Skip to content

WPML, Polylang, TranslatePress, PiedWeb AI Polyglot:which WordPress translation plugin should you choose?

By Updated at 18 min read

Putting a WordPress site into several languages is a solved problem, and solved well. WPML and Polylang have ruled the market for years. TranslatePress translates the visual output, Weglot and GTranslate host everything in the cloud. For 90% of sites, one of these tools is exactly what you need, and my advice fits in one sentence: pick the one that fits your case.

That leaves the 10% of cases where none of these tools lands right. It was by hitting that wall on one of them, a very specific store, that I ended up building PiedWeb AI Polyglot.

The fastest route: the quiz just below finds your plugin in 30 seconds, a straight answer on screen. If you would rather dig in, the rest compares the options one by one (the 3 main ways to translate a WordPress site, a summary table, a buying guide), before telling, for the curious, the story behind this plugin.

In 30 seconds: which plugin to internationalize your WordPress site?

Four questions, and the quiz points you to the solution that fits your case. It condenses the whole article: you will be able to check its verdict in the table and the “Which one should you pick” section right after.

Three ways to translate your site

Most people do not care “where” their translations are stored, and that is fair. What really matters is three things: who writes the translation, how much it costs over time, and whether Google finds your pages in the other languages.

There is still one detail that looks technical but turns into a real issue on the day it counts: who owns your translations? As long as everything runs smoothly, you never ask the question. The day you switch tools or stop paying, it is the only one that matters. It is what separates the three families below.

One clarification before we start: we are talking about translating your site content (your pages, your posts, your product sheets), not the buttons and menus of your theme, which are another tool’s job (Loco Translate), off topic here.

1. You keep control, the translations stay with you

With WPML, Polylang or TranslatePress, the translations live in your own site and belong to you. You (or a translator) hold the pen.

WPML is the most complete and the best known. It is also the only one paid from the start (39 to 199 € a year), with automatic translation billed on top. Very powerful, but it files your translations in its own database: the day you remove it, getting them back becomes complicated. (For Easy Digital Downloads stores, there is a dedicated module, edd-multilingual, but it still requires a WPML license.)

Polylang does the opposite, all in lightness: one version of your page per language, linked together. Free to start, and each language can even carry different content. In exchange, you manage as many pages as languages, and the store or automatic translation go through the paid options.

TranslatePress lets you translate by clicking straight on the text shown on screen. Very visual, ideal if you built your site with a page builder (Elementor, Divi).

The common thread: your translations are with you, you keep control.

2. An outside service translates everything, automatically

Weglot, Linguise and GTranslate do the work for you, on their servers. You connect the service, pick your languages, and the site translates itself.

Weglot is the reference: a few minutes and it is live, with good SEO. Linguise plays the same tune in a French version, a bit cheaper. Unbeatable for ease, but with two trade-offs: you pay a subscription that grows with the number of words and languages, and your translations are stored with them. If you stop paying, the site is no longer translated.

GTranslate has a free version, the little dropdown that translates the page on the fly. Handy as a stopgap, but those pages do not exist for Google: you will not gain a single visitor in the other languages. For real multilingual SEO, you have to move to their paid plan (around 300 € a year), hosted with them too.

The common thread: zero effort, but you rent your translations. The bill comes back every month and climbs with volume.

3. Translated by AI, but renting nothing

A third way, more under the radar: keep the translations with you (like family 1), have them produced by AI (the speed of family 2), but with no subscription or outside service, and nothing that locks you in. That is the approach of PiedWeb AI Polyglot, an open source plugin I actively help develop, designed mainly for stores backed by a developer. I cover it in the second part; for the comparison, remember: with you, open source, translated by the AI of your choice.

The comparison table of i18n plugins available for WordPress

PluginWhere your translations liveLock-inAuto-translationPrice
WPMLPlugin database, with youHighCredits (paid)39 to 199 €/yr
WPML for EDD
edd-multilingual
WPML module for Easy Digital DownloadsHighVia WPMLFree + WPML license
PolylangOne post per language, with youLowPaid option / manualFree / Pro 99 €/yr
TranslatePressPlugin database, with youMediumDeepL / Google (auto)Free / from 99 €/yr
GTranslateGoogle widget, or cloud (paid)TotalAutomatic (Google)Free (widget) / ~300 €/yr
WeglotWeglot serversTotalIncluded (automatic)Free limited / from 150 €/yr
LinguiseLinguise serversTotalIncluded (automatic)1-month trial / from ~150 €/yr
PiedWeb AI PolyglotInside WordPress, with youNoneThe AI of your choiceOpen source (GPL)

A word about the address of your translated pages. Most of these plugins let you choose: a subfolder (yoursite.com/en/), a subdomain (en.yoursite.com) or a separate domain. GTranslate’s free version does not create a real page (nothing Google can show in its results). PiedWeb AI Polyglot, on the contrary, accepts only one formula: one domain per language.

