The Amazon Search Terms Report, and How to Read It

There is one report in an Amazon account that contains information no tool can supply: what real shoppers typed before they bought from you. Everything else is inference.

It is also the report most accounts glance at, sort by spend, and close. Two decisions come out of it every week, and both of them compound.

The search terms report shows the actual queries that triggered your ads, against the keywords or targets that matched them. The keyword report shows what you bid on. Those are different things and the gap between them is where the money goes.

A broad keyword can match hundreds of queries. Some of them are your product described differently, some are a competitor's product, and some are nothing to do with you at all. Only the search terms report tells you which.

Every decision below starts with a single row, and each row answers more than whether a term sold. Read its columns in the order the shopper moved through them.

The term and what caught it. The customer search term sits beside the keyword or target that matched it and the match type. A good term caught by the wrong keyword is a structure problem rather than a term problem: the keyword is reaching wider than it was meant to.

Impressions and clicks. Many impressions and almost no clicks means shoppers saw the ad and chose something else on the page. That is usually the main image, the price or the review count beside the competition, and no bid fixes it.

Clicks and orders. Clicks without orders means the shopper reached the listing and left. Either the term promised something the product is not, or the page did not answer the question the term asked. Read the term again before blaming the listing: "waterproof" on a product that is only water resistant will never convert, and should not.

Spend against sales. The ACoS on a row only means something against the product's break even ACoS, the share of the price left before advertising. A term at 35 percent loses money on a product that keeps 30 percent and makes money on one that keeps 45. ACoS, TACoS and ROAS sets out the arithmetic.

The window. The sales and order columns name how many days after the click they count. A row covering the last few days of the period is still filling in, and judging it now means judging an unfinished number.

Any search term that has converted more than once, at an acceptable cost, deserves its own exact match keyword with its own bid. That takes the term out of the auction lottery a broad match runs and gives it a bid set by what it is worth.

The same terms belong in the listing, because a query that converts in advertising is a query the listing should rank for organically. That is the loop between paid and organic, and it only runs if somebody carries the terms across.

Any search term with clicks and no sales, past the point where the absence is meaningful, becomes a negative. How many clicks is enough depends on your conversion rate: if you convert at one in ten, ten clicks with no sale is normal and forty is not.

Negatives are the cheapest work available in any account. They cost an afternoon and they stop paying for traffic that was never going to buy.

The mistake is negating too early on too little data, which removes terms that were converting at a normal rate for your category. That is covered in the article on negative keywords.

Open the report for an automatic campaign or a product targeting campaign and part of the customer search term column will not be words at all. It will be ten character codes beginning with B0. Those are ASINs: the ad was shown on another product's detail page, and the shopper clicked from there rather than from a search. Amazon puts the page where the ad appeared in the same column a query would have gone in, which is why the rows look like a mistake the first time anyone sees them.

They are some of the most useful rows in the report. An ASIN that converts is a competitor, or a complementary product, whose shoppers buy yours, and it deserves a product targeting campaign of its own with a bid set by what that page is worth. An ASIN that spends and never converts is usually a product cheaper than yours, or one with far more reviews, where your ad is the weaker offer on the page. That one becomes a negative product target, which stops the ad appearing on that page without touching anything else in the campaign.

In an automatic campaign the match type column says which of the four targeting groups found each row: close match, loose match, substitutes or complements. Substitutes and complements are where most of the product codes come from, and close match is where most of the converting words come from. If one group produces most of the waste, lower that group's bid rather than negating its rows one at a time, because the group will keep finding new rows faster than anyone can negate them.

Product research. Terms appearing repeatedly that describe something adjacent to what you sell are demand you are not serving, and they are demand recorded in your own account rather than in a keyword tool's estimate.

The same is true of the words shoppers use. A catalogue written in the brand's vocabulary and searched in the customer's is a catalogue with a translation problem, and this report is the dictionary.

Weekly on the campaigns carrying most of the spend, monthly on the rest. The cadence matters more than the depth: a report read every week catches a term burning budget after seven days rather than after a quarter.

