Every other lever in an Amazon account trades one thing for another. Raise a bid and you buy placement with margin. Lower it and you save money by appearing less. Negative keywords are the only lever that removes spend without removing a single sale, and most accounts use a fraction of what they should.
The reason is simple: adding negatives is reading work rather than adjusting work, and reading is easier to postpone.
It tells Amazon not to show your product for a term, regardless of what your keywords say. It is the counterweight to broad match and automatic campaigns, which are deliberately loose so they can discover demand you had not thought of.
Loose matching without negatives is not discovery. It is a budget with no floor.
Open the search term report for the last thirty days, sort by spend, and read the top fifty. Each term goes into one of three piles.
Converting: promote it to its own exact match keyword with a bid of its own, and add it as a negative in the campaign that discovered it, so the two stop bidding against each other with your money on both sides.
Spending without converting, but plausibly relevant: leave it if it has not had enough clicks to be judged, and negate it once it has.
Simply wrong: negate it immediately and without hesitation. Somebody searching for a different product is not a shopper you are failing to convert. No bid is low enough to make them buy.
Negative exact blocks one term. Negative phrase blocks everything containing it, which is powerful and is where the damage happens.
Adding a common word as a negative phrase can silently remove a family of terms that were converting, and because they simply stop appearing, nothing in the account announces the loss. When in doubt use exact, and reserve phrase for words that are unambiguously wrong for your product.
At ad group level for terms specific to one product, and at campaign level for terms wrong for everything in it. Putting everything at campaign level is quicker and eventually blocks a term that was fine for a different product in the same campaign.
This is easier when campaigns are separated by the job they do rather than lumped together, which is the structural argument made in
campaign structure for a product launch.
A negative added during a launch two years ago is still blocking that term today. Products change, listings improve, and a term that could not convert against your old title may convert perfectly well against your new one.
Once a quarter, read the negative list. It is a short job and it is the only way anything ever comes off it.