Negate a relevant search term once it has enough clicks that zero orders would be unlikely if it converted like your product: about 3 divided by your conversion rate. Amazon's suggested 20 clicks suit a product converting 15%. At 5%, a search that sells normally shows 20 clicks without an order over a third of the time.
Zero orders is a result that a good search produces by chance. The click count decides how often you block one.
Zero orders happens to good searches
Amazon's guide to negative targeting recommends evaluating a keyword after it has received at least 20 clicks. That is a floor, not a verdict, and how much it tells you depends on how often your product's clicks turn into orders. If a search converts at the same rate as your product, the chance that it shows no order at all is one line of arithmetic:
The arithmetic
Chance of 0 orders = (1 − conversion rate) ^ clicks
Clicks for under a 5% chance ≈ 3 ÷ conversion rate
Run it for three conversion rates:
| Conversion rate | 10 clicks | 20 clicks | 30 clicks | 40 clicks | 60 clicks |
|---|---|---|---|---|---|
| 5% | 60% | 36% | 21% | 13% | 5% |
| 10% | 35% | 12% | 4% | 1.5% | 0.2% |
| 15% | 20% | 4% | 0.8% | 0.2% | under 0.1% |
Read the 20-click column. For a product that converts 15% of its clicks, 20 clicks with no order would happen to a normal search about once in 25, so it is fair evidence. For a product that converts 5%, it happens to a normal search more than a third of the time, and a seller who negates at 20 clicks blocks a lot of searches that would have sold.
The click count your conversion rate needs
Three divided by the conversion rate lands just under a 5% chance on every row: 60 clicks at 5%, 30 at 10%, 20 at 15%. Use your product's own conversion rate, orders divided by clicks over a long stretch, on the searches you consider relevant. The search's own rate is the thing you are trying to find out, so it cannot be the yardstick.
The same count in dollars is the clicks times your average cost per click. Because cost per click divided by conversion rate is what an order normally costs you in ads, the floor works out to about three orders' worth of ad spend:
One product, worked through
Conversion rate 10%, cost per click $0.60
Ad cost of an average order = $0.60 ÷ 10% = $6
Clicks needed = 3 ÷ 10% = 30
Spend floor = 30 × $0.60 = $18, three orders' worth
A search with no orders after $18 is either the roughly one-in-twenty-five unlucky case or converting worse than the product. Before $18 it has not said much either way. Whether $6 an order is affordable is a separate question, answered by the product's break-even ACOS: the break-even ACOS calculator works it out from price, cost and fees.
An irrelevant search needs no count
The arithmetic assumes the search could sell. A search for a different product, a part you do not make or a size you do not stock is not a statistics question. Amazon lists shopping terms you don't want your ads to appear against among the things to add as negatives, with no click count attached. The count is for searches you would be glad to sell on, which is where blocking too early costs orders.
Before you block it, read its whole history
A search term review usually reads a recent window, such as the last 14 or 30 days. A search that is quiet in that window can be one that sold steadily for a year. Here is one with both readings:
| Window | Clicks | Spend | Orders | Ad sales | ACOS |
|---|---|---|---|---|---|
| Last 14 days | 18 | $10.80 | 0 | $0 | none |
| The 12 months before | 400 | $240.00 | 40 | $1,200 | 20% |
Over the year the search converted 40 of 400 clicks, 10%, at a 20% ACOS. At 10%, a run of 18 clicks with no order happens about 15% of the time, roughly one fortnight in seven, to a search that is still fine. Recent spend on it is $10.80, well under the $18 floor for that conversion rate. The fortnight is not enough to overturn the year.
A negative changes what the search can do from now on, in the ad group or campaign it is added to. Once it is blocked, the search stops producing evidence there, so nothing in that place will tell you the block was wrong. The longer history does not explain why the search went quiet. A season, a change in delivery or a competitor can all look the same, and that needs a look of its own. It does tell you what you would be blocking.
The opposite mistake happens in bids and rules: a long window that averages away a recent break. Why ninety days hides the fortnight that broke covers that side. Read both windows before acting on either.
Where this still needs a person
The chance table assumes every click on the search has the same chance of an order. Real searches vary with price changes, stock, competitors and the day of the week, so treat the figures as a guide to how much evidence you have, not as a precise probability. Your product's conversion rate is itself an estimate, and a product with few clicks has a rough one.
The 5% is also a choice. It means accepting that about one search in twenty to twenty-five that would have sold normally still reaches the floor with no order. A seller with a thin margin may accept more of that risk to stop spend sooner; a seller protecting a hero search may want less. Relevance is a judgment too: whether a shopper typing that search could want your product is something only you can decide.
What to do with this
- Work out your product's conversion rate: orders divided by clicks, over three months or more, on searches you consider relevant.
- Divide 3 by it for the click count, and multiply that by your average cost per click for the spend floor.
- Filter your search term report for zero orders and spend above the floor, highest spend first. In Vendlen, save that as a filter with the condition in its name, so the floor is visible six months later; the Search terms workspace describes saved filters.
- Block irrelevant searches on sight, whatever their clicks. For the rest, pull the longest history you have before deciding. A search that has sold before needs a reason to have stopped, not just a quiet window.
- Choose the narrowest negative that matches what you learned: negative exact or negative phrase. If you block by rule in Vendlen's Automation, the Negation rule has a 365-day orders checkbox, on by default, that skips any term with an order in the past year.
Questions sellers ask
Is 20 clicks enough to negate a search term on Amazon?
It is the minimum Amazon suggests, and it suits a product that converts about 15% of its clicks: a search converting at that rate shows 20 clicks with no order about 4% of the time. At a 5% conversion rate the same happens about 36% of the time, so wait for nearer 60 clicks.
How much should a search term spend before I negate it?
Take the clicks your conversion rate needs, 3 divided by the conversion rate, and multiply by your average cost per click. That comes to about three times what an order normally costs you in ads. With invented figures, a 10% conversion rate and a $0.60 click give $6 an order and an $18 floor.
Should I negate search terms with clicks but no sales?
Negate them on sight when they are irrelevant to the product. A relevant search should first reach the click count your conversion rate needs, and its longer history should be checked: a search with orders last year and none this fortnight is a different case from one that has never sold.
Can I undo a negative keyword?
Yes. Amazon notes that adding a negative is not permanent and you can remove it. The cost is that a blocked search produces no new results where it is blocked, so nothing will tell you the block was wrong unless you go and look at the search elsewhere.