Yes. WPML's active language is detected automatically and the bot reads the matching translation. ICL's language switcher and the standard WPML query filters are respected, so any post type WPML translates is grounded in the right locale. No separate WPML connector is required - the integration works through WordPress's standard locale signals.

Both work the same way. Any plugin that exposes the current language to WordPress is supported, including TranslatePress's slug-based routing and Polylang's separate-post model. The bot reads the locale from the request and pulls the matching content. WPML, Polylang, and TranslatePress have all been tested in production deployments.

It can, but you get better quality by giving it the human translation as context whenever available. AI translation is fine for casual chat replies, but for policy or pricing language - where exact wording matters - the bot should quote the human translation verbatim. The system prompt is the right place to enforce that distinction.

Yes. Multibot lets you assign a different provider or model to each locale-scoped bot. You might use GPT-4 for German formal contexts, Claude for nuanced French, and a cheaper model for casual locales where the cost-quality tradeoff is different. The configuration is per-bot, so changes don't affect other languages.

Yes. The active locale is stored on every log entry so you can filter and review independently. The Spanish team reviews Spanish chats, the German team reviews German ones, and translation gaps surface cleanly rather than getting buried. Per-locale QA is the only way multilingual AI stays good past launch.

No - one license covers every language on the same WordPress install. The only variable cost is API usage, which scales with traffic but not with the number of locales. That's why locale-aware AI economics work in WordPress where they don't on per-seat SaaS chat tools that charge per language or per region.

The widget supports RTL layout via CSS variables and the model providers all handle Arabic and Hebrew text natively. The locale detection picks up the RTL flag from WordPress, and the system prompt for that locale should be written in the target language to match the cultural register. RTL-specific QA is the same per-locale flow as any other language.

The bot can fall back to the default-language version with a system-prompt rule that tells it to translate the gist on the fly, or refuse and link to the existing language version. Most teams choose the second option for policy content - where exact wording matters - and the first for casual content where a translated approximation is acceptable.