AI translation tools are reshaping ESOL, but not for the better

I’m marking a set of entry-level ESOL writing tasks.

The prompt is simple: Write about your daily routine.

It’s a task I return to often because it reveals the development process of learning writing. At entry level, writing begins with limited vocabulary and partial understanding. Learners test what they know, approximate what they don’t, and build meaning step by step. A first draft is rarely accurate, as it’s not meant to be. Instead, it shows how a learner is thinking through language.

But as I read through the scripts, it becomes obvious.

The sentences are structured with precise grammar. Ideas flow in a way that doesn’t match what I know of the learner’s current level. The writing is both fluent and accurate but the problem is it’s not their own.

When I ask how the work was produced, the answer is direct: a translation tool.

AI translation tools are not neutral supports in ESOL classrooms. When they shift from supporting learning to completely replacing it, they disrupt the messy, essential processes through which language is truly acquired.

There’s no doubt these tools have value. In multilingual classrooms, they can support communication, reduce isolation, and help learners access meaning quickly. When used carefully, they can help build confidence — especially at lower levels, where linguistic barriers are highest.

The issue therefore isn’t that learners use them but what they significantly replace in the process of language acquisition.

In practice, what begins as support can quickly become the default. Many ESOL learners are adults managing complex lives: long working hours, caring responsibilities and limited study time. Faced with these pressures, the efficiency of a translation tool is understandably appealing. But pedagogically, it comes at a cost.

In the moment that a learner would normally search for a word, attempt a sentence, or make an error, that fundamental part of the process is being skipped.

The learner is no longer forming the sentence for themselves, and meaning is no longer being worked out step by step. The authentic mistakes that they can connect their understanding to through feedback are never made, which creates a significant gap in opportunities to learn from.

This pattern isn’t limited to writing. In class, smartphones are ever-present. I encourage their use for soft support: to check meaning and aid participation. Yet it’s increasingly common to see tasks mediated entirely through translation apps. Learners photograph texts and instantly translate them. Speech is also recorded and converted before it’s even attempted. The space where learning would normally happen is not only erased but the confidence needed for any speaking activity is crippled.

Language learning is uneven and uncomfortable. It depends on repetition, hesitation and error. These aren’t inefficiencies – they are the process, and every learner needs reassurance that it is normal. When that process is removed, it becomes very hard to track and evidence real development.

This is why recent claims that AI tools or apps like Duolingo could replace ESOL provision are misguided. There’s no doubt these technologies can support vocabulary and basic interaction, but they don’t replicate the relational, developmental heart of language learning.

ESOL classrooms are more than spaces where language is delivered. They’re environments where learners practise, receive feedback, negotiate meaning, and build confidence. They’re also sites of connection – often vital for those facing isolation from their wider communities.

A translation tool doesn’t know its learner and can’t build interpersonal exchanges, see hesitation, spot error patterns or respond contextually to learners’ needs. It also can’t assess progress meaningfully. In formal ESOL settings, teachers must evidence development over time. That requires visibility of a learner’s own language production, including its limits. When learning outputs are generated externally, it becomes harder to recognise or measure what a learner can actually do for themselves.

The question, then, isn’t whether AI tools are positive or negative but how they’re positioned. When they are used selectively, they can support understanding and skills development. However, we must also acknowledge that they risk displacing the very processes that make language learning possible when used uncritically. Instead of outrightly banning them, the task for educators becomes ensuring they remain scaffolds for language learning, not substitutes.

In ESOL, progress is found in the gradual, imperfect work of building language, one uncertain step at a time. That work cannot be outsourced to produce perfect sentences that are void of the messy, sometimes awkward but meaningful language found in authentic learner work.