1. Instant gratification, instant dependency?

In my reading class at the ELC, the projector glowed with an English text while students’ phones hummed with instant Chinese translations. They nodded rhythmically, as if comprehension were flowing smoothly. Yet when I asked a simple question without devices, silence flooded the room. That silence taught me more than their nods: they were not reading English; they were outsourcing thought. In the commercial culture of ELCs, this dependence on an instant AI translator is no accident. Families pay for speed, brochures promise rapid results, and students internalise the message that learning should be immediate, measurable, and painless. Some colleagues argue this was just translanguaging – students mobilising multiple linguistic resources (Carstens, Reference Carstens2016; Caruso, Reference Caruso2018). But translanguaging is not mere one-to-one translation; it is a creative orchestration of languages to make meaning (Caruso, Reference Caruso2018; García & Kleifgen, Reference García and Kleifgen2020). In my classroom, genuine translanguaging often sounds like jazz – improvised, imperfect, alive. For example, a student might explain an IELTS reading passage to a peer in Mandarin, sprinkle in English terms like “global warming,” and then shift back to Mandarin to check understanding. The exchange is not polished, but it breathes. By contrast, when the same student pastes the passage into an AI translator, the output is like karaoke with the original vocal track: technically correct, but it carries someone else’s voice. In the for-profit system of China’s ELCs, karaoke-like perfection fits the promise of speed – but it hollows out the struggle that makes learning durable.

Worse still, Chinese face culture can intensify this dependency. Many students dread the shame of mistakes. AI offers them a mask: sentences correct, answers safe, dignity preserved. The benefit is real – timid students dare to participate – but the cost is high: error, the most valuable teacher, is erased. Swain’s (Reference Swain1993) Output Hypothesis reminds us that being pushed to produce language, even imperfectly, drives development. As such, when AI pre-corrects these moments, it robs learners of the productive struggle that builds linguistic muscle (Leeser, Reference Leeser2008; Russell, Reference Russell2014).

To counter this, I began to design “useful friction” into my classes. For example, students were encouraged to first summarise a paragraph in their own English, then consult AI to compare and refine. The pauses in class grew longer, the air thicker with thought, but genuine learning began. In these moments of friction, students realised that access to meaning is not the same as ownership of meaning. The act of wrestling with a sentence – even risking getting it wrong – engaged them in true comprehension and use of English. By slowing down the immediate gratification, we paradoxically sped up their internal growth.

2. The native-speaker mirage

One afternoon in a Business English class, a student drafted a job application line: “I think I may be suitable for this position because I am willing to learn and improve.” It carried her voice – cautious, respectful, sincere. The AI tool instantly revised: “I am the best fit for this position because I learn fast and work hard.” Confident, fluent, and no longer hers. Her original “may be suitable” reflected humility, not insecurity – a cultural signal of modesty deeply valued in East Asian contexts. The AI’s edit flattened this nuance, enforcing a norm of Western assertiveness. This is what Holliday (Reference Holliday2006) terms native-speakerism – the ideology that ideal English is embodied by Western native norms. Driven by the commercial imperative of ELCs, this native-speaker mirage becomes even more seductive. Just as parents applauded the AI-polished essay for its showroom shine, students and families often equate native-like fluency with market value. Within such a system, an assertive, standardised phrasing like “I am the best fit” appears not only linguistically superior but also commercially desirable – evidence that the tuition is worthwhile. Yet this pursuit of polished nativeness obscures what matters more: context-responsive communication that preserves nuance and identity.

Therefore, I worked with the student to craft a compromise: “I believe I could be a strong candidate because I’m eager to learn and improve.” This preserved clarity while honouring her tone. It reminded me that English is not a soloist but a choir. As Kachru (Reference Kachru1992) argued, English has become a repository of diverse cultural identities rather than a unitary cultural norm. Multiple voices, accents, and registers belong in the choir; our goal as teachers is to help students find their harmony, not to make everyone sing the same melody. AI prefers one clean melody, but our classrooms must hold room for harmony and counterpoint.