Which one should you actually pick?

None of these tools is “the best” in absolute terms, they answer different profiles:

  • Weglot (or Linguise, the cheaper French alternative) if you have the budget and want to translate everything in five minutes, with no technical maintenance and immediate SEO. You pay for that comfort (and per word: watch out on a big catalog).
  • TranslatePress if you want to see on screen what you are translating, ideal with a page builder (Elementor, Divi), while keeping your data with you.
  • Polylang for a blog or a media site “the WordPress way”, where you link posts by hand and sometimes want different content or design per language.
  • GTranslate (free) only for a courtesy translation, with no SEO ambition on the secondary languages.
  • PiedWeb AI Polyglot if you have an online store and a developer on hand: one site per language, a single shared stock, translated by the AI of your choice, with no subscription or tool that locks you in.

The last one, PiedWeb AI Polyglot, I built for a client, the woodrockclimbing.com store, then released as open source (GPL), so it can serve others. The rest of this article tells why and how it works, if the topic interests you.

The Woodrock store in several languages with PiedWeb AI Polyglot: one domain per language and its language-suggestion banner

Why maintain a new multilingual plugin

All these tools are mature and I recommend them without hesitation. If I still coded another one, it is not out of “not invented here” syndrome: a concrete project, the Woodrock Climbing store, fell right into a blind spot. The brief, for this WooCommerce store:

  • One domain per language (www.woodrock.fr, woodrockclimbing.com, de.woodrockclimbing.com…), not an /en/ directory. It is an SEO and marketing choice: each market gets its URL, its hreflang, its currency.
  • A single stock. An order placed on the English domain must draw down the same inventory as the French domain. Not two product sheets to sync by hand.
  • AI translation, with my own model and my own API key, with no proprietary credit system.
  • Clean SEO: correct hreflang, zero duplicate content in Google’s sense, sitemaps filtered by language.
  • Zero lock-in. If I uninstall the plugin in two years, the content stays usable.

Worth noting: four of these five needs (domain per language, SEO, AI translation, zero lock-in) hold for any multilingual site. Only the shared stock is specific to e-commerce. The WooCommerce part is therefore just an option, more on that below.

Tight budget and velocity. There is also a more down-to-earth reason, and it is the one that justifies maintaining it: across several domains and languages, stacking a WPML license and a per-word plan quickly costs more than a homemade plugin. And in the age of agents, writing then evolving this code costs a fraction of what it was worth yesterday, I add a feature in an afternoon instead of depending on a vendor’s roadmap… at the expense of a battletested plugin.

None of the solutions above ticked these five boxes together. Hence a different architecture.

The Master/Shadow approach

PiedWeb AI Polyglot rests on one principle: one main language (the master), N shadow languages (shadow).

Diagram of PiedWeb AI Polyglot's Master/Shadow approach: an FR master linked to EN, DE, ES and IT shadows sharing stock, images and reviews

You write in your main language. The plugin creates a linked copy (shadow) for each target locale. So far, it looks like Polylang. The difference is elsewhere, in two structural choices.

First choice: everything is stored in standard post meta. Not a single custom table. A shadow is a normal WordPress post: its translated title and content live in its own fields (post_title, post_content), and three meta values (_locale, _master_id, _translation_mode) link it to its main version. Let us be honest: at the post level, the database footprint is comparable to Polylang (one post per language), the gain is not there. It is elsewhere: no homemade table, so uninstall the plugin and your posts stay perfectly valid WordPress posts. Lock-in is zero by construction, not by promise.

And the whole core works without WooCommerce: domain routing, hreflang on every page, xhtml:link annotations and sitemaps filtered by language, RSS feeds per locale, a banner that suggests its language to the visitor, automatic hiding of links pointing to a draft, admin interface, translation pipeline. None of that depends on the store. For a blog, a brochure site or multilingual documentation, it is already complete, with no dependency.

Second choice: WooCommerce is only an optional layer. When the store is present, the plugin enables it, and that is where virtualization makes full sense: you share instead of copying. Stock, images and reviews are never duplicated, they are read from the main version at request time.

  • Shared stock: an order on any domain draws down the same inventory.
  • Aggregated reviews: a review left on the English domain is attached to the main product, and the average rating is computed across all languages. Your stars do not restart from zero in each market.
  • Pooled images: shadow sheets inherit the featured image and gallery of the main product, without duplicating a single media file.
  • Currency per domain: daily conversion of the main product’s prices via ECB rates, plus a Google Merchant feed per locale.

Without a store, these four lines do not activate. The rest runs everywhere.

