Customize WPML’s AI Translation – Context, Publishing Lifecycle, Spend Control, Edit-Learning

For the editor PTC plugs into and how translators see translations side-by-side, see Translation editor.

PTC That Sounds Like Your Site

The single highest-leverage thing you can do for translation quality is tell PTC what your site is about. A blog for hobbyist photographers reads differently than a B2B SaaS marketing site; PTC needs to know which one it’s translating for so the tone, vocabulary, and register of every translation match.

You give PTC three short pieces of information: your product or service name, what your site is about, and who your audience is. PTC reads them as context on every translation. A preview shows you the assembled description in plain English, so you can rework the wording until it matches your reality.

Sites that fill this in carefully get translations that read like an in-house translator wrote them. Sites that leave it blank get generic translations that often need cleanup later.

Worth doing once per site, whether it’s brand-new or has been running WPML for years.

Control Over the Publish Lifecycle

Different teams have different review processes for translated content. WPML supports three working models:

• Publish immediately. Translations go live as soon as PTC produces them. Fastest. Best for sites where AI quality is good enough to ship without a human pass.
• Hold for review. Translations stay as drafts until a Translation Manager approves them. Best for sites with editorial oversight on every piece of translated content.
• Publish and queue for later review. Translations go live straight away, but appear in a “needs review” queue that someone can work through afterwards. Best for sites that want speed and tracking: fast publication with quality assurance happening in parallel.

The right choice usually maps to whether there’s a dedicated editorial reviewer on the team. No reviewer > publish immediately. A reviewer who approves everything > hold for review. A reviewer who audits a sample > publish and queue.

PTC by Default – Or Another Engine If You Really Want One

PTC is pre-selected for new sites and is the engine WPML recommends. It’s the only engine that gets edit preservation, glossary support, and the learning loop described at the bottom of this page.

If you’d rather use DeepL, Google Translate, or Microsoft Translator, you can. They’re available as legacy options. They translate, but they don’t learn from your edits and they don’t apply your manual glossary. New sites should start with PTC; existing sites on legacy engines benefit by switching. The quality difference is measured: PTC scored 90.0 against DeepL’s 77.7 on the same source content, with about 18× fewer issues per page. See the full comparison.

The available-words counter shows how much translation capacity you have, with a path to top up if needed.

Spend Control – Who’s Allowed to Trigger AI Translation

On larger teams, AI translation cost can climb faster than expected as more translators ramp up. You can restrict AI translation to specific translators so spend stays predictable.

Administrators and Translation Managers always have access. Among other users with the Translator role, you decide who can trigger AI translation and who can’t.

Most sites don’t need this. Every translator on a small team should be allowed to use AI translation. The control is there for the cases where you need it. For the broader team-setup picture – who counts as a Translator vs Translation Manager, how external services plug in, and how the CAT-tool path works – see Translator workflows.

How PTC Learns From Your Edits

The settings above tell PTC how to behave today. This one teaches PTC to keep getting better from your corrections.

When you fix something in one of PTC’s translations – change a translated word, swap an external link for its target-language version, prefer a country-specific variant of a language, soften the tone – that correction can become permanent. PTC spots the pattern, and on every translation that follows, the same correction applies automatically. You teach the engine once instead of fixing the same thing on every page.

Four kinds of patterns get detected: terminology choices, link translations, country-specific variants of a language, and tone (casual vs. formal). When PTC spots one, it shows you a suggestion on the Improve Translations tab in the Translations area, where you accept or dismiss.

You decide how much PTC changes without asking. Three options exist on a spectrum:

• Review everything. All suggestions land on the Improve Translations tab; nothing changes until you accept. The most cautious setting and the default.
• Auto-apply selected categories. Trust PTC to apply, say, link translations and glossary terms automatically while still reviewing country and tone changes. Mix and match.
• Auto-apply everything. PTC applies all four kinds of patterns without asking. The most automated; useful when you’ve built trust with PTC and want to stop reviewing each suggestion.

If you’d rather PTC didn’t watch your edits at all, you can turn the learning loop off. The corrections you make then stay as one-off fixes – nothing carries forward.

This loop is PTC-only. On the legacy engines, every edit is a one-time fix; nothing learns, nothing carries forward.

For the full explainer, see How PTC learns your brand voice and translates like you do.