To resist the mirage that “native-like” equals “better,” I now run critical AI literacy sessions. Students compare their drafts with AI’s revision and ask pointed questions about the differences. For example:

• • Whose values are encoded in the AI’s suggestions? Is the AI privileging a more direct, individualistic tone, and if so, why?

• • What identity or voice does this phrasing project? Am I sending an overly aggressive self-image that isn’t true to me?

• • Whose English is being modelled here?

Sometimes AI’s edits help – catching a grammatical error or offering a clearer word. Other times, they steamroll the author’s intent or cultural nuance. The key is choice. By examining AI’s output, students learn to choose which suggestions serve their purpose and which betray it. This echoes Cummins and Early’s (Reference Cummins and Early2011) advocacy for identity texts: writing that reflects learners’ own linguistic and cultural selves. Our AI literacy practice aligns with this: It gives learners the power to author themselves in English, deciding when to use the colloquial “I learn fast and work hard” and when to retain the polite “I may be suitable.” Without such critical practices, students might easily mistake nativeness for superiority, losing sight of the goal of context-responsive communication. After all, as experts (e.g., Kachru, Reference Kachru1992; Pennycook, Reference Pennycook2017) in English as an International Language note, learners do not need to internalise every cultural norm of native speakers to communicate effectively. Clarity, appropriateness, and authenticity often matter more than sounding like a generic native speaker.

3. Illusions of fluency: mirrors and markets

If dependency on instant answers erodes internalisation, and if chasing native-like perfection erases nuance, then AI’s cruellest trick is fluency without foundation. In China’s ELCs – where “short-term visible gain” is the business model – AI is the perfect accomplice. A messy essay becomes a model composition in seconds; listening comprehension tests feel effortless with AI-generated transcripts and translations. Parents see results, institutions showcase progress, and students walk away believing they are suddenly fluent.

But this fluency is a flattering mirror. It reflects not who the students are, but who they wish to be. I have watched learners beam with pride after producing AI-assisted writing, only to falter when faced with ordering a simple coffee abroad. “The language centre lied to me,” one former student complained. What he misread as personal failure was in fact systemic: an illusion produced by the marriage of commercial pressure and technological seduction. As reflected in Pennycook’s (Reference Pennycook and Pennycook1998) notion of English as commodified practices and Xiong and Yuan’s (Reference Xiong and Yuan2018) discussion of neoliberal language learning, English learning in such contexts often becomes commodified – a purchasable identity rather than a lived linguistic practice. This illusion aligns neatly with neoliberal logic: Institutions sell acceleration, students consume polish, and families pay for the appearance of proficiency. Yet the collapse of this house of cards is inevitable, and with it comes disillusionment – not only with the school but sometimes with English itself.

Norton’s (Reference Norton2013) notion of investment in language learning helps explain this disillusionment. Learners invest in acquiring a language when they anticipate meaningful returns – be it social mobility, peer acceptance, or career opportunities – for the effort they put in (Darvin & Norton, Reference Darvin and Norton2016; Norton, Reference Norton2013). In other words, students commit themselves to learning English in hopes of gaining cultural capital or tangible benefits that justify their hard work. If AI-generated fluency yields only hollow returns (e.g., a perfect essay that students themselves cannot replicate or understand), the learner’s sense of investment can collapse. Why continue investing real effort when the payoff was an illusion? In such cases, students may become demotivated or even abandon the project of learning, feeling, as Norton (Reference Norton2013) would put it, that their imagined future with English has been snatched away. The tragedy is that the market cashed in on their hopes, providing just enough of a mirror to win the sale, but not enough substance to deliver the promised mobility.

To counter these illusions, I instituted “AI-Free Mondays.” On those days, phones stay in bags, and the only tools are human voices and maybe a pencil. The tasks were simple, even mundane: narrate a childhood memory to your partner, debate the merits of your favourite breakfast, ask a classmate for directions to a local landmark. The first attempts were halting, broken, filled with laughter at mistakes, and groping for words. But then something real surfaced: eyes meet eyes, hands sketch shapes in the air, and laughter stitches the silences. As Swain (Reference Swain and Byrnes2007) reminds us, output – especially imperfect, negotiated output – is crucial for language development because it pushes learners to notice gaps and construct meaning collaboratively.

