Leonardo Losoviz, creator of Gato GraphQL and Gato AI Translations for Polylang, joined WP Tavern’s Jukebox to talk about how AI has reshaped multilingual WordPress sites. His main message: if you can reach more users by translating, do it — and AI now makes that fast, cheap, and reliable, while still leaving room for human oversight when needed.

Why translate? Legal and moral reasons aside, Leonardo argues the practical business case: translated sites reach more visitors and expand markets. Historically, translations were expensive and slow — hiring professionals, managing copy, and integrating translated strings into WordPress took time and money. Today, AI can produce high-quality translations for cents (you pay token costs to providers like OpenAI, Anthropic, or Gemini). That means translating to multiple languages is affordable and can be done in bulk.

Quality and when to use humans
AI translations are excellent for general content; Leonardo estimates they’re accurate the vast majority of the time. However, for technical, industry-specific, or legally sensitive text you still want a human translator to review and correct domain-specific terms and acronyms. A pragmatic approach is to use AI for the bulk of translation and hire professionals only to check and polish tricky spots, reducing time and cost from hours to minutes.

How Gato AI Translations works (and the Polylang approach)
Leonardo’s plugin wraps AI translation services into a workflow for Polylang users. Polylang’s model creates separate posts for each language (rather than keeping one post and translating strings at runtime), which has advantages: translated posts are static entries that can be cached and exported, and runtime string lookup doesn’t incur translation costs on page load.

Typical flow:
– Finalize your source post in the origin language.
– From the posts list you can bulk-select and trigger “translate” actions. The plugin duplicates the post for each selected language and fills in translated content, categories, tags, featured image metadata, etc.
– URLs can use language codes (e.g., /fr/slug) or subdomains (fr.example.com), and Polylang adds hreflang metadata so search engines understand translation relationships.

Managing media and non-text content
Polylang associates translated metadata with the same physical media file rather than duplicating the image file itself. That means captions, titles, and alt text can be language-specific without multiplying storage. Leonardo warns against blindly replicating images with embedded text into every language; instead prefer language-agnostic images or overlay translated text in the page builder so you only translate text, not the image itself.

Practical checks before translating
To avoid wasted token spend and repetitive work, don’t translate until the origin post is final. Common issues that force retranslation include wrong headers, typos, embedded text in images, and third-party embeds (e.g., videos in another language). Leonardo keeps a checklist: finalize headings, captions, alt attributes, and any embedded content before triggering bulk translation.

UI and future WordPress integration
Right now, Leonardo’s plugin provides a simple “translate” action from the posts list that creates translated posts automatically. Looking ahead, WordPress 7.0 brings foundational pieces that will improve AI workflows: an AI Connector for integrations and Phase Three collaborative editing (Google Docs–style comments) inside the editor. Those features could enable AI agents to join editor conversations, flag uncertain translations, or leave inline notes for human reviewers — making review loops smoother.

Different translation strategies
Plugins approach multilingual sites differently: Polylang creates separate posts per language; WPML can translate strings at runtime; others like TranslatePress or Weglot have their own models. Each strategy has trade-offs (performance, cacheability, complexity). Leonardo prefers Polylang for its static entries and caching friendliness, and notes Polylang’s free version is sufficient for many sites (Polylang Pro adds extra features and previously offered DeepL integration). However, Leonardo believes modern AI models outperform older machine-translation services like DeepL for many use cases.

SEO and discoverability
Properly implemented, translations improve discoverability in other languages. Polylang’s hreflang handling tells search engines that pages are translations of one another so users searching in specific languages see the matching pages. Translating into new languages can boost traffic from new regions — but Leonardo also warns of the “arms race” effect: when translations become trivial, everyone will do them, and you may only be keeping up rather than gaining a long-term competitive advantage.

Costs and competition
AI reduces per-article translation costs to cents, so the barrier to translating to many languages is low. That democratization is good but means the strategic edge is diminished as competitors adopt the same tools. Still, for most businesses wanting international reach or compliance, translating is now a practical expectation rather than an expensive experiment.

Tips and takeaways
– Finalize your source content before translating to avoid repeat costs.
– Use AI for the bulk of translation; hire humans to verify high-stakes or technical content.
– Prefer language-agnostic images or overlay translated captions to avoid duplicating image assets.
– Choose the plugin strategy that fits your site: separate posts (Polylang) vs runtime strings (WPML) vs hosted services.
– Watch WordPress’ AI Connector and collaborative features — they’ll enable richer editor-AI collaboration and smoother review flows.

