What's New at Crowdin: November 2025

This month, Crowdin launches tools to make localization faster and more reliable: AI Pipeline, a next-gen AI workflow that self-corrects translations, ensures consistency, and follows glossaries and style guides more reliably; improved Global Search in Crowdin Enterprise; Website Context Viewer for better context when translating headless CMS; Volunteer Certificates to recognize contributions; and platform updates. Plus, new episodes of The Agile Localization Podcast feature Jan Amann (i18n expert, creator of next-intl) on how internationalization shapes product architecture, and Shashi Bhushan (Localization Workflow Strategist) on designing AI-enhanced localization workflows that remain human-centered.

AI Pipeline: Next-Gen AI for Localization

Internally, we called this feature AI Integration Gen 4. Across the previous three generations, we learned what makes AI more productive for localization tasks. Now we’re ready to release the result: the AI Pipeline.

Why we built it this way:

• Gen 1 worked almost like a basic MT connector. Crowdin sent a batch of strings to an LLM, took whatever came back, and saved it as the translation.
• Gen 2 is our current native AI integration. We added strong RAG to improve quality, built automatic loops to revert translations when they fail Crowdin’s QA checks, introduced prompt-engineering tools, integrated MCP tools in this month’s release, and expanded with features like AI Proofreader.
• Gen 3 took a different path — Agentic AI. The goal was to let the AI figure out, on its own, how to achieve the best translation quality. It performed RAG, iterated on translations, and automatically refined results. In the lab, it worked extremely well. In production, though, it’s slow, costly, and demands major workflow changes that the market isn’t ready for (for example, linguists running pre-translation instead of managers).

Based on this experience, we now have: 

Gen 4 — AI Pipeline. The fundamental principle is this: AI models are capable of producing the translations you need. The reasons they fail on the first try are usually:

• They lack context. There are lots of ways to translate certain text “correctly”. LLM just does not know what kind of translation would work for you. (no Style Guide, TM, Glossary, project metadata, and more).
• LLM hallucinates, still.
• Prompt overload: LLMs struggle to translate a large file and follow complex prompt instructions simultaneously.

Point #1 is a permanent industry challenge; Crowdin will continue working to give you better context extraction and management tools.

Points #2 and #3 are exactly what AI Pipeline solves. If you’ve been using the native Crowdin AI integration, you might have noticed the model occasionally ignoring a glossary term or skipping an instruction. As the number of instructions in a prompt increases, the probability of the AI ignoring one of them grows.

We believe this is how LLMs should be utilized:

Step 1: Share context and your request with the AI (similar to our Gen 2 native integration).

Step 2: Add specific steps asking the LLM to verify the output against your constraints. Let the AI correct itself. (e.g., “Check if terminology is correct,” “Verify adherence to Style Guide,” or “Check if translation fits UI constraints”).

We also added Language Specific Prompts, a highly requested feature from our native integration.

One more addition: File Consistency Check. This is critical for continuous localization. For example, if you add a new paragraph to the middle of a help center article that has been translated previously, you can add a pipeline step that checks whether the new translation “fits” the tone and flow of the existing translations in that file.

Technically, all of this was possible in the native integration via AI QA checks, for example, but the configuration wasn’t straightforward. While Agentic AI (Gen 3) can also loop to verify itself, it acts as a “black box”; we can’t trace or control exactly what happens under the hood.

AI Pipeline is the deterministic middle ground. It is simple to configure and offers predictable latency and costs (you decide exactly how many steps to run).

There is more to this than we can cover here. We’ve found that different content types and target languages require different pipeline architectures for maximum efficiency. For example, when localizing UI with screenshots, context extraction from the image should be a discrete step in the pipeline.

We will publish more blueprints on how to use the AI Pipeline efficiently soon. In the meantime, please experiment and let us know what you think.

Global Search: Improvements

You asked for it, and we made it happen. The global search at Crowdin Enterprise has a fresh, user-friendly design that has/helps you to:

• Context everywhere: See your search phrase, search context, and additional info depending on whether you’re searching source strings, translations.
• Recent searches: Quickly revisit past queries.
• Find source strings instantly with multiple search types and project filters.
• Search translations across languages.

Results that click through:

• Translations and source strings highlight your search term and link to the editor.
• Files are clickable and open in the editor; project names link to their pages.

