The Amazon Search Terms That Cost Us Money Weren't Zero-Sale Ones

✨ Written with AI assistance from our own advertising and settlement data, and reviewed by Avanta Global EOOD before publication. Learn more

Short answer: in 91 days of our own Amazon ads, 543 search terms produced no orders and took 44% of our ad spend. Not one of them had enough clicks to prove it was bad. Terms with no orders after a few clicks later converted as well as our winners. The money we could actually recover sat in a different place: 11 terms that did sell, but above break-even ACoS, starting with a single term at 73% ACoS.

What we measured

One product on Amazon Australia, Sponsored Products only, from 1 July to 29 September 2026. The search-term report was checked against the campaign report for the same months, and the two agree within 1%.

Search terms that received clicks	657
Ad spend	A$2,477
Clicks · orders	1,848 · 241
Conversion rate (orders ÷ clicks)	13.0%
Average cost per click	A$1.34
ACoS	24.4%

To turn ad sales into profit we used the contribution per unit measured from our settlement for the same period. After referral fee, FBA fee and product cost, but before ads, it came to A$10.46. At full price our break-even ACoS is about 33% (how that is calculated).

Where the spend went

Group	Terms	Spend	Units sold	Profit after ad spend
Sold, ACoS at or under 33%	103	A$1,006	279	+A$1,912
Sold, ACoS over 33%	11	A$388	21	−A$169
No orders	543	A$1,083	0	−A$1,083

On paper the obvious move is to add the 543 zero-order terms as negatives. Their spend is almost half the total and returned nothing.

Why the zero-order terms could not be judged

None of the 543 terms reached 20 clicks. 538 had between 1 and 9 clicks; five had 10 to 19.

At our 13% conversion rate, a perfectly good term has a 28% chance of zero orders after 9 clicks, and a 50% chance after 5. To be 90% sure that zero orders means a bad term, it needs about 17 clicks. Almost none of our terms get there.

We tested it on our own data instead of trusting the arithmetic. We split the 91 days in two (1 Jul – 15 Aug, 16 Aug – 29 Sep) and followed the terms that had zero orders in the first half:

Terms in the first half	Conversion in the second half
Zero orders on 1–4 clicks (43 terms)	13.6% (16 orders on 118 clicks)
Had orders (31 terms)	14.4%
Zero orders on 5+ clicks (8 terms)	6.8% (4 orders on 59 clicks)

The terms that looked like waste after a handful of clicks converted almost exactly like the proven ones. Cutting them would have cut sales. Only the group with five or more empty clicks was clearly weaker, and even that group sold.

The tail also regenerates. In the second half, 313 terms appeared that had not existed in the first. Together they ran at 33.5% ACoS, roughly break-even. That is the cost of discovery: the long tail pays for itself, about, while it finds the next terms worth keeping. You cannot add future search terms as negatives.

Where the money was actually lost

The terms that measurably lost money were the ones that did sell:

Term	Clicks	Orders	ACoS
postage bags	93	5	73%
padded mailer bags	46	3	53%
two competitor product pages (ASIN targets)	26 + 27	2 + 2	56% · 59%
package bags	17	1	77%

These have enough clicks to judge and a real conversion rate, and they convert too slowly for their cost. "postage bags" alone spent A$121 for five units. At A$10.46 contribution each, that is about −A$68 on one term.

Negating a term that sells is rarely the fix. The fix is price: bid it down until its ACoS sits under break-even, or move it to exact match with its own bid, and let the data say whether it holds.

How to find the terms that cost you money

1. Download the search-term report for at least 60–90 days, and check that its spend matches the campaign report. If it does not, you are missing rows.
2. Work out your conversion rate (orders ÷ clicks) and the clicks a term needs before zero orders means something. At 13%, that is about 17; at 20%, about 11.
3. Sort terms that have enough clicks by ACoS against break-even, not by ACoS against a target. Break-even comes from your price and your fees, and it moves with them (our fee change moved ours by seven points).
4. Bid down terms that sell above break-even. Add negatives only for terms that have enough clicks and are clearly irrelevant.
5. Leave the thin tail alone, but watch it as a group. If it runs well above break-even as a whole, lower the bids on the campaign that feeds it rather than chasing terms one by one.

The break-even ACoS calculator does step 3's arithmetic, and the free PPC audit runs these checks on your own report. AgentXray does the same from your connected ads and settlement data. Each bid or negative it suggests is a draft, priced against that SKU's break-even, and you approve it before it reaches Amazon.

FAQ

Should I add every search term with no sales as a negative keyword? Not on a few clicks. On our data, terms with zero orders after 1–4 clicks later converted at 13.6%, the same as our winners. Wait until a term has the clicks your conversion rate requires.

How many clicks before a search term is a loser? It depends on your conversion rate. The chance of zero orders by luck is (1 − conversion rate) raised to the number of clicks. Keep that under about 10% before you act.

Is wasted ad spend the same as zero-sale spend? No. Some zero-sale spend is the cost of finding new terms. Terms that sell at a loss are harder to spot and easier to fix.

About this article

Figures come from Avanta Global EOOD's own Amazon.com.au Sponsored Products search-term reports for one SKU, 1 July – 29 September 2026 (657 search terms, 1,848 clicks), cross-checked against the campaign report. Contribution per unit (A$10.46) comes from our settlement line items for the same period, with product cost at A$7.40. "Units sold" and "orders" use Amazon's advertising attribution. Measured on 30 September 2026. Currency is Australian dollars.