In those unaugmented conversations, students discovered that communication is more than the sum of grammatically correct sentences – it’s a human exchange, complete with hesitations, gestures, and the negotiation of meaning. These raw, imperfect moments carried more pedagogical value than a dozen flawless AI-generated paragraphs. Students left the class realising that fluency isn’t the same as perfection on paper. Fluency means you can stumble through explaining your idea and still be understood – and that you can understand someone else in turn (Usler, Reference Usler2022). The confidence built on this foundation is rock solid, not the brittle shine of an AI mirror.

4. The silent classroom and the tempted teacher

AI is not only reshaping students; it is reshaping us as teachers. There are days I feel the seduction myself: paste a draft into an AI, get neat corrections and feedback, and move on to the next task. The efficiency hums, fatigue recedes. But the more I outsource my teaching senses, the more students ask – sometimes innocently, sometimes pointedly – “What do we need you for?”

This question is both a threat and an invitation. If AI can dispense correct English faster than I can, then my value must lie elsewhere. Indeed, educators are coming to realise that our role is shifting from being mere dispensers of language rules to becoming sense-makers and facilitators of learning (Cheng, Reference Cheng2014).

Yet in China’s ELCs, where education often operates under commercial imperatives (Yan, Reference Yan2025), this human role is constantly tested. When students’ visible progress must be demonstrated to satisfy parents and sustain enrolments, the temptation is strong to outsource the hard work of learning to technology. But it is precisely in such market-driven contexts that the human teacher becomes indispensable. What AI lacks is not data but the capacity for connection – the empathic attunement, intuitive judgement, and ethical discernment that anchor genuine teaching. Empathy allows teachers to perceive what algorithms cannot: the emotional undercurrents of learning – the anxiety before speaking, the frustration of misunderstanding, and the quiet pride of being understood. This emotional literacy is vital in private centres where students’ confidence can determine retention rates. Teachers’ intuition, what Schön (Reference Schön2017) called “practical wisdom,” guides the micro-decisions that sustain engagement: sensing when to slow down a discussion, when to turn confusion into laughter, or when to gently redirect a student overwhelmed by performance pressure. Ethical judgement, too, is indispensable. In a system where education can easily slide into performance metrics and consumer satisfaction, teachers mediate between institutional demands and learners’ long-term growth (Noddings, Reference Noddings2012). They make daily decisions about when to prioritise authentic learning over speed, fairness over convenience, or depth over marketable results. Most of all, trust remains the invisible currency of education. As Bryk and Schneider (Reference Bryk and Schneider2002) argue, learning thrives on relational trust, not transactional promises. In private ELCs, where parents are paying customers and teachers often face evaluation through satisfaction surveys (Zhou et al., Reference Zhou, Nguyen and Dang2025), maintaining authentic trust is both fragile and vital. Students take intellectual risks only when they feel safe; they persist only when they believe their teachers care beyond the report card. No algorithm, no matter how refined, can replicate the subtle choreography of eye contact, tone, humour, and presence that nurtures that trust.

Yet recognising what AI cannot do is only the first step. The deeper question is: How should we, as teachers, engage with what AI can do? If trust, empathy, and judgement mark our irreplaceable strengths, then our challenge is to harness AI without surrendering those very qualities. The answer, I have come to believe, lies in mediation – in positioning ourselves not against technology, but between it and our learners.

As one educational leader once remarked, “Who will teach students to use AI wisely? It will be their teachers, not the AI itself.” In this sense, the teacher’s task is not to resist AI, but to mediate it. In Vygotsky’s (Reference Vygotsky1978) terms, mediation describes how learning is shaped by social interaction and cultural tools. Teachers act as human mediators, guiding students to interrogate AI’s feedback rather than consume it uncritically. When I ask my students, “Why might the AI suggest this phrasing?,” I am not only prompting linguistic reflection but cultivating a habit of mindful engagement – helping them see that understanding language involves choice, not compliance. Perhaps that is where our new value lies: not in outpacing AI, but in humanising it – grounding digital efficiency in ethical awareness, empathy, and shared meaning. In a market that prizes fluency as a commodity, the teacher’s quiet resistance is to make learning personal again.

There is also an equity dimension to consider. In theory, AI could democratise access to knowledge (Vesna, Reference Vesna2025) – it’s like having a tutor available 24/7. But in practice, without teacher guidance, it may widen the gap. For example, students in big cities with premium programmes get the latest tools and explicit training on how to use them; those in smaller towns often get neither. The well-guided become gracefully augmented; the unguided become quietly confused. This reflects a broader concern educators have raised: The AI revolution in schools could exacerbate digital divides if we’re not careful (Vesna, Reference Vesna2025). Disparities in income, access to technical devices, and even English learning backgrounds mean that some learners benefit from AI tools, while others are left behind or misled. As teachers, we must actively work to close this gap. That might mean lobbying for resources so all students have access, or it might mean creating additional support for those who don’t have tech at home. It certainly means teaching critical digital literacy to every student, not just the privileged few. In short, our value now is not in outspeeding ChatGPT at grammar suggestions, but in being the mentor who ensures that every student can ride the new escalator safely, without missing the view and lessons from the stairs.

For myself, I’ve taken this as a call to refocus on the human core of teaching – not only to bring what AI cannot but also to ensure that its benefits do not deepen inequality among learners. For instance, I make a deliberate effort to notice when a usually talkative student grows quiet, and later check in to learn that the student has been having a difficult week. I design moments that build community – like when the whole class applauded a student who bravely delivered her first presentation without AI support, celebrating courage and progress over polish. I try to make space for humour, vulnerability, and dialogue, because these are what turn classrooms at ELC from transactional spaces into communities of care and shared meaning. Such small acts remind me that empathy is not simply a feeling, but a pedagogical practice – one that keeps teaching profoundly human.

Yet empathy alone is not enough. Within the same ELC, students come from vastly different family backgrounds – some have laptops, stable Wi-Fi, and parents who encourage them to explore AI tools, while others rely on shared devices or limited access. Their starting points in AI literacy are therefore unequal. To truly “humanise” AI, we must also work to democratise it. That means making sure every student, regardless of family wealth or digital familiarity, can benefit from AI as a learning partner rather than be left behind or misled by it. Sometimes, this means designing activities that don’t assume access to technology at home; other times, it means explicitly teaching critical digital literacy, so that all students – not just the privileged few – learn to question and navigate AI responsibly.

5. Conclusion: holding two truths

I return to that mother’s whisper: “This is worth every RMB.” She was not wrong, in a sense. AI had given her daughter a quick taste of polish and a boost of confidence. In a results-driven system, those things do matter. But my role as a teacher is to ask the longer-term question: will that confidence survive outside the mirror? Will it endure in the wild, messy reality of using English in real life, where there is no AI whispering perfect phrases in your ear?

After this journey, I now hold two truths with equal conviction. First, AI is a remarkable scaffold. It lowers barriers, broadens input, and accelerates practice in ways I couldn’t have imagined a decade ago. It can democratise learning by providing personalised support at scale, and it can free up class time for more meaningful interactions. Second, AI is a seductive crutch. Its very strength – delivering instant, effortless assistance – can lull teachers and students into complacency. It can also flatten voices, anesthetise error, tempt institutions to sell shine over substance, and even risk turning classrooms into theatres of spectators rather than workshops of active learners. Both of these truths coexist.

My task now is one of calibration and balance. I need to let AI support students without replacing their own thinking. I must use the mirrors AI provides without mistaking them for windows. Above all, I keep reminding students (and myself) that fluency is not about sounding perfect, it’s about sounding human. The goal is not to beat the Turing test (Turing, Reference Turing1950); it’s to communicate your ideas in your voice, as clearly and authentically as possible.