Auto vs manual Amazon campaigns: run both, forever.

Auto versus manual is not a strategy decision. It is a labour decision — who does the keyword research, Amazon's algorithm or you — and on its own it decides nothing about whether you make money. Both campaign types hand the same auction the same question: how much is this click worth? Neither one answers it for you. A badly-bid manual campaign and a badly-bid auto campaign lose money at exactly the same rate. So the useful version of this question is not "which type," but "which type does which job, and what sets the bid inside it."

What the two actually are

Strip away the folklore and the difference is narrow. In both, Amazon runs the auction and charges you for the click. The only thing that changes is who picks the search terms.

An auto campaign lets Amazon choose which search terms your ad competes on, sorted into four targeting groups — close match, loose match, complements, and substitutes. You set a default bid (optionally one per group) and the algorithm decides where to show. A manual campaign inverts the control: you choose the keywords or ASINs and the match type, and set a bid per target. Amazon still runs the auction — you just stop letting it pick the words.

That is the entire technical distinction. Everything else written about auto versus manual is advice about how to sequence the two, and most of that advice is a decade old and quietly leaking money.

The standard playbook — and where it leaks

The textbook flow goes: launch an auto campaign to let Amazon discover converting search terms, mine the search-term report for winners, then "graduate" those winners into a manual campaign where you bid them precisely, and add them as negatives in the auto so the two stop competing. It is not wrong. It is just slow, leaky, and built for an era when keyword discovery was the hard part.

The first leak is self-competition. Until you negate a graduated term in the auto campaign, both campaigns bid into the same auction for the same shopper — you are paying to outbid yourself, and Amazon is delighted to let you. The second leak is latency: harvesting on a weekly cadence means a search term that started converting on Monday keeps running at a generic auto bid until you get round to the report. The third is the quiet one — most sellers harvest the winners and never prune the losers, so the auto campaign keeps spending on search terms that will never convert while attention drifts to the shiny new manual campaigns.

Why "graduate winners to manual" is only half true

The instinct behind graduation is sound: a term you know converts deserves a deliberate bid, not whatever the algorithm felt like that hour. But the conclusion people draw — that manual is the "serious" campaign and auto is training wheels — is backwards in 2026. Auto is not a beginner tool you outgrow. It is a permanent discovery engine, because shopper language never stops drifting: new competitor names, new seasonal phrasings, new long-tail variants that no human keyword list anticipated. The moment you switch auto off entirely, you go blind to everything you have not already thought of.

So the mature structure is not auto then manual. It is auto and manual, indefinitely, each doing the job the other is bad at. Auto keeps surfacing terms you did not know to bid on. Manual lets you price the terms you do know against margin. The graduation step is real, but it is a continuous flow, not a one-time promotion — and the negatives that keep the two from colliding have to flow just as continuously.

The EU wrinkle nobody warns you about

Auto campaigns are far less plug-and-play across marketplaces than the US-centric advice admits. Amazon's loose-match and substitute targeting lean on its understanding of shopper intent in each language — and that understanding is uneven. An auto campaign that behaves sensibly on amazon.de can spray a Spanish or Italian listing across barely-related search terms, because the semantic match is weaker and the competitive data thinner. Running one bid logic across DE, FR, IT, and ES as if they were one market is how a tidy auto campaign turns into a wasted-spend machine in three languages at once.

This is the same trap we describe in our work on US-to-EU expansion: sellers running cross-marketplace copy without per-region tuning report roughly a 28% ACoS hit. Auto campaigns amplify that, because you have handed targeting to an algorithm that is sharper in some of your languages than others. Per-marketplace thresholds are not a nicety here; they are the difference between auto helping and auto bleeding.

What actually decides profit — and it is neither type

Here is the uncomfortable part. You can run the cleanest auto-and-manual structure in your category and still lose money, because the structure says nothing about the bid. A campaign type is a container; the bid is the decision. And the only bid that protects profit is one priced against the margin of the specific ASIN being advertised — not a flat default, not last week's ACoS, not whatever the bulk-edit nudged everything to.

Break-even ROAS = 1 ÷ margin. A product carrying 35% margin after COGS and fees breaks even at about a 2.9x ROAS — a 35% ACoS. A bid is only worth placing if the expected revenue from the click clears that line with the profit you actually want on top. Auto or manual, that math is identical; the campaign type just decides who chose the search term, not whether the click is worth winning.

We made the same point about ROAS versus ACoS: the framing debate is a distraction from the number both metrics ignore, which is your margin. Auto versus manual is the structural cousin of that same distraction. The container matters far less than what you pour into it.

How to actually structure it in 2026

1. Run auto permanently as a discovery layer, with a conservative default bid and the four targeting groups split so you can see close match separately from substitutes. Treat it as a sensor, not a workhorse.
2. Run manual alongside it for the terms you already know convert, bidding each one against the ASIN's break-even, not a blanket target. Use exact match for your proven money terms and phrase or broad for controlled expansion.
3. Close the loop continuously, not weekly. Every term that proves itself in auto gets a deliberate manual bid and an immediate negative in the auto, so the two never bid against each other.
4. Prune as hard as you harvest. The search terms that spend without converting are the ones that decide your true cost of advertising; negate them on the same cadence you promote winners.
5. Tune per marketplace, never blended. A bid, a default, and a negative list that work on amazon.de are a guess on amazon.it. Set thresholds in each currency and each language as CPCs differ by region.

The one-line version

Run both, forever — auto to find the words you did not know to bid on, manual to price the words you do — and stop pretending the choice between them is your strategy. Your strategy is the bid: every click priced against the margin of the ASIN behind it, in the currency and language of the marketplace it runs in. The campaign type just decides who picked the keyword. The bid decides whether you keep the money.

See the margin context attached to a single bid, or read how negatives cut the spend auto campaigns leak.