Multilingual strategies in Drupal in the GenAI era

Launching a multilingual website is a strategic decision that opens doors to new markets, increases user trust, and strengthens your brand's identity globally. At the same time, managing a multilingual digital ecosystem has always been a balancing act.

Anyone who has managed an enterprise platform knows that the challenge doesn't lie so much in the translation technology itself, but in orchestrating the processes: exponentially growing content volumes, review cycles that slow down time-to-market, data governance, and operational costs.

Today, Generative AI (GenAI) has brought brutal acceleration to this scenario. The promise of instant, near-zero-cost translations is seductive, but brings with it new risks: loss of brand consistency, hallucinations from probabilistic models, quality levels not always up to par from generalist models, and the difficulty of maintaining rigorous editorial control over thousands of auto-generated pages.

At SparkFabrik, we work daily on complex Drupal-based projects, serving clients who manage large digital ecosystems, from universities and public institutions that need to publish important tenders and official informations, to enterprise companies with broad product portfolios and global presence.

For these organizations, linguistic precision isn't an aesthetic detail, it’s not just to “look and sound good”: it's a requirement for brand identity and reputation (and, in certain contexts, also for compliance).

In this scenario, Drupal confirms itself not only as a solid choice, but as the enterprise CMS best positioned to transform the GenAI revolution into a concrete operational advantage, even for multilingual needs, and without sacrificing quality.

The Multilingual Challenge

Let's address the central issue right away: multilingualism is not a trivial matter of translating words from language A to language B.

If it were that simple, a Google Translate plugin would suffice. Multilingualism is strategy. It's international technical SEO, it's cultural adaptation (localization), it's evolutionary maintenance of content that must remain synchronized over time.

In short, having a multilingual presence is a multifaceted strategic decision for the brand. And in this area, the choice of Content Management platform is the founding decision for any internationalization strategy.

Drupal stands out in the enterprise CMS landscape, excelling in the structural management of these complexities thanks to its architecture that conceives multi-language as a native attribute of data.

At the same time, a major "Achilles' heel" of any multilingual system has always been the automation of translation workflows, in terms of balancing costs and quality. Traditional methods, such as sending files via email to agencies or using old-generation automatic translators, are now obsolete for the pace and quality levels required by today's market.

Our thesis is clear: the only viable path for modern organizations is the intelligent use of GenAI, but rigorously accompanied by strategic human control.

Why has multilingualism become a current topic again?

To understand the scope and relevance of the multilingual content topic, it's necessary to look at the historical context we're experiencing.

First of all, in the digital landscape formed in recent years, we've witnessed a convergence that has ultimately led to an explosive increase in the quantity of content and translations:

• The maturation of Generative AI (GenAI) technologies.
• An explosion in digital content production, fueled mainly by GenAI which has empowered teams of all sizes.
• A consequent increase in the demand for localization of produced content (a recent report indicates a surge in enterprise translation demand of 30% annually). Moreover, translations fit into a more general market trend towards consistent and personalized content for end users (we discussed this in the context of omnichannel with Drupal).

But it's not just about content quantity; there's an equally important increase in pressure on speed. Marketing campaigns, communications, and other content must be released simultaneously in all languages. There are no more weeks of time for manual localization.

Third, the need for quality. Enterprise organizations face a crossroads: continue to rely on manual processes, now unsustainable in terms of costs and time, or embrace automation while risking compromising brand reputation with low-quality translations (not just literal, but in terms of brand tone-of-voice).

GenAI can represent the "Holy Grail" that balances quantity, speed, and quality in this area. At the same time, however, there's the need for control: in a world where content is machine-generated, editorial governance becomes the last bastion of brand identity. Both fine-tuning AI systems according to each brand's identity and human supervision and review become essential.

It's also worth considering the impact of GenAI on editorial teams: content management teams should not be replaced, but empowered, freeing them from repetitive tasks to focus on creativity and qualitative supervision, including in terms of localization (in this sense, Drupal fully embraces this approach to AI).

Last but not least, making (or maintaining) a brand multilingual is a strategic decision that opens doors to new markets and strengthens the brand internationally or globally. The interest for brands in this strategy is absolutely evident, now made significantly more accessible to organizations of all sizes thanks to GenAI.

Drupal and multilingualism: what works, what's changing

Drupal needs no introduction when it comes to multilingual capabilities; in fact, Drupal's centrality in the enterprise sector is largely attributable to its architectural maturity regarding multilingual data structures.

Unlike other CMSs that require heavy plugins to manage translations, Drupal handles multilingualism at the Core level. This means that every entity (from nodes to content blocks, from taxonomies to menus) is natively translatable.

