How to localize your app with AI translation

By the end of this guide you will have your app's interface text translated into at least one new language, reviewed, and shipping from your normal release process. To localize your app with AI, you move the text out of your code into translation files, send those files to a machine translation or LLM-based tool, check the output, and load the results back into the app.

Which tool you use depends on how much of the process you want to automate and how much human review you need. This guide walks through the steps and compares seven options from our Developer platforms category, from a free GitHub Action to enterprise translation platforms.

Key takeaways

• Separate text from code first. AI translation works on files of strings (such as JSON or XLIFF), not on text buried in source code.
• The main decision is how much process you need: a GitHub Action for a small team, a translation management platform for ongoing work, or a managed service with human reviewers.
• Check the plan limits that matter for apps: word allowances, placeholder handling, supported file formats and who can edit.
• Machine translation is a draft. Have a fluent reviewer check legal, medical, financial and brand-critical text before it ships.

What you need before you start

• Source text in one language. Pick a source language and make sure the wording is final enough to translate. Changing source text later costs time.
• Strings in files. Your app should use an internationalization library or format such as JSON or XLIFF. If text is hard-coded, extracting it is the first job.
• A list of target languages and markets. Start with one or two.
• A glossary. Product names, feature names and terms that must not be translated.
• Access to a reviewer. A fluent speaker for each language, whether in your team, a contractor or a service.
• Time. A first language can be translated in under an hour once strings are in files. Review and testing take longer.
• Difficulty. Moderate; some developer work is required.

Which kind of AI tool fits this job

An automation inside your repository. ai-i18n is a free MIT-licensed GitHub Action that translates XLIFF and JSON files with a model you choose (Anthropic Claude, OpenAI, Ollama or AWS Bedrock). It records a hash of each string so only new or changed text is sent, and it can commit the results. You pay only for the model usage you bring.

A translation management platform. These keep strings in projects, offer an editor and glossaries, and add machine translation; Lokalise and Phrase also list translation memory. Lokalise offers AI and machine translation, tasks, in-context editors, an API, CLI and over-the-air delivery. Tolgee is open source, offers in-context editing inside your running app, an AI translator that uses glossaries, and either a free cloud plan or self-hosting. Phrase is a capacity-based enterprise suite covering translation management, software localization and machine translation.

A one-off service for a specific format. LocalePack translates Chrome extension messages.json files into a ready _locales ZIP and charges once per order. It is a narrow tool, but a clean fit if that is your format.

Managed or enterprise translation. LILT combines contextual AI models with connectors, adds human expert verification on its Enterprise tier, and prices through sales. Tolq combines machine translation with professional linguists at quality levels you choose per request, billed per word.

Step 1: Extract your strings

You end this step with all user-facing text in translation files rather than in code.

1. Adopt an internationalization library for your framework, or use your framework's built-in support.
2. Replace hard-coded text with keys that point to your source-language file.
3. Use placeholders for variables, such as names and counts, rather than joining sentences in code.
4. Handle plurals with your library's plural rules, not with manual if statements.

Check it worked by switching the app to a pseudo-language or a test language and seeing the text change everywhere.

Step 2: Write a glossary and style notes

Good input makes better output.

1. List the terms that stay in the source language, such as brand and feature names.
2. List terms that must always translate the same way.
3. Note the tone you want: formal or informal, and how you address users.
4. Add short descriptions to strings that are ambiguous, for example whether "Order" is a noun or a verb. Tools like LocalePack read description fields as context, so the idea is widely useful.

Check it worked by having a colleague read the glossary and confirm there are no obvious gaps.

Step 3: Choose a tool and a workflow

Match the tool to your team and format.

1. If you are a small team on GitHub with JSON or XLIFF files, start with ai-i18n.
2. If several people will translate and review, shortlist Lokalise, Tolgee or Phrase.
3. If you want human linguists in the loop, compare LILT and Tolq.
4. Check supported formats and file sizes. LocalePack limits source files to 500KB.
5. Check plan limits against your content. Tolgee's Free plan allows 30,000 words and 3 seats, and Lokalise bills by processed words per year.

Check it worked by confirming your file format and word count fit the plan you chose.

Step 4: Translate the first language

Run one language end to end before adding more.

1. Upload or connect your source file. For ai-i18n, point the action at your files and list the target languages. For platforms, create a project and import your strings.
2. Add your glossary where the tool supports one.
3. Run a dry run if available. ai-i18n has a dry-run mode that previews changes.
4. Start the translation and wait for the results.

Check it worked by opening the translated file and confirming keys, placeholders and formatting are intact.

Step 5: Review and correct the output

Machine output is a draft that needs a human eye.

1. Have a fluent reviewer read the strings in context. Tolgee and Lokalise both list in-context editing, which shows text where it appears in the app.
2. Check placeholders, plurals, numbers, dates and punctuation.
3. Look for text that is too long for buttons and labels.
4. Pay extra attention to legal, medical, financial and safety text, and have a professional translator check it.
5. Save approved translations so they are reused. Tolq's translation memory only reuses identical sentences you have approved.