Automatic campaigns deserve the closest reading, because their whole job is discovering terms you did not think of. An automatic campaign nobody mines is a discovery budget with nobody collecting the discoveries, which is the point of running automatic and manual together.

Illustrative figures from one week's report for one product, to show the decisions in the order they are made. The product sells at $40, keeps 35 percent of its price before advertising and converts at about one order in eight clicks, so its break even ACoS is 35 percent.

Row one: the product's own description of itself. 60 clicks, 9 orders, ACoS 18 percent, found by a broad keyword. It has converted more than once at a cost well inside break even, so it is promoted to exact match with its own bid.

Row two: a close variant. 14 clicks, 2 orders, ACoS 30 percent. Inside break even, on thin data. It stays where it is and is read again next week.

Row three: a material the product is not made of. 22 clicks and no orders. At one order in eight, about three were expected by now. It becomes a negative exact, and if the material turns up in other rows too, a negative phrase on the material.

Row four: a competitor's product code. 30 clicks, 1 order, ACoS 120 percent. The page it appeared on sells a cheaper product with far more reviews. It becomes a negative product target.

Row five: 4 clicks and no orders. Too early to judge. Negating it now is the mistake the article on negative keywords warns about.

Five rows: three changes and two deliberate waits, each written down with the date so next month's report can check it.

A decision is only half the job. The other half is putting it in the right place, and this is where most accounts leak. A term promoted to exact match in a manual campaign should also be added as a negative exact in the automatic or broad campaign that found it. Otherwise the two campaigns bid against each other for the same query, the report splits its data between them, and neither bid means anything.

Negatives go at the level that matches the problem. A single irrelevant query is a negative exact on the campaign that paid for it. A whole theme that is never yours, such as a material you do not sell, a size you do not make or a use the product cannot serve, is a negative phrase, and it can sit across every campaign at once. Sponsored Products takes negative exact and negative phrase only, so there is no broad negative to reach for, and a phrase negative on a single common word removes far more than intended. Read what it would have blocked before saving it.

Then read the report a week late. A sale is credited to the click for days afterwards, and the report's own column names say how many, so the last few days always look worse than they will. Negating a term on Monday for having no sales since Friday is negating on data that is not finished. And download the report every time: the console keeps a limited history, and a year of search terms is the only record you will have of how the language of your category changed.

Volume. It reports what your ads matched, not what the market searched, so a term absent from it may be a term nobody searched or one your bids never reached.

And it withholds some queries, especially rare ones, so the rows never quite add up to the totals. That is Google and Amazon both protecting individual searches, and it is worth knowing before somebody spends an afternoon reconciling the difference.

Small accounts and long tail categories meet a different problem: hundreds of rows, each with two or three clicks, none of them enough to judge on its own. Waiting for any single term to reach forty clicks can take a year.

Read words, not terms. Split every search term into its single words and add up the clicks, spend and orders for each word across the whole report. A word that appears in sixty different terms with ninety clicks between them and no orders is a clear signal, even though no term containing it had more than three clicks.

Then pairs of words. The same sum for two word phrases catches what single words miss. "For kids" may be wasting money on an adult product even where "kids" alone appears in terms that convert.

Widen the window. Read the last two or three months together rather than the last week, from reports downloaded each time so the history is yours to add up. The decisions come more slowly, which suits a budget too small to waste.

Act on the theme, not the row. A wasted word becomes a negative phrase, read first against everything it would block. A word that converts across many terms is a theme worth an exact match keyword of its own, and a place in the listing.

Keep the threshold honest. The rule from decision two applies to the word as it does to a term: converting at one in ten, a word needs well over ten clicks without a sale before its absence means anything.

Grouped this way, a report too thin to read row by row still produces the same two decisions every week.

Reading it is part of the weekly routine in Amazon account management, and the terms it produces go in two directions: into the campaigns as keywords and negatives, and into the listings as the words shoppers actually use.

The equivalent report exists on Walmart with thinner data, and on Google Ads where the same discipline applies to a much larger match surface.

If nobody has read yours in a month, that is usually where the first savings are.