Plus, “All Records” is here for dictionaries — TM and glossary entries now take you straight to the right results with filters applied automatically.

New Integration: Crowdin for Canva

Canva recently introduced native AI translation, yet demand for a Crowdin integration hasn’t slowed.

Enterprise teams need a way to ensure translations meet their specific quality criteria. This means translations verified by professional linguists who strictly adhere to style guides, glossaries, and brand terminology — something that is hard to achieve when asking linguists to work directly inside a design tool.

We are excited to unveil the first version of the Crowdin for Canva integration. We know how many businesses rely on Canva for their design workflows. Now, those designs can be made multilingual with Crowdin.

• When requesting translations, the integration allows you to attach screenshots of the designs, so linguists understand exactly how the text fits the visual.
• Push text to Crowdin, wait for high-quality translations, and pull them back directly into your Canva design elements.
• Apply translations to specific pages to check for layout breaks before exporting.

Attachments in Comments

We know that sometimes text just isn’t enough. You can now attach images, videos, and audio files directly to comments in the Crowdin Editor. This is the fastest way to share context---whether you are a project manager or a linguist.

Here are a few use cases we envision:

• Did a button break because the German translation was too long? No need to write a paragraph describing it — attach a screenshot of the broken UI.
• Often, text is part of a complex animation in a computer game. A linguist or manager can attach a screen recording to show exactly how the text appears and how long it stays on screen.
• Creative translations often rely on cultural context. Attach a movie poster or a pop-culture meme to explain a specific pun or the mood of a phrase.
• If you are localizing game audio scenes, the best way to convey the right intonation is to attach the original audio recording for the linguist to hear.

We’re sure there are many more cases where extra context can save the day.

P.S. Speaking of comments in the editor: We are considering adding reactions (emojis) to comments. Is this something you would find useful? Let us know if you want them and how you’d like them to work!

New App: Website Context Viewer

Localizing websites was somehow an “easier” task than UI localization, for example, because the pages carried their own context (usually as HTML or Markdown). A linguist could see the entire page or a big part of the page they are working on. That is not the case for websites built on a Headless CMS. Localizing headless CMS is more like localizing software UI: linguists are often presented with tiny content blocks in isolation. The smaller the text chunk, the harder it is to figure out the best translation.

Website Context Viewer is our latest tool to help address the problem.

Last month, we introduced Website Context Extractor. It solves a related but different problem: it helps AI understand context by using SERP API to locate where text appears across your site. Automated extraction is helpful, but not perfect – sometimes two blocks have the same text but mean different things. Plus, Website Context Extractor produces textual descriptions as artifacts that might overload linguists, and it uses AI to extract context, which still poses a risk of hallucinations. The new Website Context Viewer provides the opposite type of context: explicit, visual, and human-friendly. Project Managers (or the API) can pass URL information during content upload. When that’s available, the app renders the actual live page directly inside the Crowdin Editor. The matching works like a simple “Cmd+F” when the page loads, highlighting the text. But even just seeing the exact page where the text appears gives linguists much better context and improves translation quality.

Integration Note: Not every Headless CMS makes it easy to link text blocks to URLs—some need engineering work. Others are simpler. This app comes with a WPML plugin that automatically provides the context so linguists can see previews when translating WordPress sites. If your linguists feel like they’re “flying blind” with a Headless CMS, ask your engineering team if they can map content blocks to URLs during upload. If they can, this app fixes the context issue.

In short:

• Website Context Viewer – ideal for people (accurate visual context, no AI guessing).
• Website Context Extractor – ideal for AI (helps it understand your site without burdening linguists).

Other New adn Updated Crowdin Apps

• WPML Companion: automatically processes WPML XLIFF files on import, extracting the original website URL and adding it to each string’s context so translators can see where the content appears. (Created to be used with Website Context Viewer)
• Kustomer: app to automate importing knowledge base articles from Kustomer into Crowdin, translate them using TM/MT/AI, then export back; translations are pushed as drafts and require manual publication. Make sure to configure the source language before using the app.
• Contentful: Data Center Selection: A New update lets you choose a data center during authorization for flexibility and regional compliance.
• Bashini: Indian-language MT for Crowdin; adds Bashini as a machine translation provider specializing in Indian languages, including those often missed by other MT providers. Learn more on their website.
• Pin Apps: now works strictly at the project level, allowing quick access per project without affecting the entire account.
• SCIM via Azure AD: Pending User Deletion Control: new settings allow controlling the removal of pending users during Azure AD synchronization.
• Simple Term Extractor: New Iteration includes updated UI, dropped the Machine Learning method; after term extraction, a full-screen grid shows all term details (type, part of speech, gender, etc.) with bulk actions available.

