Every Amazon account has a report that lists precisely what shoppers typed before they clicked. Almost every account we take over has been reading it occasionally, acting on half of it, and calling that optimisation.
Harvesting is not a research exercise you do before a launch. It is a loop, run on a schedule, in two directions at once.
The two directions
Promotion. A search term inside a broad or auto campaign converts repeatedly. It is currently being bid on indirectly — Amazon decides what to pay for it based on the keyword that matched it. Promoting it means creating an exact-match target for that term, where you set the bid, watch it alone, and stop it competing with everything else in the ad group.
Negation. A search term has spent real money across several clicks and produced nothing. Left alone it will do the same next week and the week after, because nothing in the system removes it. Adding it as a negative keyword is the only thing that stops it.
Accounts that only promote look busy and get more expensive. The negation half is where the ACOS movement actually comes from, and it is the half that gets skipped because it feels like subtraction rather than work.
When a term has earned promotion
Not on one sale. A single conversion tells you almost nothing, and promoting on it fills the account with exact-match targets that never repeat.
The threshold we use is two or more conversions at an ACOS at or below the product’s break-even, over a window long enough to be more than noise — usually 30 days. That is deliberately conservative. A term that clears it is a term you can bid on with intent rather than hope.
Once promoted, the term needs negating in the campaign it came from. This is the step most often missed, and it is the reason so many accounts have the same term competing against itself in three places, bidding up its own click price.
When a term has earned negation
The instinct is to wait for a round number of clicks. Ten clicks, no sale, negate.
That rule is wrong in both directions. On a $4 product, ten clicks at $1.20 is wildly past the point of loss. On a $200 product, ten clicks may be a perfectly normal path to a sale.
Judge it against contribution margin, which is the same number that sets your break-even ACOS. If a term has spent more than the margin on one unit and returned nothing, it has failed at the only job it had. Whether that took six clicks or forty is not the interesting part.
Two qualifications worth holding:
- Check the term before you kill it. A term with strong impressions, poor CTR and no sales is often a relevance problem in the listing, not a bad term. Negating it hides a fixable listing issue.
- Seasonal terms recover. A gift term that failed in July has not failed. It was out of season.
Phrase versus exact, and why it matters
Negative exact blocks one term. Use it when one specific search failed and its neighbours are fine.
Negative phrase blocks a family. Use it when the pattern itself is wrong for you — a competitor brand name, a size or format you do not sell, a use case your product genuinely does not serve. One well-chosen negative phrase does the work of forty negative exacts and keeps the list readable.
The failure mode is a negative phrase applied too broadly, quietly suppressing terms you wanted. Anything blocking a family deserves a second look before it goes in, and a note of why.
Search term overlap between campaigns
Harvesting creates a second problem if it is not policed: the same search term ends up targeted in two places. An exact-match campaign and the broad or auto campaign it was harvested from both bid on it, and Amazon picks which one serves.
You can see it in the search term report. Filter to one search term and check whether it appears under more than one campaign or ad group. If it does, those two are competing with each other, splitting the data you need to judge the term and often serving from the wrong bid.
The fix is the step below that people skip: negate the promoted term, as a negative exact, in every campaign it came from. Then run the overlap check once a month to catch what slipped through. Overlap is one of the first things an Amazon account audit looks for, because it quietly inflates reported ACOS.
Running it so it survives
The loop only works if it is boringly consistent, which means it has to be small enough to do every week:
- Pull the search term report for the last 7 days, plus a 30-day view for judging promotion.
- Sort by spend, descending. The top of that list is where the money is, and it is where the report is worth your attention.
- Negate the failures — spend above unit margin, zero conversions, no seasonal or listing explanation.
- Promote the proven — two or more conversions at or below break-even, over 30 days. New exact-match campaigns go into the same Amazon PPC portfolio as the product they sell, so budget and reporting stay grouped by product line.
- Negate promoted terms in their source campaign, every time, without exception.
- Log what you changed and why, so next week’s decisions are made against a record rather than a memory.
Step six sounds like paperwork. It is the step that makes the difference between an account that improves for six months and one that improves for three and then drifts, because whoever picks it up next can see the reasoning rather than guessing at it.
This loop is the core of the weekly cycle in our Amazon PPC management, and it is the single routine most consistently absent from the accounts we inherit.
