This episode of the Jukebox podcast features Leonardo Losoviz, a WordPress developer behind Gato GraphQL and Gato AI Translations for Polylang. Nathan Wrigley and Leonardo discuss how AI has changed the cost, speed, and workflows for translating WordPress sites, practical considerations for site owners, and what to watch for as the platform evolves.
Who is Leonardo Losoviz
Leonardo has worked with WordPress since 2012. He created Gato GraphQL and more recently focused on tools for non-developers, launching Gato AI Translations for Polylang to make translating sites with AI easy and affordable. He has spoken about the hidden pitfalls of WordPress translation and explains how to turn the idea of translating a site into a reliably executed plan.
Why translate your site
If legal requirements make translation mandatory, there is no choice. But even when not compelled by law, Leonardo argues it is usually worth doing: translating opens access to new audiences and markets. Because AI has dramatically reduced both cost and time, translation is now feasible for many more languages than before, allowing sites to reach visitors around the world with minimal expense.
How AI changes the translation equation
AI-based translation quality is now very high for general content. Leonardo estimates that AI handles roughly 99% of straightforward translations well, though specialist or technical material may still need a human review. The economics have shifted: instead of paying a professional translator for many hours, you can use AI for the bulk of the work and engage a human only to polish ambiguous or domain-specific terms. Token costs from providers like OpenAI, Anthropic, or Gemini are now a fraction of historical human-translation rates.
User experience and workflow
Gato AI Translations is built as a Polylang add-on. Its workflow is simple: from the posts list you can bulk-select content and ask the plugin to create translated versions in chosen languages. For Polylang, translations are created as separate posts — one per language — which has several benefits: the translated pages can be cached, exported, and served directly, and Polylang links translations together so search engines understand the relationship via hreflang tags.
This approach differs from some plugins that keep one post and translate strings at runtime. Both strategies have trade-offs. Static translated posts are fast and easier to cache; runtime string translation can be more dynamic but may be more resource-intensive.
Minimize wasted effort: finalize before translating
A common waste of time and money is repeatedly translating content that is still being edited. Leonardo strongly recommends finishing and proofreading the original post before initiating translation. He maintains a checklist covering typos, correct heading structure, image captions, alternative text, and embedded media — all things to verify so translations are accurate and do not require reruns.
Handling images and non-text content
Translating images and embedded charts is technically possible but often impractical. Leonardo cautions against duplicating every image for every language because that creates storage and maintenance overhead. Instead, use language-agnostic images when possible and avoid embedding text inside images. For captions and media metadata, Polylang supports translating the attachment metadata without duplicating the actual image file: the attachment has separate entries per language pointing to the same media file.
Quality control and interaction with AI
When using an automated translation plugin, the AI returns a translated string that you can add directly to the translated post. There typically isn’t a conversational back-and-forth unless you use an interactive AI interface like ChatGPT. Leonardo foresees deeper integrations where AI can flag uncertain translations in the editor UI and prompt human reviewers to check or refine specific phrases.
WordPress, collaborative editing, and future directions
WordPress 7.0 introduces foundational pieces that will enable better AI interaction and collaborative editing (phase three of Gutenberg). Leonardo expects this to enable workflows where an AI assistant and human editors can exchange comments in the editor, similar to Google Docs. That infrastructure could allow the AI to mark low-confidence translations and surface them for human review directly within the post editor.
Phase four of Gutenberg, which would more fully integrate translation features, remains further in the future. For now, the community can build on 7.0’s AI and collaboration foundations to create richer translation workflows.
SEO and URL structure
Polylang offers options for languages in URLs, either language-code paths like mysite.com/fr/slug or subdomains like fr.mysite.com. Because translated posts are separate entries and Polylang provides hreflang relationships, search engines can correctly serve the language-appropriate page to users. This setup keeps translations SEO-friendly while maintaining clear mapping between source and translated content.
The arms race effect
Leonardo points out a pragmatic downside of low-cost AI translation: when translation is cheap and easy, everyone will do it. That reduces the competitive advantage of being multilingual, making translation effectively something you do to avoid falling behind rather than to leap ahead. In short, AI raises the baseline for internationalization and turns translation into a necessary investment for many businesses.
Practical takeaways
– Use AI for the bulk of general-content translation, but hire a human to review specialized terminology.
– Finalize and proof your original content before translating to avoid wasted token costs and repeated translations.
– Consider a plugin strategy that fits your site: static translated posts (like Polylang) are fast and cache-friendly; runtime string translation suits other use cases.
– Keep images language-agnostic when possible, and translate image metadata rather than duplicating files.
– Watch WordPress 7.0 and the evolving AI connector and collaborative editing features, which will enable smarter editor-based workflows and AI-assisted review.
Where to find more
The full episode and links are available via WP Tavern podcast pages and WordPress.tv for Leonardo’s talk. Search for the episode on wptavern.com/podcast or look up Leo Losoviz for additional resources, his plugins, and the checklist he mentions for preparing content before translation.
If you are considering making your site multilingual, the combination of low-cost AI translation and careful review workflows makes it far easier to reach new audiences — just be mindful of quality checks and the ongoing need to keep content finalized before translating.