However, the ability to store translations is useless without an efficient operational process to create and manage them. This is the domain of modules like the Translation Management Tool (TMGMT).

Let's analyze in more detail the multilingual aspects in Drupal's Core and in TMGMT.

Multilingual content and localization in Drupal Core

Drupal incorporates multilingualism into its main Core, at the deepest level of its application framework. This means that robustness and scalability are guaranteed, not depending on third-party plugins that can break at any moment.

More specifically, Drupal integrates language support at the Entity and Field level. Every content element is an entity (be it a page, a block, a taxonomy term, a menu, or a media asset). The native translation system allows creating language variants for each entity while maintaining a single unique ID.

At the same time, you can configure which specific fields of content must be translated (e.g., product titles and descriptions) and which should remain unchanged (e.g., product codes, numeric technical specifications, global images). This not only optimizes translation costs by reducing word volume but also ensures the integrity of technical data across markets.

Drupal's linguistic architecture therefore operates on four levels:

Furthermore, organizations can choose whether to maintain a symmetric structure (every page exists in all languages) or asymmetric (specific content for local markets), managing everything within a single instance or through a centrally governed multisite architecture. The logic that determines which variant to serve to the user is also configurable: URL prefixes (e.g., /it/), top-level domains, authenticated user preferences, or browser settings.

Equally important, Drupal's granular permission management is a fundamental aspect for more structured organizations, allowing precise role-based permissions and review, approval, and publication pipelines for each language or region to be set.

In short, Drupal supports flexibility essential to support the most complex international product, content, and SEO strategies.

Drupal's Architectural Superiority compared to competitors

When compared with alternatives like WordPress or Adobe Experience Manager (AEM), Drupal's native architecture offers indisputable business advantages.

• Comparison with WordPress: WordPress typically requires plugins like WPML or Polylang. These often store translations as separate posts linked by metadata, which can lead to database bloat and query inefficiency at scale. Drupal's entity-based translation stores translations within the same entity record, optimizing performance, simplifying API queries, and ensuring greater data consistency.
• Comparison with Adobe Experience Manager (AEM): While AEM offers robust "Language Copies," it comes with high licensing costs and often requires heavy customization for complex workflows. Drupal offers comparable enterprise capabilities (granular permissions, workflow integration, multi-site management) without licensing fees, significantly reducing Total Cost of Ownership (TCO) and allowing budget to be reinvested in innovation and content quality.
• Comparison with Headless CMS: The evolution of digital architectures towards "Composable" and "Headless" models has made a CMS's ability to act as a central repository for multilingual content even more critical. Drupal, thanks to its API-first approach, natively exposes translated content via JSON:API and GraphQL. Importantly, data is exposed in a structured format consumable by any frontend (React, Vue, Angular), facilitating omnichannel distribution without complex middleware for language logic management.

TMGMT as a workflow orchestrator

While Drupal Core provides the ability to store translations, it doesn't fully manage the operational translation process. This is where the Translation Management Tool (TMGMT) comes in.

Used by over 10,000 high-traffic sites, it's a suite of tools that standardize the translation process. In enterprise contexts, and for anyone managing advanced editorial workflows, TMGMT truly becomes the beating heart of the system.

Manual translation management (export copy-paste via email) is the main bottleneck for scalability. TMGMT solves this problem by introducing an abstraction and automation layer.

First of all, TMGMT allows completely decoupling the content source from the translation provider. We can therefore see two levels:

1. Sources: TMGMT can extract text from any Drupal element (Nodes, Blocks, I18n Strings). It doesn't matter if the content resides in a paragraph, a custom field, or a configuration string; TMGMT normalizes it into a translation-ready format.
2. Translators: Thanks to its plugin architecture, TMGMT is agnostic about who performs the translation. It can be a human user, an external agency connected via XLIFF files, or an automatic translation service. Today, LLMs are also included among translators, offered by various providers (OpenAI, Gemini, Ollama, Lara…).

The advantage of this flexibility is clear: it allows changing translation providers without having to rewrite code or retrain editorial staff, drastically reducing vendor lock-in risk.

Governance functionalities are another central added value of TMGMT. It allows assigning translation jobs to specific users, managing granular progress states ("pending", "translated", "reviewed", "accepted"), and having an overview of what has been translated and what hasn't. This structured approach ensures that translations aren't published blindly, but according to advanced review and validation pipelines.

Finally, an advanced functionality (particularly useful for high-update-volume sites) is Continuous Translation Jobs.

