Ask how to choose keywords for advertising and most answers give selection criteria: relevant, has volume, converts well.
None of that is wrong, but those are SEO selection criteria. Advertising carries one more constraint: you also have to be able to win it.
A term can be relevant, high volume, and converting well for whoever runs it — and if several hundred sellers are already bidding while your conversion rate is average, entering it mostly contributes to that term's average price.
Excluding three kinds beats picking
Before mining for new terms, take these three out of the ones you already have.
One: terms where you already rank on page one organically.
Every term from the reverse lookup carries rankPosition and adPosition. When rankPosition.position is already strong and ads are still running, that spend buys a position you hold anyway.
It does not have to stop — defending a head term is sometimes right — but it should be a decision with a reason. Monitoring Amazon keyword rank gives a way to compute how many terms are in that state.
Two: terms marked I by conversionType.
The order keyword lookup sorts terms into four classes, and I is "ineffective impressions" — visibility with no conversion. In advertising that class is pure cost. How to optimize Amazon keywords covers what to do with each of the four.
Three: terms that look wrong when you read them. The data does not know what your product actually is. High volume, decent relevance score, and you can still see that the person searching wants something you do not sell. Trust that reading — it beats any field.
Four fields decide whether you can win
Among what the keyword miner returns, these four exist to answer the competition question:
| Field | Meaning | How to read it |
|---|---|---|
products / adProducts | Products / advertising competitors | Their ratio is how crowded the ad slots are |
monopolyClickRate | Click monopoly rate | High means a few ASINs take the clicks |
titleDensity | Title density | High means everyone has the word in their title |
supplyDemandRatio | Supply-demand ratio | Supply as a multiple of demand |
The first row is the most overlooked and the most useful.
The documented example values are products 1645 and adProducts 34. So that term has over sixteen hundred products under it and only 34 of them advertising — roughly 2% ad density.
That ratio is a different thing from "how many products are on this term." A term can have very many products, making organic placement hard, while few people advertise on it, leaving the ad slots relatively open. That is exactly the kind of term advertising is cheapest on, and looking only at search volume and product count will never surface it.
monopolyClickRate supplies the other side: even where few advertise, if a handful of ASINs monopolize the clicks, a new entrant still struggles for volume. Read the two together.
titleDensity tells you the organic difficulty. High title density with low adProducts is the classic "cannot get in organically, can buy in" — that term deserves ad spend rather than three months of title optimization.
One combination worth testing first
Put those together and one class of term deserves to go first:
Mid search volume (not a head term) +
adProductslow relative toproducts+ highpurchaseRate+ abidrange you can carry
The profile: demand is real (there are purchases), competition has not arrived yet (low ad density), and the price is manageable. These terms rarely appear on any "top keywords" list, because such lists sort by search volume and these sit in the middle.
bid, bidMin and bidMax give the price range. Whether you can carry it depends on your conversion rate rather than your budget — which Amazon PPC keyword bidding works through.
Filter server-side rather than exporting and sorting
Nearly every returned field has a matching min/max request parameter, so the judgments above can be written as query conditions:
| What you want | Conditions |
|---|---|
| Ad slots not yet crowded | maxAdProducts |
| Demand that is real | minPurchases or minPurchasesRate |
| Manageable price | maxBid |
| Not monopolized | maxMonopolyClickRate |
| Hard organically but worth buying | minTitleDensity with maxAdProducts |
| Long tail | minWordCount |
includeKeywords and excludeKeywords add a root-word filter, and filterRootWord collapses by root.
This beats exporting a few thousand terms and filtering in a spreadsheet, and it costs fewer calls — the filtering happens server-side, so what comes back is already the shortlist.
After the terms are live
Selection is the start. After a period, go back to the order lookup and watch conversionType migrate: I says the term was the wrong pick, E says it was right.
Choosing negatives is a separate job — see Choosing Amazon negative keywords.
Three things this data cannot do
It cannot give you your own ad reports. Actual spend, actual ACOS and the search term report are authorized data in Seller Central. Everything here is market-side distribution.
It cannot predict your results. It describes the term's competitive structure, not your performance. Two sellers with different conversion rates get different outcomes on the same term.
It is not real time. It returns statistics over a window, for setting selection strategy rather than intraday adjustment.
Questions
How many terms belong in one ad group? There is no single number, but fewer terms makes it far easier to tell which one is working. Dozens mixed into one group and you finish the run not knowing where the money went.
Is low adProducts always good?
Not always. Confirm purchases or purchaseRate is not zero first — few advertisers can also mean the term does not convert and others already found that out.
Should I bid on brand terms? Your own brand terms are usually cheap and convert well. Bidding on someone else's touches compliance boundaries; check the rules before deciding, and this post does not advise on that.
Are long-tail terms worth running?
Yes, particularly the ones maxAdProducts surfaces. Individually small, but low competition density and usually more precise conversion, which makes them good for testing which demand actually buys.