For more details, Leonardo’s talk (Invisible Gotchas of WP Translation) and related links are available in the episode notes on WP Tavern’s podcast page. If you want an efficient, affordable multilingual site today, AI-based translation plugins make it possible — but plan your workflow carefully and keep humans in the loop for the cases that matter most.

This episode of the Jukebox podcast (WP Tavern) features Leonardo Losoviz, a long‑time WordPress plugin author behind Gato GraphQL and Gato AI Translations for Polylang. Leo joined Nathan Wrigley after his WordCamp Asia talk, “The Invisible Gotchas of WP Translation,” to explain how AI is changing multilingual sites and to share practical advice for site owners and developers.

Why translate at all?
Leo separates two obvious cases: legal obligations and opportunity. If you’re compelled by law to provide content in another language, you must do it. If you aren’t required, translating still makes sense: it enlarges your audience and can drive more visitors. With AI making translation cheap and fast, the cost/benefit calculus has shifted—if your potential audience speaks other languages, it’s usually worth translating.

How AI changes the equation
AI models now produce very high quality translations for most general content. That means the heavy lift of paying human translators for full translations is often no longer necessary. Instead, humans are best used to review edge cases: technical jargon, industry‑specific terminology, ambiguous phrases, or legal copy where absolute precision matters. The cost model is now token‑based (fractions of a dollar per translation), so you can translate into many more languages for very little money and only spend on human review where it’s needed.

Gato AI Translations + Polylang: workflow details
Leo’s Gato AI Translations plugin integrates with Polylang. Key points about how this approach works:
– Polylang creates separate posts for each language (one English post produces a French post, a Spanish post, etc.). That static approach helps caching, static exports, and performance compared with translating strings at runtime.
– Gato enables bulk translation from the posts list: select a post, choose “translate,” and the plugin will duplicate and translate content to the target languages you’ve configured.
– Media and taxonomy translations: Polylang keeps language metadata for categories, tags, and media attachments. Media files aren’t duplicated on disk; translations create separate attachment entries with translated titles/captions but reference the same physical file.
– URLs and SEO: Polylang lets you serve languages via language codes (site.com/fr/slug) or subdomains (fr.site.com). It also outputs hreflang relationships so search engines understand which pages are translations of each other.

User interface and future WordPress features
Currently Leo’s plugin handles translation without interactive AI prompts—translation returns a completed string which you insert into the translated post and then edit in WordPress as needed. Looking ahead, WordPress 7.0 introduces two relevant pieces:
– An AI Connector API that opens standardized access to AI services from within WordPress.
– Collaborative editing (phase three of Gutenberg), bringing comments and chat‑like interactions inside the editor.
Together these could let AIs flag uncertain translations, add sticky comments, or participate in editor conversations—so a translation could be automatically created and then annotated with AI suggestions that a human reviewer can accept or correct.

Practical checklist and tips
Leo emphasizes not translating unfinished posts. Every translation run costs tokens (and creates work if you must retranslate). Before translating, validate your source content: fix typos, set correct headings, ensure images have proper alt text, and confirm any embedded media is appropriate for the target audience. He keeps a pre‑translation checklist to avoid wasted effort.

Handling images and non‑text content
Technically AI can extract text from images and regenerate localized versions, but Leo advises caution. Doubling or tripling image assets for each language quickly becomes unmanageable. Prefer language‑agnostic images or use overlay text elements that are translated as normal text. Polylang already stores translated captions and metadata without duplicating the image file itself.

Performance and architecture choices
There are two common strategies among multilingual plugins: create separate posts for each language (Polylang approach) or store one post and translate strings at runtime (some other plugins). Static duplicate posts are often better for caching and scale because they avoid runtime string lookups and expensive translation calls on each request.

The competitive landscape: an arms race
A major observation: because AI makes translation cheap and easy, everyone can do it. Early adopters once gained a big advantage, but as adoption becomes widespread, translating becomes a baseline expectation rather than a unique differentiator. That may still be reason enough to translate—to avoid falling behind competitors.

Final thoughts
AI has dramatically lowered the barriers to multilingual WordPress sites: faster, cheaper translations with quality sufficient for most content. The best pattern now is to automate the bulk translation with AI, use human reviewers for critical or technical content, finalize source posts before translation, and choose a plugin/architecture (like Polylang’s separate posts) that suits your caching and SEO needs. Emerging WordPress features—AI connectors and collaborative editing—promise tighter AI/human workflows inside the editor, making translation and review even smoother.

To hear the full conversation and see Leo’s WordCamp Asia talk, look for the Jukebox episode on wptavern.com/podcast or search WP Tavern in your podcast app.