This feature reverses the traditional paradigm: instead of waiting for an editor to manually create a translation "package," the system proactively monitors content. When content is created or updated, TMGMT detects it and the new content is automatically added to a Job, then sent to the translation provider.

This mechanism eliminates "dead times" and the risk of drift between original and translated content, essential for maintaining consistency in e-commerce ecosystems or real-time news.

However, until recently, there was a traditional limitation. The options were polarized: on one hand manual translation (high quality, high costs and time), on the other classic Machine Translation (low quality, low cost). An effective "bridge" to services capable of combining automation speed with enterprise-grade publication quality was missing.

GenAI is changing this paradigm, fitting exactly into this space and enabling hybrid workflows that were previously unthinkable.

AI translations + human-in-the-loop: speed is important, but not at the expense of quality

LLM-based (Large Language Models) language automation today allows managing translation volumes that would have been humanly and economically impossible just a few years ago. Think of translating thousands of product sheets, technical knowledge bases, or historical news archives.

However, speed cannot become an excuse for quality degradation.

For institutional, strategic, or core business-related content, human input remains essential. AI, however advanced, can lack sensitivity to specific cultural context or may misunderstand tone nuances crucial to the brand. The winning strategy we're observing isn't replacement, but the hybrid approach: AI + Review (Human-in-the-loop).

Here arises a critical problem: many try to solve the issue by connecting Drupal to generalist models like ChatGPT or Gemini via generic APIs. While technically possible, this approach is often ineffective for enterprise. Generalist models are "know-it-alls": they translate a poem with the same statistical probability as they translate a technical manual, often inserting hallucinations or losing necessary terminological consistency.

Enterprise and Academic clients cannot afford these risks. A legal term translated approximately or an overly colloquial tone in institutional communication can create real damage.

When quality is a fundamental KPI, relying on generalist systems means shifting cost from translation to massive review, canceling the economic advantage.

If we want to leverage GenAI's power in contexts where accuracy is central, we need a specialized AI model. We need a technology partner that has solved the quality problem at the root. It's in this scenario that we introduce Lara Translate.

Integration with Lara Translate: why we built it

While Drupal Core provides the ability to store translations and TMGMT provides the logistical infrastructure and integration with providers, output quality depends on the translation engine.

If generic Large Language Models (LLM) have demonstrated impressive fluency, they often lack the domain specificity and terminological consistency required for enterprise use.

This is where specialized Language Models like Lara Translate stand out. It's an AI created by the Italian company Translated, a company specialized vertically in translations and high-quality AI technologies.

Our choice to integrate it into Drupal stems from the specific need of an institutional client to integrate a quality translation provider. From an in-depth analysis of available market solutions, Lara consistently positions itself a step above standard automatic translation, approaching the performance of the best human professional translators.

But what differentiates Lara from other solutions? The difference lies in the project's DNA. Lara is the LLM developed by Translated, a company operating in the professional translation sector since 1999.

Unlike generalist models trained on the entire web (including low-quality content), Lara was trained and fine-tuned on a proprietary dataset of millions of professional translations.

We're talking about decades of work done by over 500,000 professional linguists for 397,000 enterprise clients, in more than 200 languages, for a total of over 25 million real professional translations.

Lara "learned" to translate by watching how the best humans work, not by reading online forums. This specialization in training data is what guarantees superior output.

To bring this power into our projects, at SparkFabrik we developed and released the TMGMT Lara Translate module, a plugin that introduces Lara as a translation provider for all content in Drupal.

The plugin allows editorial teams to send content to Lara and receive translations directly in the Drupal interface, keeping intact all TMGMT's governance, review, and workflow functionalities.

The result is a fluid process: no more copy-paste, all the advantages of multilingual in Drupal, combined with automatically high quality. But to reach this quality level, some peculiar functionalities have been developed in Lara (and are fully supported in Drupal).

Additionally, Translated also offers the possibility to integrate professional human review (human-in-the-loop) for those translations requiring an extra layer of guarantee. As seen, Lara is a highly performant GenAI model in translation tasks precisely thanks to Translated's human-centric philosophy, which led to training based on millions of human professional translations (you can learn more here).

Distinctive features of Lara Translate integrated in Drupal

• Translation styles.
  Companies don't communicate in a single way. A legal contract requires absolute precision, while a marketing campaign requires creativity. Similarly, Lara doesn't translate flatly but natively integrates three distinct translation styles.
  • Faithful: Ideal for technical manuals, legal contracts, and content where terminological precision is vital.
  • Fluid: Perfect for general editorial content, blog posts, and news.
  • Creative: Designed for marketing and storytelling, where AI takes the liberty to adapt the message to maximize emotional impact.

