Keyword harvesting takes a search term that has sold through an auto or broad campaign and adds it as its own keyword, usually exact match, with a bid of its own. Adding it does not stop the source campaign from serving that search. If you want the traffic to move, negating the term in the source is a separate step.
Harvesting adds a route to a search. Whether the old route closes is a second decision.
Two routes to one search
Amazon's own guide describes the loop: use the automatic campaign's search term report to find the keywords worth targeting in a manual campaign. Most write-ups then describe the result as the term having moved. It has not. The auto campaign that found the term is still running, and it can still serve the same search. You now have two targets that can serve one query, at two bids.
That is sometimes what you want and sometimes not, which is why it is worth treating as its own decision rather than a step everyone does automatically. It comes up again below, after the two questions that come first: which terms, and at what bid.
How many orders a term needs first
One order is a coincidence. A harvested keyword is a commitment, so it should wait for enough orders that the term's conversion rate and ACOS mean something. How many depends on the match type you create, because the looser ones reach searches the term never proved itself on:
| New keyword | What it can match | Example: orders first |
|---|---|---|
| Exact | The proven search, word for word | 5+ |
| Phrase | Any search containing it, in order | 15+ |
| Broad | Its words in any order, and related terms | 40+ |
The ceiling bid should run the other way. Amazon recommends the highest bid on exact, lower on phrase and lowest on broad, which matches the risk: a broad keyword spends on searches the original term never proved itself on.
Price the new keyword from the term's own numbers
A new keyword has no history, so its opening bid is the one number nobody can check against results yet. The term behind it already has plenty. Say an auto campaign's term earned 9 orders of $25 from 90 clicks over 60 days, at an average cost per click of $0.90:
| Where the opening bid comes from | Opening bid |
|---|---|
| The ad group's default bid, set before this term existed | $1.10 |
| 80% of the term's average CPC: $0.90 × 80% | $0.72 |
| Revenue per click × a 30% target ACOS: ($225 ÷ 90) × 30% | $0.75 |
The term spent $81 for $225 of sales, an ACOS of 36%. The last two rows are priced from that evidence; the first is not. The third is the target ACOS bid, worked through in the bid formula and its limits. The 30% target is the product's break-even ACOS less the margin you want to keep, which the break-even ACOS calculator works out from price, cost and fees.
Vendlen's Import rule prices the new target this way: a percentage of the term's own average cost per click, clamped by an optional minimum and maximum bid, or the ad group's default bid when you leave the percentage blank. It reads candidate terms from all campaigns advertising the ASIN, the same campaign or the destination ad group, and by default does not create a target that already exists.
Negate the source, or leave it running
Both choices are defensible. They cost different things:
- Negate the term in the source. The search now reaches you only through the new keyword, at the bid you chose for it. If that bid is too low to win the search, you lose it, and nothing else catches it.
- Leave the source running. The search can still reach you through either target, so the search term report may keep showing the term under the source campaign.
Harvesting usually exists to control the bid on a proven search, and that argues for the negative. Leaving the source open makes sense while the new keyword finds its level. Either way, decide it once and do it on purpose. In Vendlen Automation, an Import rule only adds; a Negation rule on the source does the closing.
The expensive term with lots of orders
Harvesting is usually described for winners. The same move answers a harder case: a term that sells a lot inside a broad keyword, at an ACOS you cannot live with. Blocking it throws away real revenue. Cutting the broad keyword's bid cuts every good search it serves too.
Give the term its own exact keyword at a lower bid instead. A term with 25 orders of $24 from 400 clicks at $1.20 spent $480 for $600: 80% ACOS and $1.50 of revenue per click. An exact keyword opened at 55% of its CPC bids $0.66, and you pay at most your bid, since a bid is the maximum you are willing to pay for a click. If revenue per click held at $1.50, a click at $0.66 or less would run at 44% ACOS or below. On its own, the new keyword only adds a cheaper route; the expensive one stays open until you negate the term in the source.
Where this still needs a person
The arithmetic assumes the term keeps converting the way it did at the old bid. A different bid can change where and how often the ad shows, so the new keyword's ACOS has to be read again after it has run. And every harvested keyword is a target you now own: it needs its bid managed, and nothing about harvesting tells you when it stops deserving one.
What to do with this
- Pull 60 days of search terms from your auto and broad campaigns. Shortlist terms with at least five orders and an ACOS under your target for exact; ask for more orders for phrase or broad.
- Price each new keyword from the term: a share of its average CPC, or its revenue per click times your target ACOS. Do not let it inherit a default bid.
- Decide per term whether to negate it in the source, and write the decision down.
- Look for terms with many orders and a high ACOS inside broad keywords. Those want their own exact keyword at a low bid, not a block on the term.
- Two weeks later, read the new keywords and the source campaigns side by side, to see which target is serving each harvested search.
Questions sellers ask
Should I negate harvested search terms in the auto campaign?
If you want the search to reach you only through the new keyword, yes. The auto campaign keeps serving the search until you do. The cost is that the new keyword then has to win the search on its own bid.
How many orders before harvesting a search term?
Amazon does not set a number. One order is a coincidence; a handful at an ACOS under your target is a reasonable floor for an exact keyword. Ask for more before creating phrase or broad, because those reach searches the term never proved itself on.
What bid should a harvested keyword start at?
One priced from the term's own history: a share of its average cost per click, or its revenue per click times your target ACOS. Amazon recommends bidding highest on exact match, lower on phrase and lowest on broad.