AI in translation: 8 key questions, answered

AI continues to dominate boardroom agendas and the pressure on teams to use AI for efficiency is only increasing. But is it possible to apply AI to translation in a way that is practical, low-risk, and high-reward?

In the kickoff session of Smartling’s new AI Translation 101 series, CEO Bryan Murphy and Sr. Director of AI Solutions, Alex Yanishevsky, demystified the role of AI in translation. From exploring real use cases to offering practical advice on adoption, we shared the latest of what we’ve been seeing and how you can apply it.

Below, we recapped the conversation, addressing 8 key questions to help you approach AI with confidence.

📺 Ready to watch? Stream the full session on demand.

🎧 Prefer to listen? Catch the podcast here.

📘 Want to go deeper? Our new ebook, Navigating the Shift: Why, When, and How to Adopt AI Translation, breaks down the strategy behind successful adoption. Check it out here.

1. What’s the current buzz about AI in translation?

AI is everywhere - on conference stages, in investor meetings, and on leadership to-do lists. Unfortunately, this buzz often turns into pressure: “How can we use AI to translate more, faster?”

At first, the big question was whether AI was reliable enough to use at all. Today, we know it’s efficient, scalable, and cost-effective. 

Now, the real questions are:

• How do we leverage it effectively? 
• How do we scale it? 
• And how do we use it while safeguarding quality, security, and privacy?

2. Where is AI actually driving efficiency today?

The most efficient use cases for AI are during pre- and post-translation to help deliver the highest quality translations. AI performs especially well on repetitive, rules-based tasks like project estimation, content triage, and post-editing. Alex highlighted several examples where customers used linguistic assets, like glossaries and style guides, to train AI and accelerate their workflows, without sacrificing quality. Tackling workflow bottlenecks like time-consuming human review steps is where AI can deliver immediate, measurable value.

One of the biggest outcomes teams can expect from AI-powered translation is high-quality, on-brand content delivered at a fraction of the cost and turnaround time of traditional translation. Companies can translate 8x the content for the same cost, liberating budget for even more multilingual content.

Proof point: Secret Escapes used Smartling’s AI Human Translation (AIHT) to reduce translation time by 25% across all languages, with even greater gains in Italian and Dutch. Their editors now have more time to focus on creative or strategic localization work.

3. How advanced is LLM translation? And is MT still better in some cases?

AI is evolving fast, but it’s not magic. While large language models (LLMs), like GPT-4, can generate more natural and expressive translations, traditional machine translation (MT) still excels in highly regulated, terminology-heavy domains. 

While LLMs do still hallucinate, it’s a solvable problem. Alex compared this issue to a highly gifted eight-year-old: just because they excel intellectually doesn’t mean they’re ready to drive a car. Likewise, while LLMs can perform at a high level, they still do silly things, like hallucinate. Alex reassured that when this issue occurs, “we have enough tools, algorithms, and logic to catch such silliness when it goes off the rails.” At Smartling, when we receive the translations from LLMs, if they don’t meet our standards, we’ll push them to the next step of adding a human in the loop. Ultimately, we use a “trust but verify” approach with LLM outputs.

Bryan put it simply, saying MT is great for “accurate translations, while LLMs tend to be better at more brand-specific, fluent, and culturally-adaptive translations.” It’s not about choosing LLMs or MT; it’s about using each where they perform best.

4. What’s the upside of using AI in your workflows?

The benefits of AI go far beyond speed and scale. It’s also changing the how of translation. From speed to scalability to cost savings, AI opens the door to more efficient global content and localization programs. When paired with human-in-the-loop workflows and well-crafted prompts, it also delivers higher quality, faster.

Until LLMs came to be, we would get an automated translation and be stuck with that; there wasn’t much you could do if you weren’t happy with the output. But now, with LLMs, we have prompts that allow us to control the style and tone of the translations, with a customized approach. This wasn’t possible before, and the implications are quite profound. 

It’s essential to remember that AI doesn’t replace human translators; it empowers them to focus on higher-value work.

“Up until now, if you were translating in 50 languages, you would have to talk to up to 50 different linguists, and talk to them about the tone and style to be changed. 

Now, we’re baking that right into the technology itself and our customers are able to have a very fine control over the translations we’re producing, in a highly automated way.”

 – Bryan Murphy, CEO, Smartling

Proof point: Therabody used AI translation to replace manual workflows, cutting translation costs by 60% and achieving a 99.7% on-time delivery rate.

5. What are the challenges of implementing AI on your own?

Implementing AI translation isn’t plug-and-play. It requires strategic planning, quality benchmarks, and expertise in prompt engineering, data governance, and model selection. There’s risk in underestimating the complexity.

Alex drew a compelling parallel between "DIYAI" (do-it-yourself AI) and pickleball. While most can grasp the basics with fundamental hand-eye coordination and athletic ability, true mastery demands dedicated practice. Similarly, investing in and implementing AI follows the same principle. You might achieve a decent translation by crafting a prompt in your own language and providing feedback. However, do you truly possess the expertise to replicate this across 50 languages? Are you a linguist or a data scientist with a profound understanding of these intricacies? This is precisely where DIYAI falls short. The better path is partnering with experts who understand both the tech and the linguistic or brand nuance needed to scale successfully.

Proof point: Gemini, a crypto exchange, shifted from human-only translation to AIHT and doubled translation speed while maintaining quality for highly technical content.

6. How can AI elevate your career?

Adopting AI is more than a team-wide productivity boost. It’s also a career accelerator. A key distinction lies between being "AI-first" and "AI-everywhere." To be AI-first, consider whether a task can be addressed with existing technology before turning to AI.

By contrast, AI-everywhere means you use AI solutions no matter what the problem, which can actually be inefficient. Alex likened this approach to using a diamond to cut glass: while this approach is possible, it would be overkill and quite costly. Likewise, tasks that can be done with simple automation don’t require the power (or cost) of LLMs. Instead, save AI for the areas where other tools fall short.

Understanding how to make your tasks more efficient, so you can do more, will advance you in your career. Bryan emphasized that the professionals leading AI adoption today are tomorrow’s thought leaders in the years to come.

7. What’s the smartest way to start testing and implementing AI?

Start small, test often, and scale based on data. 

It’s easy to get overwhelmed by AI. While executives see AI as a magic bullet for productivity, many teams in localization, product, and marketing are still unsure how (or where) to start using it effectively. 

To avoid overwhelm, start by defining your goals: are you trying to reduce turnaround time? Expand content volume without growing budget? Once you’re clear on your “why,” it’s easier to align stakeholders and evaluate the right tools.

Intentional testing is key. Build a roadmap, not a guesswork strategy. Once AI proves its value, expand strategically.

Proof point: A Fortune 100 tech company started by testing Smartling AIHT on select content, then expanded usage, saving $3.4M in the first year while delivering content 50% faster and maintaining a 99%+ MQM score.

8. What’s one piece of advice for AI-curious teams?

Having a trusted advisor is key. While DIYAI may be tempting, it’s essential to consider the expertise needed to really translate at scale (and with accuracy) before diving in with DIY. 

The big picture

AI in translation isn’t a passing trend; it’s a transformational shift. While it won’t replace human talent or traditional tools outright, it’s already reshaping how companies scale and sustain their global voice. 

Whether you’re just starting to explore AI or already in the trenches of testing, the message from our speakers is clear: prioritize clarity, think in systems, and don’t be afraid to ask the big questions.

Ready to watch? Stream the full session on demand.

Prefer to listen? Catch the podcast here.

Want to go deeper? Our new ebook, Navigating the shift: Why, when, and how to adopt AI translation, breaks down the strategy behind successful adoption, along with the ROI you can expect. Check it out here.