Example query | Match type | Action |
|---|---|---|---|
| Problem-aware | best way to keep coffee hot all morning | Broad or phrase | Test with a modest bid and check the listing answers the problem |
| Product-specific | insulated 32 oz water bottle | Phrase | Core target; monitor ACoS weekly |
| Brand or model | competitor brand stainless bottle | Exact, separate campaign | Expect lower conversion; keep the bid limited |
| Attribute-constrained | BPA free water bottle with straw | Exact | High intent; move winners to exact |
| Price or bargain | cheap water bottle bulk | Phrase | Negative if your margin cannot survive the price |
| Mismatch | water bottle holder for car | Negative exact | Block; the use case does not match your product |
Two rows carry most of the wasted spend. Price queries pull shoppers comparing against a cheaper substitute, and no bid fixes a margin problem. Mismatch queries look relevant to a keyword tool and obvious to a human, which is why the review step stays manual.
Practical rule: judge a query by what it commits the buyer to, not by how many words it contains.
Brand terms need their own campaign because they behave differently in both directions. A shopper searching your brand name is close to buying, but the click competes with your organic listing for the same sale. A shopper searching a competitor's brand name is researching, and the conversion rate you see there rarely justifies the bid it takes to hold position. Keeping these apart from generic targeting lets you read branded versus non-branded keyword performance separately instead of averaging a strong number with a weak one.
Localization breaks naive word-count logic entirely. A four-word query in one marketplace can be less specific than a two-word query in another, because word order, language, and shopping behavior all shift. Keep a separate intent taxonomy per country rather than translating one list.
If you want a second opinion on the classification step itself, OGTool's keyword selection framework is a useful sanity check on which classes deserve budget first.
For each row, run one test before you bid: does your listing deliver the attribute the shopper named, and do your reviews use that same language back? If the attribute only exists in your ad copy, exact match will convert badly and you'll blame the keyword.
Rank Keywords by Profit Instead of Volume
Set the ceiling before you rank anything. Your break-even ACoS is the profit per unit left after Amazon fees, shipping, returns, and cost of goods, divided by your selling price. That number decides which keywords are viable, and it has nothing to do with how many people search a term.
A $30 item that leaves $9 after all costs has a break-even ACoS of 30%. Any keyword running above 30% ACoS on that item loses money on every sale it produces, no matter how good the search volume looks in a tool.
High volume fails as a ranking signal because volume measures how many people are looking, not how many of them want what you sell. A broad head term with thousands of monthly searches and a 2% fit will drain a budget faster than a mid-volume phrase that describes your product exactly.
Here are starting thresholds. Adjust them to your margins and category, then keep them fixed long enough to compare periods.
| Metric | Signal | Action |
|---|---|---|
| Clicks | Fewer than 10 clicks | Keep collecting data; do not judge yet |
| Clicks and orders | 10+ clicks, zero orders | Lower the bid or add a negative |
| ACoS | Below target ACoS with steady orders | Raise the bid and test exact match |
| ACoS | Above break-even ACoS for your margin | Cut the bid or pause the keyword |
| Conversion rate | Above the category median | Move to its own campaign |
| Search volume | High volume, low conversion | Test at a small bid; do not scale yet |
The 10-click floor matters more than sellers expect. Three clicks and no sale is noise, and negating on that basis removes terms that would have converted with another week of data.
A keyword that loses money on every sale isn't a keyword problem. It's a bid you should not be paying.
Keep the ranking in one sheet, one row per keyword, with columns for clicks, orders, ACoS, unit margin, and intent class. That last column is what stops you from cutting a term for a reason that only applies to a different class of buyer.
When the margin math gets murky, work it out properly with a guide to how to calculate profit margins before you set the ACoS ceiling. Getting the ceiling wrong makes every downstream decision wrong in the same direction.
Rerun the ranking after each export. Keywords move between rows as data accumulates, and a term you paused at 12 clicks may earn a second test later in the season. The sheet is the audit trail, so keep the old versions.
Automation Tips and Blind Spots to Check
A bid rule can cut wasted spend quickly, but a rule acting on incomplete data can also shut off a keyword before its value is clear. Use automation to sort and flag terms at scale, then set limits on what it can change.
