KDP Keyword Generator

Because the field is. A box holds a fixed number of bytes, and in UTF-8 an ordinary English letter costs one while an accented letter costs two and a Japanese or Chinese character costs three. So "café" is five bytes, not four, and a Japanese keyword fills the box roughly three times faster than an English one that looks the same length. Counting characters would under-report exactly the boxes most at risk of being truncated. The 50-byte figure is the cap observed in KDP's own form, not a number Amazon publishes — we say so rather than dressing it up as a rule.

Partly, yes — that is how a long-tail phrase gets built. What wastes a box is one made ENTIRELY of words already in your title or subtitle, because Amazon already matches those fields. Our filter drops only that case. It is a backstop rather than the mechanism: a KDP title is keyword-stuffed by design, so the title's own words go into the brief and the generator is asked for phrases that add new ones.

Because every filter here removes a box rather than repairing one. A phrase over the byte budget cannot be trimmed — shortening a keyword silently re-targets it at a search the author never chose — so asking for exactly seven leaves the validator no slack and turns each near-miss into a second paid round trip. The surplus absorbs the ordinary case for free.

No, and be suspicious of tools that do for backend keyword boxes. Amazon publishes no search volume, so any number attached to a box here is an estimate at best. What we can tell you is factual: the byte count of each box, whether it duplicates another box or your title, and whether it shares any word with the rest of your listing at all.

The generator ships what it produces and our rule corpus flags it — deliberately. There is one banned-term lexicon and it sits in the checker, so a promotional phrase or an Amazon program name that slips past the prompt arrives intact and renders as a labelled warning on its own box, rather than being silently deleted somewhere you cannot see. A second list living inside the generator is exactly the drift that arrangement prevents.

It reads your book. The brief comes from your actual puzzle sets, page count, and trim size, plus the title and description already saved against the book — not from what you retype, which can be stale or a different book altogether. You can override any of it, which is exactly what an author who published under a title we did not write needs.

Yes, and the byte limit is the reason this matters more than it sounds: fifty bytes is roughly sixteen Japanese characters against roughly fifty Latin ones, so a translated box that reads fine can be over the cap. The translator re-measures every box in bytes per language. It is machine translation and labelled as such.

Generating is part of the paid Listing Studio. Checking is free and unlimited: paste keywords you already have into the KDP Listing Checker and it will report the byte counts, the duplicates, and the boxes made entirely of title words.