How to Translate App Store Keywords Without Translating Word for Word

A word can have several correct translations, and people usually search with only one of them. Sometimes they keep an English loanword. Sometimes the natural phrase is a compound word or a different concept entirely. A translator has no way to know which version people type into the App Store search field, and Apple does not publish that data. In a LocalASO sample of 753 phrase pairs, local and English searches shared a mean top-10 overlap of only 0.26, which is covered in the local versus English top 10 research.

The keyword field also counts bytes, not characters, so the same idea costs different amounts in different languages. Apple's reference states a limit of 100 bytes. Japanese characters use 3 bytes each in UTF-8, so the field holds roughly a third as many.

1. Start from the intent, not the English words. Write one sentence for each keyword: what is the person trying to do? "Track a daily habit" is the intent behind "habit tracker".
2. Produce 3 candidates per keyword. Get them from a native speaker, from your AI tool, from the competitor listings in that storefront, or all three. Include a literal translation, a loanword and a local phrase if they differ.
3. Search each candidate in the target storefront. Open the App Store for that country and search the phrase. Do apps that match your product appear? A phrase that returns unrelated results or nothing is not a good candidate.
4. Compare strength. Look at the top 10. Many results with thousands of ratings mean the phrase is hard for a new app; see keyword difficulty.
5. Choose and count bytes. Keep one or two per intent and pack the field, with commas and no spaces, never repeating words from your name, subtitle or category. Check with the keyword field checker.
6. Review with a native speaker. One reading catches wrong connotations, awkward compounds and unintended meanings.

A hypothetical habit tracker, Streakly, wants the concept "habit tracker" in Germany. Three candidates, with their lengths:

All three together are 53 bytes with commas, over half the 100-byte budget for one idea. Streakly searches each in the German App Store, looks at who ranks, and keeps the one or two that return the right apps. A single umlaut, as in "tägliche Routine", is 17 bytes for 16 characters, which is why you count bytes rather than characters.

The same idea in Japanese: "習慣化" is 3 characters and 9 bytes, "習慣トラッカー" is 7 characters and 21 bytes, "ルーティン" is 5 characters and 15 bytes. At 3 bytes a character, a field made only of such terms holds about 33 characters.

• Do not assume how Apple matches word forms. Apple does not document plural, stemming or compound handling in search for the keyword field. Test a variant before you spend bytes on it.
• Do not copy competitor names. Names of other apps or companies are not allowed in the keyword list.

An AI model is useful for generating candidates quickly and for naming alternatives you had not thought of. It does not know the current App Store results in a storefront. The check in step 3 is what makes the list reliable.

LocalASO takes your keyword, adapts it into local phrases with your own AI, and checks each phrase in the matching storefront out of 52, with volume and difficulty estimates. It is a faster version of steps 2 to 4; step 6 stays human. See the localization docs and the broader localization workflow.