Early Access to New AI Tools

You can now access both Crowdin App Builder (vibe coding) and Video Dubbing. They’re still in beta, but you’re free to explore them and start applying them to your localization projects right away.

The Automator is still under development. Keep an eye on our updates to know when it becomes available.

Vendors Marketplace Refresh

It was about time to refresh our partner marketplace. The vendors marketplace has been updated and can now be found under Store > Vendors. We’re excited to see more translation agencies offering a “pay for actual AI translation edits” model, which makes AI translations more affordable without compromising quality and rewards managers who provide context and fine-tune AI setups.

Crowdin Certificates for Translation Volunteers

While Crowdin is now a mature enterprise localization management platform, we still host hundreds of thousands of crowdsourcing projects. We never stop marveling at the incredible open projects managed with us. Minecraft, for example, is currently translated into 180 languages by a crowd of nearly 44,000 volunteers. There are very few things on our planet translated into so many languages with such high participation. 

Today, we are introducing a new app for Crowdin Enterprise designed to help organizations recognize volunteer contributions. The Volunteer Certificates app allows you to issue certificates for tenure, word count, and proofreading efforts of your volunteers. It is designed with flexibility in mind, so a manager can issue certificates for translators or configure the app to allow volunteers to generate their own certificates in a self-service mode. We are still considering what effort recognition should look like on Crowdin.com, where most open projects live.

Introducing the Updated Role: Translation Requestor

We’ve refined the Developer role so it better reflects how teams actually use it. In many enterprise setups, there are people who don’t work in Crowdin every day but regularly need to request translations (marketing, product, support, and similar teams).

This updated role now functions as a Translation Requestor. It lets colleagues quickly submit a page, document, or any other content for translation without requiring full project access.

If you have team members who “just need this translated” and don’t need deeper involvement in Crowdin, this role is designed exactly for them.

Other Platform Updates

• Showing Existing Translations in WYSIWYG We’ve improved how WYSIWYG displays existing translations. Vendors now see the translations already provided on the client side, which gives them better context. Linguists also see non-editable strings in grey, helping maintain consistency across the document. This context should have always been visible, and now the experience is aligned with that expectation.
• Main Branch Protection You can now fully lock the main branch in string-based projects. All write actions are blocked in the editor and the Strings tab --- no file uploads, deletions, string edits, pre-translation, tasks, or translation updates (including approvals and votes). Only read-only access remains. All changes to source or translations must now come through feature branches and be merged into master in the Crowdin UI.
• A new AI option is available: the Gemini 3 Pro preview model.
• Word Count for Selected Strings: You can now view the word count for any selected strings directly in the Editor.

The Agile Localization Podcast: Latest Episodes

This month, we’ve released two insightful episodes of The Agile Localization Podcast, diving deep into the technical and strategic sides of building global products.

• Internationalization Is Architecture, Not Translation: A Deep Dive with Jan Amann. In this must-listen episode, i18n expert and creator of next-intl Jan Amann argues that Internationalization (i18n) is not just about translating strings---it’s about product architecture, affecting how you model language, currency, routing, and content pipelines. Learn how to avoid costly scaling mistakes.
• How to Build Human-Centered AI Workflows in Localization with Shashi Bhushan. We sat down with Localization Workflow Strategist Shashi Bhushan (ex-Marvel, Google, Amazon) to discuss the future of AI in localization. Shashi shares practical strategies on designing AI-enhanced workflows that remain deeply human-centered, ensuring AI agents handle repetitive tasks to free up linguists for quality and creativity.

External tools

This month, we released new versions of:

• Figma plugin version 89, version 90, version 91
• JS API Client 1.49.0
• Java API Client 1.30.1
• Ruby API Client 1.13.0
• Go API Client v0.18.0
• .NET API Client 2.41.0 
• Android SDK 1.16.0
• Flutter SDK 0.8.1