We offer several AI solutions designed to speed up and improve your localization workflows:

Lokalise offers two AI translation tiers: Standard AI/MT and Pro AI.

These features all help assess translation quality, but they don't work in exactly the same way. AI scoring and AI checks share the same underlying quality assessment engine, while standalone AI LQA tasks use a separate process.

For more information, see Scoring translation quality and Lokalise Vantage projects.

It depends on the specific feature:

Yes. If a translation exceeds the character limit set for a translation key on Lokalise, the system automatically retries with an alternative model (up to two attempts) to generate a version that fits.

ChatGPT and other general-purpose AI tools can be useful for translating individual strings or preparing quick drafts. In fact, Lokalise Pro AI uses models from the GPT, Claude, and Gemini families.

The main difference is how these models are integrated into your localization workflow.

Pro AI can use your project's key descriptions, glossaries, style guides, and previous translations as additional context. You can translate large amounts of content directly in Lokalise, use quality assessment tools to identify potential problems, and automate translation processes through tasks, Workflows, Automations, and the API.

This reduces the need to repeatedly prepare prompts, copy translations between tools, and manually maintain consistent terminology across languages.

The advantage isn't necessarily a lower cost per translated word or a guarantee of better output in every case. It's having context, translation management, quality controls, and automation in one platform.

Lokalise Pro AI uses language models from three families: GPT, Claude, and Gemini. With dynamic routing, Lokalise can automatically select a suitable model based on the content and language pair.

The translation process can also use additional context, including key descriptions, glossary terms, style guides, and task instructions.

When a custom AI profile is active, Pro AI can retrieve relevant examples from your translation memory or existing project translations and use them alongside the other context sources.

These features help the AI follow your project's terminology, established wording, and writing guidelines. However, the results still depend on the content, language pair, and quality of the context provided.

Use AI profile evaluation to compare translations generated with your custom AI profile, the base AI profile, and Google Translate using your own approved translations as references.

The results include similarity-based metrics and side-by-side examples, helping you validate the value of your custom profile and identify areas for improvement. Evaluation supports profiles based on tagged or reviewed translations and TM-based AI profiles.

Learn more in AI profile evaluation and proof of value.

AI translation usage is measured in translated words. Depending on the translation method and your plan, two usage metrics may apply:

These metrics are related, but they are not the same. An AI translation can consume an AI-specific allowance without necessarily consuming processed words. Standalone AI LQA tasks consume Pro AI words but not processed words, since they evaluate existing translations without generating new ones. AI scoring in Lokalise Expert does not consume additional AI words beyond those used for translation.

For example, a first-time translation into an empty target language counts toward processed words. However, retranslating an existing target translation without changing the base language does not consume processed words, even though the AI translation itself still counts toward the relevant AI allowance.

All usage is tracked at the team level across all projects.

You translate content into German and Spanish using Standard AI/MT. The resulting translations contain:

This results in 14 translated words of Standard AI/MT usage.

If these are first-time translations into empty target fields, they also add 14 processed words.

You translate content into four target languages using Pro AI. The resulting translations contain:

This results in 197 Pro AI translated words.

If these are first-time translations into empty target fields, they also add 197 processed words. If Pro AI instead retranslates existing target content without a base language change, the Pro AI usage is still counted, but the update does not consume processed words.

Our AI solutions are incredibly versatile and can handle a wide range of content types, including websites, documents, emails, marketing materials, user-generated content, and more. It can adapt to different industries and content formats, making it the perfect solution for businesses aiming to reach global audiences.

Yes. Pro AI does take locale and region into account, but only if your project language is configured with the correct regional code. When using Pro AI, make sure that your project language is set with the regional code you actually need.

This can directly affect the output. For example:

Yes, when using AI profiles. By default, Lokalise uses dynamic LLM routing to automatically select the most suitable model for each translation task. Alternatively, you can configure the base or custom AI profile to always use a specific supported LLM, such as GPT or Claude.

Creating a foundational LLM from scratch would be resource-heavy and slower to improve. Instead, Lokalise leverages the best-performing models available (e.g., GPT, Claude, and others) and enhances them with our proprietary orchestration:

In short: we focus on delivering quality and innovation, not reinventing the wheel.

Translation quality depends on several factors, including the language pair, content complexity, available context, and the terminology involved. There isn't a single accuracy percentage that applies to every project.

Additional context can make a measurable difference. In customer evaluations conducted between March and October 2026, across 26 languages, Custom AI Profiles with full context produced translations that exactly matched approved human translations 51% of the time, compared with 40% for the same models without context.

This measures exact matches, not overall translation accuracy or acceptance rates. A translation that differs from the reference may still be correct.

To assess results in your own projects, you can use AI Scoring and AI LQA to identify potential quality issues, and AI profile evaluation to compare different translation approaches against approved references.

For important or sensitive content, human review is still recommended.

Yes. Pro AI can translate specialized and technical content, especially when you provide relevant terminology, clear key descriptions, and appropriate writing guidelines.

You can also use custom AI profiles to provide examples of previously approved translations, helping Pro AI follow established terminology and wording.

However, specialized content can be particularly sensitive to small errors in meaning or terminology. For technical, legal, medical, or other high-risk content, we recommend reviewing translations with the appropriate subject-matter experts.

It depends on your project and task settings.

For segmented projects, AI Translation tasks translate empty segments by default and leave existing translations unchanged. To retranslate existing content, enable the Translate all segments option when configuring the task.

You can also enable Apply 100% translation memory matches. Lokalise will first look for exact matches in the translation memory and use them where available instead of generating a new AI translation. Applying a 100% TM match does not consume your Pro AI words allowance.

When you use our AI solutions, only the content required for the feature is processed. For example:

We may use this data for machine learning unless you withdraw your consent. This helps us develop new AI-powered features and improve existing ones.

Yes. Our AI solutions are designed with enterprise-grade security and strict privacy controls:

Provider policies: