Closing the loop: why AI localisation only works when humans stay part of it

TL;DR

As communication is more than words, localisation is more than translation.

Success means efficient and productive workflows at scale.

Multilingual content marketing succeeds with human nuance.

Thinking of AI localisation as a substitute employee is wrong.

When localising, AI magnifies human skill, not replaces it.

‍

How localisation became a bottleneck for content teams … and how AI is part of the solution

If your market speaks more than one language, content marketing is a lot more than “writing stuff”. Indeed, content creation is just the start. And the freelancers we work with at Contentoo – experienced professionals in the top 10% of the sector – all know it. Not one is “just a writer”.

The trouble is, as you add markets and scale, it’s all too easy to add friction and costs, too. 

Smaller market with less budget? “Ah, just use AI translation.” But as a result, your target consumers feel alienated, and your market share never grows. You originated in Germany, but your largest sales now come from Spain? There’s an invisible 20% drag on your workflow, as you create in a Dialekte, customers don’t read. 

And the average content team is just 2-5 people (often a headcount of just one) with 60% active in more than one language region. Let in just a few of these “friction factors”, and you’re looking at late nights in the office, weekends with your laptop, and quiet sobbing at 2 AM as you realise you’ve got a Zoom at 7 (we’ve all been there). 

In fact, our recent survey found localisation help was the No. 1 “extra pair of hands” needed.

AI localisation tools definitely help. But that’s the key point. “Help.” Not “do it all for you”. Thinking AI is a substitute for employees is wrongheaded. These tools, wonderful as they are, remain just tools. 

AI can add speed and volume at scale. But AI translation struggles with things like cultural touchstones and brand voice – putting audiences in the “uncanny valley”, where what they read is just … *off*. They feel no human connection, no emotional anchor.

Watch Permission to Rant here.

And if you think these factors don’t matter, that tug of the heart is the secret of every great book, every hit song, every beautiful artwork. It’s where the magic happens: the sense someone “gets” you. And that’s how strong relationships start.

This article explores why localisation is such a bottleneck for modern content teams, where AI can help and where it can’t, and why the future of scalable localisation is a human-in-the-loop model. 

Finally, you’ll see how Contentoo does it – combining AI-driven workflows with human oversight and direction to help lean teams localise content at scale. Let’s get started.

‍

AI localisation has become popular …

Any business is interested in more, faster, cheaper – and with localisation traditionally a human-centric function, it seems ideal for a sprinkle of AI fairy dust. 

The driver: localisation is more than swapping one language for another. It’s about a long list of detail-intensive, time-critical, and – to be honest – boring-as-hell administrative tasks. 

Deadline orchestration, version control, rolling out changes at scale: repetitive tasks, ripe for automation. 

And AI – able to deal with unstructured data, like natural language – holds promise. That’s why 86% of teams use AI in their workflows. Work out the inputs and outputs, set expectations and constraints, and let an LLM do its thing. The results are … acceptable.

Until (actual example) a client selling electric scooters complains that your translation of “ride”, in his language, only applies to a horse. 

The checks and controls that prevent this (and far larger errors) aren’t a minor job at the end of a project. They’re the difference between a successful outcome and your company turning into an embarrassing meme. And there are thousands of them to think about.

See how SendCloud increased content production 50% … without adding to human headcount!

‍

… but AI content localisation isn’t a magic bullet

So despite AI’s huge scope for multilingual content marketing, it’s important to recognise its limitations. Here are three of the biggest.

‍

AI follows its own rules, not your market’s

Localisation isn’t about saying the same thing; it’s about saying the right thing. And even in single markets like the EU, these differ from State to State. Especially in markets like healthcare and food, translated text must comply with local definitions, prohibited claims, labelling requirements, and more.

When an LLM’s training data doesn’t include relevant facts, the AI will make things up, or “hallucinate”. Even at this base layer of localisation, an unsupervised AI translation could cost your company thousands in fines – or worse. 

‍

An ongoing struggle with cultural relevance

One layer up, effective communication means adapting tone and register to the target audience. (A taco recipe for the British market needs details a Mexican grandma would find insulting.) Idioms and slogans, figures of speech, and  commonly understood metaphors and analogies: all are language, region, and frequently demographic-specific. What, for example, was that “6-7” thing all about? 

In Spanish, you cause chaos by “arming a chicken”. When it “rains cats and dogs” in the UK, it’s all “chickens and pigs” in China. If a Russian wants to confuse you, he “hangs noodles on your ears”; if your personality changes in Japan, you’ve “put a cat on your head”. And across the Arabic-speaking world, someone who “brings all his camels” isn’t offering transport or trade: he’s burdening you with his problems. And when the late US President Ronald Reagan tried to say “Relations between America and Japan are not without issues” in Japanese, what came out was “Japanese-American bathhouse prostitutes can be a real handful.” 

AI localisation is getting better all the time, but the core problem here is that one small error can poison workflows long-term, alienating and confusing readers everywhere. And without human oversight, the risk of those small errors is always there to cause hassle. Or – at worst – cause international incidents. 

‍

Brand voice is hard to automate

When a comedian impersonates a celebrity, he’s not imitating their voice directly; he’s over-emphasising traits and quirks. (Donald Trump doesn’t really say “very strongly” in every sentence, despite the skills of China’s Ryan Chen.) Ask an AI to replicate a brand voice, and you get the same problem: it often sounds like a caricature, not human warmth. And readers are quick to spot someone who’s faking it. Again, nuance is all.

It’s because a brand voice is more than a Style Guide. It’s how your company would talk if it were a person. Quickwitted and opinionated? Bohemian and sarcastic? Authoritative and knowing? Communicating naturally takes a dose of human judgement and critical thinking – it’s why AI translations at Contentoo are always moderated by a human editor. 

AI doesn’t do brand voice well for the same reason it doesn’t write original content well: there’s no internal model of the world, no self-awareness or empathy. This quality is fundamental to today’s LLMs, and until completely different technical architectures arrive, it won’t go away.

‍

Errors can scale faster than output

With content localisation at scale, an error of translation doesn’t pop up once to be whacked down. It’ll roll out into many materials, multiple versions, and different formats. And just when you thought you’d caught it – boom! Another variant rears its head. In an environment with millions of words in dozens of languages, errors multiply and mutate like viruses. 

The key learning here: dealing with these errors takes exponential time. Catch it at the source, and it’s a few minutes of corrections. One week later, it’s in a hundred PDFs being downloaded worldwide – and that’s a hundred trouble tickets in your localisation workflow. Each needed isolation, replacement, and an apology to readers. 

AI, without the sense of personal responsibility we humans have (or our fear of job loss!), will make mistakes, roll them over into future versions, then use them as the basis for new mistakes elsewhere. AI engineers have a word for it: “model drift”. Relying on what happened within its experience, without the context of the real world beyond. And that’s not a mistake humans make. 

See Contentoo’s take on the future of content.

‍

Localisation versus translation (and why it matters)

Let’s summarise the difference between localisation and plain-vanilla translation. Translation stops at input: making sure a translation has the right meaning. Localisation is much broader: it’s about achieving the right output, or response from your audience. Whether that’s to buy your product, read your instructions, or think about your brand in the way you want.

Without this deep attention to the outcomes your localisation generates, even polished translations will fall flat. Because they’re not building your brand or reinforcing the marketing investments you’ve already made. 

Such nuanced judgment isn’t yet a skill AIs have: these blind and deaf ghosts-in-the-machine lack the embodied experiences and emotional involvement to inform such decisions. 

But with human guidance and moderation – the right prompts in the initial instructions, the right moments in automated workflows – understanding can be built, and model drift can be smoothed out, at the touch-points where it matters most. 

AI can apply information and knowledge at scale. But humans can apply wisdom.

‍

Why keeping humans in the loop is so important with AI localisation 

If the above sounds sceptical of AI, think again. We’re huge proponents of AI localisation at Contentoo, and in several areas, we’re taking a leadership role in its use.

We just don’t think it should do everything. Anyway, what tool does?

Within a large localisation project, there are hundreds, thousands, perhaps, of processes ripe for AI assistance and automation. Repetitive actions on unstructured text, sense-checking via back translation, and applying style consistency across millions of words. Things that once took humans countless hours. 

But there are distinctly human roles, too. Understanding how cultures interact. Applying real-world experiences to lift readership. And – most of all – how such creativity can improve the performance of a localisation effort. Leading to higher sales, fatter margins, and costs kept under control instead of rising faster than outputs. 

How Contentoo helps its clients speak fluent customer.

‍

How Contentoo helps teams localise content at scale

Just as an effective client relationship is about partnership, we believe effective AI localisation is a partnership that includes humans. Whether it’s a 50:50 split, 80:20, or 99:1, the real value is in how the differing skills are mixed and matched. 

AI does things faster – but takes zero responsibility and makes errors. Human talent produces great localised work – but takes time your deadline can’t offer. 

But when some core principles – like human-written brand standards and playbooks and local-language editors overseeing AI-produced content – are followed, even the largest localisation challenges can proceed efficiently, cost-effectively, and without surprises. And that’s how we do it at Contentoo. Humans in the loop.

Everyone thought AI was the magical fairy dust for multilingual content marketing. 

In fact, it was humans all along. 

To talk about how Contentoo’s approach to AI localisation could work for you, contact us.

‍

Key takeaways for content teams expanding globally

• As your localisation workload increases linearly, the opportunities for errors increase exponentially. Costs can rise faster than productivity. 
• AI is wonderful .. for some things. The opportunities are where tasks are well-defined, repetitive, high-volume, and with a clear picture of what success looks like.
• Keeping humans in the loop keeps your localisation responsive to the nuances, understandings, and cultural differences of markets worldwide. 
• The best outcomes stem from a clear division of labour that recognises the strengths of AI LLMs and skilled humans. 
• Multilingual content localisation is a team effort – and the team includes both humans and machines!

‍

FAQs about AI localisation and human editors 

FAQ 1: Why can’t AI be used for every localisation task?

It can. But even if it gets it right first time – unlikely – over time it will “drift”, making errors and compounding them over time. The longer you use AI alone, the greater your business risk. 

FAQ 2: Can I use AI to write my translations?

Yes – but it’s far more effective to involve a human editor or translator, too. An AI’s copy may be grammatically perfect but still feel “foreign” or off-brand to audiences; human editors catch these subtleties and correct them.

FAQ 3: What does a human editor add to AI localisation?

Human editors check standards against playbooks and brand guides, streamline tone and ensure cultural norms, and align content with local marketing norms and legal constraints – in many cases, using their own personal experiences of the market.

FAQ 4: Is AI useful for technical and engineering localisation?

Yes, this is where AI localisation is most effective – sectors with agreed-on processes and precise terminology. This also means, however, that errors “look bigger”, with greater consequences for business relationships and product functionality. 

FAQ 5: Will AI replace humans in the loop … eventually?

It’s more likely that roles will change than disappear. Localisation professionals have always been more than translators – and today they’re evolving into Quality specialists, prompt engineers, and workflow managers that take their skills to the next level.

‍