Check it worked by walking through the app in the new language and noting no cut-off, broken or untranslated text.

Step 6: Wire translation into your release process

Make localization part of normal development instead of a separate project.

1. Commit translation files to your repository, or sync them from the platform.
2. Run translation when source text changes. ai-i18n can run on every push, commit or schedule, and it only sends new or modified strings.
3. Fail the build when keys are missing, so untranslated text does not reach production.
4. Decide how updates reach users. Lokalise lists over-the-air delivery for updated strings, and Tolgee lists content delivery to cloud storage without a new app release.

Check it worked by changing one source string and seeing it flow through to the translated file.

Step 7: Test in the product and keep improving

Real use reveals problems that files do not.

1. Test layout with longer languages and with right-to-left languages if you add them.
2. Check formats for dates, currency and numbers.
3. Collect feedback from native-speaking users and fix recurring issues in the glossary.
4. Review quality periodically as your product text changes.

Check it worked by having a native speaker complete your main user journey and report no blocking problems.

Tools for this job compared

Tool	Good fit for	Pricing	Stated limits
Lokalise	Ongoing localization with an editor, API, CLI and webhooks	Free trial, paid from $149/mo (checked 24 Oct 2026)	Capped by processed words per year; API limited to 6 requests per second; card payment only on Explorer and Growth
Tolgee	Open-source platform with in-context editing, cloud or self-hosted	Freemium, paid from €58/mo (checked 24 Oct 2026)	Free plan: 30,000 words, 3 seats; self-hosted enterprise features need a paid license key
ai-i18n	Small teams on GitHub using XLIFF or JSON	Open source	XLIFF and JSON only; you pay your model provider; quality depends on the model
LocalePack	Chrome extension messages.json files	Paid, from $9.99 one-time (checked 24 Oct 2026)	Source files up to 500KB; no translation memory or review workflow; files stored for 7 days
Phrase	Enterprise suites with software localization and machine translation	Contact sales	Capacity-based plans; Business and Enterprise are quoted; multimedia is an add-on
LILT	Enterprise localization with connectors and human verification	Contact sales	Quoted through sales; human verification and API access are Enterprise features
Tolq	Per-word translation with human review levels	Paid, from $0.05, usage-based (checked 24 Oct 2026)	Rates vary by language; translation memory reuses only 100% matches

Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.

Common mistakes to avoid

• Translating before the source text is stable. Every wording change creates new work. Fix it by freezing copy for a release and using tools that only retranslate changed strings, as ai-i18n does.
• Breaking placeholders. Variables such as names and counts can be changed or removed by a translation tool. Fix it by choosing tools that preserve placeholders, and by testing with real values.
• Skipping human review. Machine translation can be fluent and wrong. LocalePack's listing says translations are provided as-is and critical text should be reviewed. Fix it by assigning a fluent reviewer for each language.
• Ignoring layout. Translated text is often longer than the source. Fix it by testing buttons, menus and error messages in each language before release.
• Forgetting where your text goes. Translation tools and model providers process your content. Fix it by reading each vendor's data terms and, for unreleased or sensitive text, choosing options such as Tolgee's self-hosting or ai-i18n with a local Ollama model.

How we put this together

This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.

Frequently asked questions

Is there a free way to localize my app with AI?

Yes. ai-i18n is free and open source, and you pay only for the model API you bring, which can be a self-hosted Ollama model. Tolgee has a Free cloud plan with 30,000 words and 3 seats and can also be self-hosted. Lokalise, Phrase and Tolq list free trials or free sign-up options, and their paid usage follows.

Can AI translate an app accurately?

It produces a usable draft for many languages, but accuracy varies by language, model and context. Short interface strings are ambiguous without context, so add descriptions and a glossary. Have a fluent person review the result, and have a professional check legal, medical or financial text.

What file formats do these tools handle?

It depends on the tool. ai-i18n handles XLIFF 1.2, XLIFF 2.0 and JSON. LocalePack handles Chrome extension messages.json files and says it handles web app i18n JSON files for next-intl and i18next. Platforms such as Lokalise and Tolgee work with translation keys through APIs and CLIs, and Tolgee also lists web and mobile SDKs. Check each listing for your exact format.

How long does it take to localize an app?

Once your strings are in files, machine translation is usually the quick part; LocalePack, for example, says most jobs finish in under 5 minutes. The longer work is extracting strings, reviewing in context and testing layout, which depends on your app's size.

Can I keep my content private?

You can limit exposure by self-hosting. Tolgee can be self-hosted with Docker and PostgreSQL, and ai-i18n can use a local Ollama model. Cloud tools process your text on vendor or model-provider systems, so read their data terms before sending unreleased or sensitive strings.

Compare more options in our Developer platforms category.