Set a maximum bid ceiling tied to your margin, a minimum click threshold before a rule acts, and a pause condition for keywords above break-even ACoS. Schedule search-term reviews consistently, and route proposed changes through a review sheet before applying them. Automated bid management can help optimize ad spend and identify high-performing keywords, but the thresholds still need to reflect your margins and campaign goals.
The main blind spot is what the reports leave out. Amazon's Search Term Report playbook describes a report of customer searches that generated at least one click. Searches with no clicks are absent. A tool that relies only on this report cannot show whether an unreported term has no demand or has not received exposure.
Automated targeting can also use signals beyond a direct keyword match. That makes it harder to connect an attributed sale to one query, so treat the report as evidence of observed activity, not a complete map of customer intent.
Before enabling automation, check three things:
- Inputs: Confirm which reports the tool uses and how often they refresh. Search-term data alone leaves unclicked searches out of view.
- Outputs: Check whether the tool can edit bids, budgets, negatives, or campaigns. Require review for changes you need to explain or reverse.
- Incrementality: Hold back a portion of spend and compare total orders. More attributed sales can reflect a shift in credit from organic or another paid placement rather than additional sales.
Language variants add another source of uncertainty. A keyword classification that fits one marketplace may misread intent in another, so review country-specific terms rather than relying on direct translation. Automation can surface patterns, but it cannot fill gaps in the underlying reports or prove that a sale was incremental.
Move Winning Terms Into Campaigns and Run Your First Audit
A query that converts in a discovery campaign deserves a controlled test in exact match. Give it a separate bid and budget, then check whether performance holds without letting discovery traffic compete for the same query at a different bid.
The weekly harvest routine
Export the search-term report each week and keep the files. It shows queries that received at least one click, while older rows eventually fall outside the reporting window. Saving exports preserves a record for comparisons.
Filter for queries with at least one order, then sort by orders and ACoS. Sorting by clicks can put a query with many clicks and no sales above a smaller, more useful signal. Check orders alongside cost before deciding what to promote.
Add promising queries as exact-match keywords in a manual campaign. Separate control over bids and budgets makes it easier to protect a converting term from being underfunded in discovery. For a further explanation of keyword targeting for Amazon ads, see the match-type mechanics.
Add irrelevant or unprofitable queries as negatives at the appropriate match type. Use negative exact for a specific query. Use negative phrase when a word or phrase is producing a broader set of poor matches.
Change bids in small steps and allow enough clicks to assess the result. A keyword needs roughly 10 clicks before a bid change tells you much. Daily adjustments can make it difficult to distinguish a real performance shift from normal variation.
Keep the automatic campaign running as a discovery source, and review it weekly. Turning it off stops new queries from feeding the exact-match layer. Review search terms and add negatives rather than leaving the campaign unchecked.

What the long-run evidence says about patience
Amazon reports that sellers whose Sponsored Products campaigns ran continuously for 12 months achieved, on average, an 11.2% higher return on ad spend than their ROAS during the first month, based on worldwide internal data collected from June 1, 2019 through July 30, 2021, as described in Amazon's keyword strategy guidance. Treat that as a benchmark, not a forecast for your product. Results depend on the queries you discover, the bids you set, and how consistently you act on the evidence.
Open the targeting report to compare keyword and match-type performance across a reporting window long enough to see whether a seasonal term recurs. Then open the search-term report to identify customer queries that might merit new targets or negatives. These reports do not show every reason a sale happened, so read them alongside orders, spend, and the timing of changes. Pausing or restructuring campaigns too quickly can leave too little evidence to tell a temporary dip from a poor target.
Here's how that audit loop looks in practice.
This week's first audit
Write down your break-even ACoS. Choose a best-selling product with several weeks of ad data, rather than a new launch, and work through the six steps using its exports. For each harvested keyword, record the customer query that produced it so you can compare the next export against the original signal.
Schedule the next export before closing the report. A consistent comparison makes it easier to see whether a keyword still earns its place or whether new search evidence calls for a change.
Next Point Digital supports marketplace brands with product listing optimization, marketplace SEO, and predictive bid management. If your search-term exports are piling up unused, Next Point Digital can help build a campaign structure around them through listing optimization, marketplace SEO, and predictive bid management.