• Context awareness and document coherence.
  Unlike old systems that translated sentence by sentence losing the thread of discourse, Lara analyzes the entire document. It understands relationships between sentences, maintains consistency of grammatical gender and references throughout the text, ensuring natural flow.
• Glossaries.
  Allow specifying correct translations for specific terms and phrases that are crucial for your particular context. This ensures Lara applies the right terminology consistently across all translations.
• Trust Attention.
  Lara uses a proprietary mechanism to "weigh" information. During generation, it prioritizes data from verified professional translations over less reliable sources. These also include revisions, corrections, and "error memory."
• Lara Feedback.
  Thanks to its dataset that also includes real corrections, Lara is able to "explain" its translation choices, providing an unprecedented level of transparency for an AI system (the so-called "AI Explainability").
• Access to experts.
  Translated's ecosystem allows, when AI isn't enough (for example for ultra-sensitive content), activating professional human translator services through the same pipeline. The transition from AI translation to on-demand professional human translation is thus made more immediate.

How to use Lara as a translation provider in Drupal

If you're familiar with TMGMT, it will be immediate to start using Lara. If you're new, here's a quick procedure overview (common to other providers).

1. Obviously make sure you've installed and activated TMGMT and installed the TMGMT Lara Translate plugin.
2. Go to Translation Management → Providers. Create an instance for Lara by adding your API credentials (you'll obviously need a Lara account). The settings allow you to customize the module according to your specific context, for example selecting the default style and linking glossaries.
3. Through TMGMT, choose the entities to be translated (nodes, paragraphs, etc.) and necessary languages. You thus create Jobs to send content to Lara.
4. Lara automatically translates and returns output to Drupal. Here you see Lara's quality: translations respect specific context, tone, and terminology.
5. Typically, output requires minimal human editing. Additionally, Lara supports review by highlighting ambiguities and providing explanations.
6. Once approved, translations are automatically published.

The hybrid approach, native but modern

As you may have noticed from the procedure, using Lara seems absolutely native in Drupal, especially if you've already worked on a multilingual site with TMGMT. What's different is the "engine" behind the scenes, a super specialized LLM.

Even with Lara as the basis of the automatic translation process, the human role in the process isn't eliminated or diminished. This is the concept of "Human in the Loop" (HITL), which here takes on a dual meaning.

• Quality AI as base. Lara provides a high-quality "first translation" that is often already final, drastically reducing editing time.
• Editorial control in Drupal. Thanks to TMGMT, the human editor can review the translation directly in the CMS before publication and manually edit the content. Thanks to output quality, these are typically minor interventions, especially if Lara is correctly configured with glossaries and brand tone. The reviewer is thus empowered and transformed into a strategic supervisor.
• Professional translations. For more specific and particular cases, it's possible to request professional translator services from Translated, the parent company behind Lara.

Adopting this technology stack generates immediate and measurable economic impact: the company can reduce translation budget by up to 80% or, with the same budget, translate 5 times more content, opening new markets previously unreachable due to cost limitations.

Indeed, 2025 market data highlights an enormous disparity between human and AI translation costs, and the hybrid approach allows having the best of both worlds: the following table offers an indicative estimate (see details here and here).

But the advantages of this approach don't stop at economic aspects. Equally relevant are:

• Time-to-Market acceleration, with consequent increase not only in speed but also in competitiveness in local markets
• Brand consistency, through use of correct terminology and unified tone of voice, otherwise difficult to obtain with fragmented human teams. We discussed consistency extensively in terms of Design System, but equally important is consistency in textual content.
• Operational scalability: the marketing team (or external support figures) doesn't need to grow linearly with the number of content and supported languages. It's possible to automate translation of "low-risk" content and focus human attention on sensitive content, strategy, and other high-value aspects.

Use cases

Adopting this architecture (Drupal + TMGMT + Lara Translate) isn't a theoretical exercise, but a practical solution to real problems. Not surprisingly, this integration was born from a client's request in a real business case.

It's the ideal configuration for high-content-volume sites that cannot afford the costs of a traditional agency for every single word, but also cannot accept the poor quality of raw machine translation.

Think of projects where tone of voice, consistency, and clarity are non-negotiable assets: international marketing portals, technical product documentation, legal or institutional sites. In these contexts, automation must be intelligent.

An immediate example? Think of an enterprise e-commerce with 50,000 SKUs: it can automatically translate product descriptions (in Fluid style) and technical specifications (with Faithful style), reserving human budget for reviewing technical details, marketing campaign pages and the home page, maximizing ROI.

Business Case: The Digital University

Let's look in more detail at a specific business case. A concrete example of the value of this solution is the work done for a prestigious Italian University (a real client for whom we originally developed the module).

