Amazon PPC keyword harvesting: when to graduate a search term, what to bid and how to prove it worked

1. 1Pull

   Search term report, 30 to 60 days, archived

2. 2Total

   Sum each term across every campaign

3. 3Judge

   Click floor = 1 / conversion rate x 2.5

4. 4Graduate

   Exact keyword, bid from margin

5. 5Isolate

   Negative exact in the source campaign

6. 6Prove

   Term's total orders up, TACoS flat or down

What Amazon PPC keyword harvesting is, and why it still matters

Auto, broad and phrase campaigns are discovery tools. They show your ad on queries you never typed, and the search term report tells you which of those queries a shopper actually used before clicking. Harvesting turns that list into decisions: graduate the terms that convert, negate the ones that burn money, and keep watching the rest.

The reason to graduate a winner is control. Inside an auto campaign one bid covers hundreds of queries. In an exact match campaign the term has its own bid, its own budget and its own placement settings, so you can push it when it is profitable and pull back when it is not.

Some tool vendors now call harvesting obsolete because their software moves terms automatically. The tool changes who presses the button, not the logic. Someone still has to decide what counts as a winner, and that decision is where most accounts go wrong.

The two-sided move most accounts get half right

A harvest has two halves. First, add the term as an exact keyword in a manual campaign. Second, add the same term as a negative exact in the campaign it came from. Skip the second half and the auto campaign keeps bidding on the same query, so two of your own campaigns compete for one shopper and your reports split the result in two.

Use negative exact, not negative phrase, in the source campaign. Negative phrase also blocks every longer query that contains the term, and those longer queries are often where the next winners come from.

• Exact keyword added in a manual campaign, with its own bid
• Negative exact added in the source auto, broad or phrase campaign
• Branded and competitor terms kept in their own campaigns, never mixed with generic terms
• A short note logged: term, date, source campaign, starting bid

Set the harvest threshold from your conversion rate, not a click count

Published guides disagree on the click floor. Seller Labs says avoid judging a term on fewer than 20 to 30 clicks, Trellis uses about 15 to 20 clicks with no order before negating, and Hector AI cuts terms that miss target after 10 or more clicks. None of them ties the number to your product, and that is the gap.

The right floor depends on how often your listing converts. If your product converts 10 percent of clicks, you expect one order every 10 clicks, so 20 clicks with no order is a real signal. If it converts 4 percent, you expect one order every 25 clicks, and 20 clicks with no order tells you almost nothing.

A simple rule that holds up: judge a term only after it has had about two to three times the clicks you would normally need for one order. Take 1 divided by your conversion rate, multiply by 2.5, and use that as your click floor for both graduating and negating.

• 10 percent conversion rate: about 25 clicks before you judge
• 5 percent conversion rate: about 50 clicks
• 3 percent conversion rate: about 80 clicks, so most low converting terms need a longer lookback
• Graduate when the term has 2 or more orders inside that click window and ACoS at or below your break-even target

Price the new exact bid from margin

Most guides tell you to set the harvested bid "a little higher". A number from your own economics is better. The maximum you can pay per click and still hit your target is target ACoS multiplied by the term's conversion rate multiplied by the average order value.

Start the exact bid at or slightly under the term's observed cost per click in the source campaign, then let it rise toward that ceiling only if impressions are too low. Hector AI suggests setting new exact campaigns to the down only bidding strategy, which keeps Amazon from raising the bid while the term proves itself.

Worked example with illustrative numbers

These numbers are illustrative, not from a real account. Say you sell a $30 product with a 30 percent break-even ACoS and a listing that converts 10 percent of clicks overall.

In the last 60 days your auto campaign shows the search term "bamboo drawer organizer" with 42 clicks, 5 orders and $150 in sales on $33.60 of spend. That is a 12 percent conversion rate, a $0.80 cost per click and a 22.4 percent ACoS. Your click floor is 1 divided by 0.10 times 2.5, so 25 clicks. The term clears it with 42 clicks and 5 orders, under target. It graduates.

The bid ceiling is 0.30 times 0.12 times $30, which is $1.08. You start the exact bid at $0.85, just above the observed $0.80, and add the term as a negative exact in the auto campaign.

Four weeks later the check is not the exact campaign's ACoS alone. Add up orders for that search term across every campaign. If the term earned about 2.5 orders a month before and now earns 4 across the account at a similar ACoS, the harvest worked. If it still earns 2.5 and they simply moved from auto to exact, you gained control but not sales, and the bid may need to rise toward the $1.08 ceiling.

The retention window you cannot ignore

Search term data does not live in the ad console forever. Intentwise and Trellis describe a window of roughly 60 days for the console search term report, while older Amazon coverage reported by PPC Land describes a 65 day custom range. Check the current limit in your own console, but plan as if anything older than about two months is gone.

That matters for low volume products. If a term needs 80 clicks to judge and gets 20 a month, the evidence expires before it is complete. Download the report every week or two and keep your own archive, or use a tool that stores it for you, so a slow term can still graduate on four months of data.

Keyword harvesting playbook: 6 steps

Run this every one to two weeks. Weekly suits accounts spending more than a few thousand dollars a month; smaller accounts can work every two weeks.

• PPC manager, Amazon Ads console, Reports: download the Sponsored Products search term report for the last 30 days, or 60 days for low volume products, and add it to your archive. Metric: no gaps in your saved history.
• PPC manager, spreadsheet or tool: total each search term across all campaigns, then apply your click floor from 1 divided by conversion rate times 2.5. Metric: every term is tagged graduate, negate, watch or ignore.
• PPC manager, Campaign Manager: add graduates as exact keywords in a manual campaign with a bid from the margin formula, and as negative exact in the source campaign. Metric: zero graduated terms still spending in the source campaign a week later.
• PPC manager, Campaign Manager: negate terms that passed the click floor with no orders, using negative exact, and check that none are branded or core terms first. Metric: spend on zero order terms falls week over week.
• Brand owner or strategist, Business Reports and the ad console: after four weeks compare total orders for each harvested term across campaigns, and total ad sales against total sales. Metric: harvested term orders up, TACoS flat or down.
• PPC manager, notes log: record what changed and why, so the next review does not undo it. Metric: every bid change has a dated reason.

Mistakes that make harvesting backfire

• Graduating on one lucky order, then paying a high exact bid for a term that never converts again
• Leaving the term live in the auto campaign, so your own campaigns bid against each other
• Using negative phrase in the source, which also blocks future long tail winners
• Pausing the auto campaign once it has been harvested, which stops discovery for the next season
• Judging success on the exact campaign's ACoS alone instead of the term's total orders

When this advice does not apply

Very new listings with under a few hundred clicks in total do not have enough data for these rules; spend the first weeks on discovery and listing fixes. Products with very low conversion rates may need a lookback longer than the console keeps, so the archive step is required, not optional. Branded terms follow a defence logic, not a profit threshold, and seasonal products should be judged against the same season, not the last 30 days.

If you would rather not run this every week, PPC Spider's Amazon PPC management includes weekly optimization and negative targeting on every plan, from $995 a month up to $10k of monthly ad spend, with daily bid and budget management from the $1,250 Growth plan. Every plan is a flat retainer, never a percentage of ad spend.

Get the next change before your competitors do

We send the marketplace changes that matter each week, with what to do next.

Work emailSubscribe

By subscribing you agree to our privacy policy. Unsubscribe any time.