Localize App Store keywords with AI: what works, what to check

A direct translation of "habit tracker" may be grammatically correct and still not match what people type. People in different countries often use a different word for the same idea, or keep an English term. The AI step is useful for generating the candidate phrases at speed; it is not an authority on which one people search. Apple does not document how its search treats word forms in different languages, so test with rank data instead of assuming.

LocalASO adds a mode option to its board_add_keywords tool, described on the board_add_keywords page. With mode: "localize" it adapts every term to each country's language through your configured AI provider, and it never translates the same term and country twice. In the app, bulk localization uses the fast model setting (default haiku for the Claude Code CLI and the Anthropic API), as one batched call per language. Every request asks for strict JSON, which is validated and retried if invalid, and localizations are cached.

The AI request goes from your Mac straight to the provider you picked; LocalASO has no server. That is the setup described in the AI providers docs.

When an agent calls the tool through the MCP server, one constraint applies: the app encrypts the keys it stores, and the server cannot read them. Localize mode therefore needs a provider that works without a stored key. The default Claude Code CLI does. An Anthropic API key saved in the app is not available to the server, and the docs name only the Claude Code CLI as working there, so test any other provider before you rely on it. If you mainly localize from the app itself, the provider you picked in Settings is used. For a local-only setup see Ollama for ASO.

1. Add your English keywords with mode: "english" first and check them, so you know which ones are worth localizing.
2. Add the best ones with mode: "localize" for a few countries at a time.
3. Run keywords_check (details). The result shows the localized phrase, rank, volume and difficulty for each country.
4. Drop phrases with no volume signal and keep those with a Push or Maybe verdict.
5. Have a native speaker review the final field. The keyword field holds 100 bytes, and non-Latin scripts use more bytes per character, so a localized list may hold far fewer words than the English one.
6. Remember Apple's fallback rules for locales that share a storefront; see the localization guide.

"Add these 10 keywords to Streakly, localized to Germany, France and Japan. Then check them and show me, per country, the phrase the AI chose and its rank." This should trigger board_add_keywords with countries: ["DE","FR","JP"] and mode: "localize", then keywords_check. It is an example prompt, not recorded output.

• The AI can pick a phrase that is correct but unused. Data from the check is the test.
• Volume and difficulty are estimates from rating counts, not Apple search counts.
• LocalASO covers 52 storefronts, each in its local language, so large lists across many countries add up to many checks; at most 60 run per keywords_check call.

Learn the surrounding workflow in the App Store localization guide or compare locales with the localizations tool. Prompts are on AI agent prompts.