• The context.
  A University is a huge editorial machine with hundreds of people working in different languages: institutional sites, department sites, news, research highlights, competition announcements, regulations, course program descriptions, administrative information... are just part of the content managed by university editorial teams. And typically, they must be published in different languages. In the education context, Drupal proves to be the ideal CMS.
• The problem.
  Manual translation times are incompatible with news speed. Solutions like Google Translate and continuous copy-paste (which also break formatting) are now unthinkable. But even using generalist LLMs, significant quality limitations were encountered, resulting in significant resource investment in the review phase. An alternative system was needed, high-quality and integrated directly into Drupal, in a workflow familiar to operators and able to guarantee full governance.
• The solution.
  After thorough research, Lara was identified as a provider and we implemented the TMGMT Lara Translate module.
• The new workflow.
  Today, University editors create content in Italian (or the initial language) on Drupal. With one click, they select target languages and send the job to Lara directly from the editing interface. Lara returns a high-quality translation, respecting academic terminology (thanks to specific training, use of glossaries, and customized instructions) and keeping HTML tags intact. The content returns to Drupal in the "To be reviewed" state. The editor takes a quick look, approves, possibly optimizes, and publishes.
• The result.
  Multilingual publication times have been reduced by 80%. Translation costs have plummeted, allowing many more contents to be translated with the same budget and high quality. Editorial control has remained firmly in the hands of the University, without duplications or data loss.

Conclusions, recommendations, and next steps

GenAI has had a disruptive impact on the entire content world. Yet, despite how it may seem, the GenAI era doesn't ask us to choose between automation and human quality, but to orchestrate them to leverage the best parts of both.

Managing a multilingual ecosystem is a strategic lever that directly impacts growth, Time-to-Market, and brand reputation. In a world of tool abundance, some fundamental details make the difference: quality, workflow, supervision.

The combination of Drupal CMS, with its solid, API-first, and inherently secure architecture, TMGMT, to effectively manage the localization process, and Lara Translate, with its specialized contextual intelligence, finally offers a concrete answer.

Brands are no longer forced to sacrifice quality on the altar of speed, nor to drain operational budgets to ensure terminological consistency on a global scale. The identified hybrid solution and the "Human-in-the-Loop" approach (validated through real case studies) are the ideal compromise. Editorial teams can free themselves from repetitive, low-value "linguistic data entry" work and elevate themselves to curators of global strategy, focusing on the cultural and communicative nuances that make brands unique in every market.

Recommendations for Decision Makers

For decision makers who intend to transform this vision into operational reality, the recommended roadmap is articulated in four essential steps:

1. Audit current flows: Map the existing "content-translation" lifecycle. Identify bottlenecks caused by human intervention, manual file management, or email exchanges. Effectively, how much time passes from creating master content in Italian to its actual publication in Chinese, German, or Arabic? If the answer is still measured in weeks rather than hours, the competitive gap is growing.
2. Adopt the structural stack: Implement multilingual management in Drupal with the TMGMT module. For enterprise sites, it's not optional, but an architectural requirement necessary to "decouple" content creation from its translation.
3. Opt for specialized AI: Start an initial pilot on non-critical segments, replacing generalist LLMs or manual processes with Lara Translate. Leverage the model's unique ability to understand the entire document context and programmatically adhere to your brand's style ("Faithful", "Fluid", "Creative") to drastically reduce the time and cost of human review.
4. Define governance: Establish clear guidelines on which types of content require human post-editing versus AI-only translation, using TMGMT workflow states to enforce these rules. For critical content, consider maintaining manual intervention by localization professionals.

By shifting the focus from manual translation to strategic supervision of reliable and contextual AI, companies can overcome language barriers with unprecedented speed and quality.

SparkFabrik, through its deep technical and strategic expertise in Drupal and the development of tools like the Lara connector for Drupal, positions itself as a key technology partner to guide organizations in this transition, transforming the challenge of linguistic complexity into a structural competitive advantage.

If your organization is exploring adopting Drupal as a CMS that's robust, reliable, and customizable, introducing multilingual strategies, or AI integration for its digital initiatives, we invite you to:

1. Explore our case studies of enterprise Drupal implementations
2. Contact our team for an assessment of your specific needs
3. Discover how our Drupal services suite can support your AI strategy

This article is part of our series dedicated to Drupal CMS. To explore other aspects of the platform, we invite you to consult our previous articles on features and benefits, comparison with alternatives, migration strategies, security and compliance, composable architecture, Design System, Drupal headless omnichannel, and Drupal AI overview and news.