One structural consequence remains: the main language is the superset. You can translate only part of your pages (a French page with no English version stays absent from the English domain), but never the other way around: you cannot have a page specific to a secondary market without first creating it in the main language. The plugin therefore targets a precise case, a strong main market that carries the structure and the SEO strategy, rolled out as is into the other languages, and not country sites with diverging content.

“Bring your own translator”: flat files, AI at the controls

This is the stance that gives the plugin its name: it translates nothing itself. wp polyglot translate is a router, not a translator. It is up to you, or to your AI, to provide the text.

On Woodrock, it does not go through the WordPress admin but through flat files. The database stays the source of truth, but you edit it like a Git repo. wp polyglot export dumps all the content into a readable tree:

wp polyglot export
#   polyglot-flat/
#   ├── _index.md          ← overview: id, slug, how many languages translated, which are stale
#   ├── product-100/
#   │   ├── fr_FR.html      ← main: front matter (title, slug, price) + HTML body
#   │   └── en_GB.html      ← shadow: the AI translates the body here
#   └── ...

An agent reads these files (cheap in tokens: Read, grep), translates the body of the main version into each locale, then writes the target-language file. On save, a hook resyncs the file back to the database under optimistic locking: each file carries an etag, a compare-and-swap token. If the server changed in the meantime, conflict (409), you replay the change on the fresh version. Never a silent overwrite. And a translation tweaked by hand stays locked (_translation_mode = manual): nothing overwrites it without --force.

And because an agent slips up, that lock is only half the safety net: wp polyglot lint checks every translation against its master (tags, links, leftover source text), wp polyglot doctor validates the whole multi-domain setup (hreflang, currencies, feeds), and internal links are auto-fixed after each import. The AI writes, the plugin verifies.

No proprietary credits billed per word, no black box: I choose the model, the prompt and the cost (the subject of my two years of hands-on AI).

This loop does not come from nowhere: it is exactly the one from Pushword, the flat-file CMS that powers this blog, designed from the start to be edited by an AI. PiedWeb AI Polyglot just carries that idea into WordPress: the content lives in files, the agent translates, the plugin syncs. You choose your workflow: editing in the WordPress admin, via the API or in flat files, depending on your preference and on whether you work solo or as a team.

Translating day to day: three workflows compared

Architecture is one thing; the daily gesture is another. Take a concrete task, translating a product sheet into English, and look at what it amounts to depending on the tool.

With Polylang, you open the product list, click the “+” in the English column, and WordPress creates an empty draft. You write (or paste) the translation in the Gutenberg editor, language by language, product by product. No learning curve, but everything is manual: no queue, no batch processing, no command line. On ten pages, ideal; on five hundred sheets, it is another job (and automatic translation goes through Polylang Pro with the DeepL API, or a third-party addon).

With WPML, you go through Translation Management: you select the content, assign it to a translator (you, an agency or a service), and you translate segment by segment in the Advanced Translation Editor, pre-filled by DeepL or Google, with translation memory and glossary. It is a real translation workshop, built for humans and providers. To step outside the frame, WPML can export XLIFF files to process in an external tool then reimport them, but that path is heavy and sits poorly with the built-in automatic translation. AI here means their credits, not your API key.

With PiedWeb AI Polyglot, no queue and no editor: as we saw, wp polyglot export, an agent edits the flat files, the hook resyncs. To translate five hundred sheets with the LLM of your choice, you script a loop; to fix a single one, you edit a file (or the sheet in wp-admin, which then locks to manual). Formidable for automating and scaling, unsuited if you specifically wanted to click in an interface.

At bottom, the right workflow depends on who is at the keyboard: you by hand with Polylang, human translators with the WPML workshop, a scripted AI with PiedWeb AI Polyglot. None is better in absolute terms, they do not address the same translator.

What the plugin does not do (on purpose)

Honesty calls for listing the trade-offs, because they are deliberate:

  • One domain per language, full stop. No subfolder /en/ routing. It is a choice that keeps the routing logic fast and the SEO model clean, but if you want yoursite.com/en/, it is not for you.
  • Some server config. You need to be able to point several domains at the same WordPress and set WP_HOME dynamically in wp-config.php. A developer or a managed host, not a two-click shared blog.
  • Single-site only. No WordPress Multisite. For this path (one site per language), the reference stays MultilingualPress.

If one of these points blocks you, refer back to the guide above: one of the other tools will be a better choice, and that is perfectly fine.

The plugin is open source (GPL), available on WordPress.org and on GitHub if you want to try it, read the code, open an issue, or burn a few thousand tokens to fork your own version. In the age of agents, people keep saying code is worth nothing anymore. In practice, a little, still.

AI, SEO and Web strategy session

Still choosing between multilingual architectures?

Content volume, team, SEO, budget, and technical dependency: test your constraints against the right model before migrating.

Validate my technical choice

€350 excl. VAT, up to 90 minutes

Email • RSS • Twitter

Get notified of new posts

Keep reading: