Google Translate for business use – risks and alternatives

If you run a small business, you might rely on Google Translate for quick translations. It’s free, fast and always there. But what if that “free” translation costs you a client, leaks your data or lands you in legal trouble?

In this blog post, we explore the data security, quality and legal risks associated with free translation tools. We’ll also show you professional alternatives that fit your budget.

Data security risks of using Google Translate for business

First, let’s clear something up before we go on. We’re talking about the free, web-based Google Translate service here. The one that lets you translate text, images, documents and websites from a web browser. We’ll talk about Google’s fee-based machine translation solutions next.

When you paste content into Google Translate, you may be exposing your business to risks without realising. That’s because Google’s Terms of Service state that data entered can be used to improve its services.

This means:

• Sensitive client data could end up stored on Google’s servers
• Commercial secrets may no longer remain confidential
• You lose control over how your data is processed and stored

This actually does happen. In 2017, employees of Norwegian energy firm Statoil found out that details of their performance appraisals were available online. All that sensitive information was accessible to anyone via a simple Google search.

If you handle work under NDA, free tools like Google Translate pose a genuine data privacy threat. Let’s take a look at a few less risky alternatives.

Secure machine translation alternatives to Google Translate’s free service

Google Cloud Translation Hub and Translation API

Google Cloud offers various paid-for machine translation options.

Its pay-as-you-go Translation Hub offers self-serve document translation. There are two pricing tiers, and both tiers allow you to define a data retention policy.

Google’s Translation API may be more suitable for large LSPs or enterprise customers. Its Advanced option is optimised for customisation. It’s complicated to set up, though, and Google doesn’t offer direct technical support.

As for data security? Google Cloud has data centres in various locations around the world, and you can select which region to locate your apps. This is helpful if you have specific data security requirements to meet. You can find out more about Google’s commitment to data security in the Google Cloud Trust Center.

Planet Languages’ take? As a provider of machine translation post-editing services, we opted not to use Google Cloud. Why? The complex set-up for starters. And as a UK-based company that works with EU-based clients, we wanted greater reassurance on GDPR compliance. So we looked to Europe.

EU-based machine translation providers

German firm DeepL offers AI-powered machine translation. Its paid-for Pro plans offer maximum data security, including GDPR compliance and certification to ISO 27001.

ModernMT is another European machine translation provider. They’re GDPR compliant and ISO 27001 certified.

Legal and compliance issues with free translation tools

It’s worth pointing out that most AI translation providers only guarantee data security with their paid-for plans. If you’re accessing their free web browser-based tools, they’re likely to be using your content to train their models. Just like Google Translate.

Many industries have strict compliance requirements for handling their content. If you’re exploring paid-for machine translation options, check that the terms comply with:

• GDPR
• Client confidentiality clauses
• Industry-specific compliance standards

This is particularly important if you’re handling financial data or medical records. Steering clear of free translation tools will help you avoid substantial fines and incalculable reputational damage.

Quality problems that damage your professional image

Do paid-for AI translation plans mean you get better quality? Not quite. When we talk about quality, there are lots of factors to consider:

• The model you’re using. We’ll go into the different types of AI translation below.
• The language pairs you’re working with.
• Your content type and the domain.
• User expectations – do you just need the gist? Or do you need to impress your readers?

Let’s take a closer look.

“This sounds like Google Translate” isn’t the insult it once was…

Translators used to dread receiving feedback like this back in the noughties. But machine translation technology has come a long way since then. Google began using neural machine translation in 2016. The move away from statistical machine translation brought with it improved fluency. Fast-forward to today, and Google Translate is pretty impressive for everyday use and gisting or informational purposes.

…but it falls short in most professional contexts

When you take a closer look at the output, cracks start to appear. How important is your brand’s reputation? Even minor errors or omissions can undermine your credibility and confuse your readers.

So, what are the alternatives to Google Translate when quality matters? Does AI translation live up to the hype?

Not all AI translation models are created equal

When quality matters, it’s important to understand what you can expect from AI translation models. Some might be a better fit for your content and industry than others.

Generic neural machine translation models

Google Translate and its paid-for options are based on Google’s pre-trained neural machine translation model (NMT). This model has been trained with generic data from a variety of domains. The aim is to produce reasonable output regardless of content type. Other providers in this category include Amazon Translate, DeepL and Microsoft Translator.

Customisable neural machine translation models

In some cases, it’s possible to train these models with your own data. Google Cloud’s Advanced option allows you to do this. You can also customise DeepL’s output in some languages by adding a glossary. Often, this helps produce better-quality output. Having said that, there’s usually a fair amount of trial and error involved in customising these models.

Domain-specific models

These models have been trained with data from a specific domain. Let’s say you’re a fintech company. You’re likely to get better output from an AI translation model that’s been trained on financial content than a generic model. Some models have already been pre-trained with domain-specific data. Systran is one such provider. Or you can train your own model.

Generative AI

The language services industry has used AI in one form or another for years. Yet it’s only recently that everyone has started talking about AI’s potential. It happened with the arrival of generative AI. And specifically, Large Language Models (LLMs).

Generic LLMs

LLMs are a type of generative AI that focus on text. If you’ve ever tried using ChatGPT to translate something, you’ve used a generic model. You can try prompting or Retrieval Augmented Generation (RAG) to refine the output. The trouble is, ChatGPT wasn’t designed with translation in mind. You might get fluent-sounding text, but it often hallucinates and/or misses out key details. It’s a risky choice if you need an accurate translation.

LLMs that are purpose-built for translation

There are now LLMs that have been trained specifically for translation tasks. Examples include Lara and Widn. However, even with purpose-built LLMs, the result is often unpredictable.

Reality check – if quality matters, human oversight is crucial

Different AI translation models support different languages. While data security may be high on your wish list, the most secure model may not offer the languages you need.

Is quality high on your wish list too? How important is it that your translated text uses your approved terminology and tone of voice, and accurately reflects your brand image?

Output quality typically varies between AI translation models and between language pairs. They also tend to perform better with higher-resource languages. That’s because providers have been able to draw from a bigger dataset in these languages. For lower-resource languages, automated solutions should be avoided for now, unless you have a robust post-editing process in place.

There is usually a trade-off with any automated solution

Generic AI translation models have been trained on vast amounts of data. Some of which may contain biased, harmful language. You may see examples of exclusionary language in machine-translated output.

Neural machine translation models tend to produce accurate but clunky translations. LLMs, in contrast, often produce very fluent-sounding output. But they tend to be slower and more expensive than NMT. They can also miss key details, invent content, hallucinate and behave unpredictably.

None of these models guarantee complete accuracy. And often, it’s actually quite tricky to implement AI translation providers’ API solutions if you’re not tech-savvy.

“But I really do only need the gist”

We hear this a lot. Let’s look at how Google Translate coped with translating clauses in a Spanish contract template into English. Two high-resource languages. Shouldn’t be too much of a problem, right? In Spanish contracts, you often see clauses enumerated with ordinal numbers (first, second, third, etc.). So, how did Google Translate fare?

1. Primero. First.
2. Segundo. Second.
3. Tercero. Third.
4. Cuarto. Room.
5. Quinto. Fifth.

Not quite! At least, not in this context. The Spanish noun cuarto does indeed mean room. But it’s also an ordinal number. If Google Translate falls short with such relatively simple wording and in high-resource languages, how does it handle more complex content and other languages? How confident are you in its ability to give you the gist in lower-resource languages?

When Google Translate mistakes cost more than professional services

While Google Translate is free upfront, mistakes can lead to losses far greater than professional translation fees. The same is true for paid-for AI translation tools.

What happens if you lose business due to poor communication? Or if you need to reprint marketing materials to correct mistakes? Or if you face fines from breaching confidentiality or compliance rules?

Those free or low-cost AI translation tools can quickly become very expensive. Read more about the true cost of machine translation and why post-editing matters.

Don’t wait for a costly mistake to happen to you. At Planet Languages, we’re here to help. Contact us to explore secure, budget-friendly translation options.

Professional translation alternatives that fit your budget

The good news is that professional translation is more accessible than you might think. Here are some budget-friendly alternatives for your business:

• Machine translation post-editing. AI translation tools often serve as a good starting point. We select the best-performing model for your language pair and content type. Professional linguists edit the output so that it’s fit for purpose. There’s an ISO standard for this service. It’s ISO 18587, and it applies to full post-editing. Planet Languages was among the first companies in the world to obtain certification to this standard.
• Professional human translation for critical content. Our recommended approach when precision matters. Think contracts, certified translations and high-stakes content.
• Tailored translation packages. We can customise language solutions based on your precise requirements, strategic priorities, volumes and budget.

At Planet Languages, we combine technology with professional linguists to deliver secure, high-quality translations that protect your reputation.

Key takeaways

• Free machine translation ≠ secure.
• Paid-for AI translation tools don’t guarantee high quality.
• You need a human expert if quality matters.
• Free and low-cost tools can end up being more expensive than professional translation.

Conclusion: protect your business with professional translation solutions

Using Google Translate for business may seem convenient. But it can damage your professional image, breach confidentiality rules and cost you clients. Affordable professional alternatives are available to keep your business communications effective, accurate and secure.

Protect your business with professional translation that fits your budget. Contact Planet Languages for affordable, secure translation solutions tailored to your business.

About the author

Bethan Thomas has worked in the language services industry for 20 years. She helped Planet Languages achieve certification to ISO 18587 soon after the standard was first published. Our continued certification to this standard demonstrates commitment to incorporating AI into workflows responsibly.