Back to journal
KeywordsMulti-marketplacePPC fundamentals

Amazon backend keywords are a 250-byte budget, and German listings run out first.

The Search Terms field is measured in bytes, not characters, and Amazon’s guidance is that going over stops it indexing rather than trimming it. What belongs in it, what Amazon explicitly says does not, and why an umlaut costs double.

The Mirox team7 min read

The backend search terms field is one line, under 250 bytes, and it fails closed: exceed the limit and Amazon's guidance is that it stops indexing the field rather than trimming it to fit. Most listings treat it as a keyword dump. It is closer to a budget — and if you sell in German or French, it is a smaller budget than the character count on your screen suggests.

What the field actually is

Every listing has one Search Terms attribute, hidden from shoppers, that Amazon reads when deciding which queries a product is eligible to appear for. It is not a ranking dial and it is not a bid. It is a statement of what the product also is, in words the listing copy never got around to using.

Amazon's own guidance on it is short and mostly a list of things not to do. Do not include brand names — including your own — or other product identifiers. Do not duplicate content that already appears in the title or the bullets. Do not repeat a keyword: once is enough. Do use synonyms, hypernyms and spelling variants of what the visible listing already says, which is the case Amazon cites directly: if the title says "whiskey", the search terms should carry "whisky".

Read that list again and notice what it rules out. Everything most sellers put in this field — the brand, the title words, the same phrase in four arrangements — is explicitly named as the wrong content. The field is for the vocabulary gap, not for a second copy of the listing.

Bytes are not characters, and this is where EU listings lose

The limit is measured in bytes. An ASCII character is one byte, so an English seller can treat the limit as roughly 249 characters and be right. A character with an umlaut or an accent is two bytes in UTF-8, and every one of them silently spends double.

A German search-terms field of 230 characters containing 25 instances of ä, ö, ü or ß is 255 bytes. It looks comfortably inside a 250 limit if you are counting what you can see. It is over, and the failure mode is not a truncated tail — it is a field that stops working.

This is the sort of thing that is invisible until someone measures it, because nothing in the listing looks wrong. You can check any field in a browser console in one line — new TextEncoder().encode(s).length — and the number that comes back is the number Amazon is using. Do it per marketplace: the same product's German field and its English field are not the same length even when they are the same words.

It is the same class of error as the one that makes translated campaign structures underperform native ones: an assumption that holds in English quietly stops holding one marketplace over, and nothing raises its hand.

What this has to do with what you pay per click

Backend keywords do not gate your advertising. You can bid on any keyword you like, whether or not the listing is indexed for it, and Amazon will take the bid.

The connection runs the other way. A listing that is not relevant to a term is not competitive for it, so the bid required to win an impression is higher and the conversion behind that impression is worse. Sellers meet this as a keyword that "does not spend", or one that spends and never converts, and they treat it as a bidding problem. Sometimes it is a relevance problem wearing a bidding problem's clothes, and the fix is upstream in the listing rather than downstream in the bid.

Which is worth saying plainly, because the reverse mistake is more common: filling the backend field with high-volume terms the product is not, in the hope of catching their traffic. That does not make the product relevant. It spends your byte budget on terms you will lose, and pushes out the long-tail vocabulary you would have won.

A twenty-minute pass that is usually worth doing

  1. Measure before you edit. Byte-count the field on every marketplace you sell in. Anything at or over the limit is the first thing to fix, because it may be contributing nothing at all right now.
  2. Delete the brand. Yours and everyone else's. Amazon names brand terms and product identifiers as content that does not belong here, and your own brand is already indexed from the fields that carry it.
  3. Delete every word already in the title or the bullets. This is usually where half the field goes, and none of it was doing anything.
  4. Spend what you freed on vocabulary the listing lacks: the regional word, the older word, the misspelling shoppers actually type, the generic category term your marketing copy is too specific to contain.
  5. Pull the terms from your own search-term reports, not from a keyword tool. Queries that already converted for you are evidence; a tool's volume estimate is a guess about someone else's catalogue. The same reports feed the negative list nobody maintains, so you are reading them anyway.

What it will not do

Backend keywords will not rank a product that does not convert, will not rescue a listing with no reviews, and will not change what a click is worth. They are a cheap correctness fix, not a growth lever — the value is in removing a silent failure, not in adding a new advantage.

They also sit outside the part of the account that moves weekly. Match types decide who reaches the auction and the bid decides whether their click clears margin. The backend field decides whether the listing was a plausible answer to the query in the first place. All three have to be true; only one of them is worth looking at every week.

The one-line version

One field, under 250 bytes, measured in bytes rather than characters, containing only the words your listing does not already say — no brand, no repeats, no title copy. Over the limit it indexes nothing, and in German and French you will hit that limit sooner than the screen suggests.

Common questions

How long can Amazon backend keywords be?
Under 250 bytes for the whole Search Terms field — one field per listing, not one per line. Amazon measures bytes, not characters: a-z and 0-9 cost one byte each, but accented and non-Latin characters cost two or more, so a German or French field runs out sooner than its character count suggests.
What happens if I go over the byte limit?
Amazon’s guidance is that the field stops being indexed — not that it gets trimmed to fit. Going three bytes over does not cost you three bytes of coverage; it risks costing you the whole field. Seller Central shows a warning next to the field when you exceed it, which is the only signal you get.
Should I repeat words from my title in the backend?
No. Amazon asks you not to duplicate content that already appears in other attributes, and repeating a word does not weight it. The field is for what the listing does not already say: synonyms, spelling variants, hypernyms, and the words shoppers use that your copywriter would not.
Do backend keywords affect my PPC?
Not directly — you can bid on any keyword regardless of what is in the field. The effect is indirect and real: backend terms are one of the ways a listing becomes indexed for a term, and bidding on terms your listing has no relevance for is how you end up paying more for less. Indexing does not win the auction, but it changes what the auction costs.

Backend keywords are a listing job, but the search-term reports that should feed them are a PPC one. We read those reports for free and send you the converting queries your listing is not indexed for — useful whether or not you ever run a campaign with us.

Further reading

What this looks like on your account

Watch the AI before a cent moves.

Simulation Mode is free, needs no card, and runs on your real account read-only for 30 days. Watch the agents make every call, then switch on live bidding when the traces convince you.

30 days free in Simulation Mode · paid tiers from $169/mo