Artificial intelligence has arrived in the language classroom whether teachers invited it or not. Students at every level now have access to tools — ChatGPT, Google Translate, DeepL, Duolingo’s AI features, and many others — that can translate, conjugate, write essays, hold conversations, and even explain grammar rules in seconds. For world language teachers working with adolescents, this creates both an opportunity and a genuine challenge: AI can scaffold meaningful learning, but it can also silently do the learning for the student, leaving behind the illusion of progress without any real acquisition.
This post examines what the research and classroom experience tell us about AI’s role in second language acquisition (SLA), particularly for learners in grades 6–12, and then offers practical, structured approaches that teachers can use to design lessons and assessments that are AI-resistant — not because teachers fear technology, but because they understand that proficiency cannot be outsourced. But, as always, I am looking forward to learning from each other in the comments! How do you handle AI in your classroom?
Is AI Actually Useful for Second Language Learners?
The Case for AI as a Learning Tool
When used intentionally, AI-powered tools offer real benefits to language learners.
Immediate, low-stakes feedback. One of the most persistent challenges in SLA is the gap between instruction and meaningful practice. A student in a class of 28 might speak for only a few minutes per week. AI tools like language tutors and chatbots can fill that gap by offering unlimited practice opportunities outside of class. Research consistently shows that comprehensible input and output — the act of producing language and receiving immediate feedback on it — are essential for acquisition. AI dramatically expands how often a student can engage in both.
Differentiation and pacing. AI-based platforms can adjust to individual learners’ levels in ways a single teacher managing multiple preps simply can’t. A struggling student can revisit a grammar concept through an interactive AI explanation at 9 p.m. without embarrassment, while an advanced student can push toward more complex tasks. For middle schoolers especially, who are acutely sensitive to peer judgment, the private nature of AI practice reduces anxiety — a variable that research in SLA identifies as a significant barrier to acquisition.
Pronunciation and listening practice. Text-to-speech and speech recognition tools have become remarkably accurate. Students learning tonal languages like Mandarin, or languages with unfamiliar phonemes like French or Arabic – or Italian – benefit from being able to hear authentic pronunciation on demand and receive feedback on their own speech. This is something even an excellent human teacher cannot provide individually at scale.
Vocabulary acquisition support. Tools like spaced-repetition systems embedded in apps, or AI-generated vocabulary-in-context examples, align well with what cognitive science tells us about how vocabulary is learned: through meaningful, repeated encounters in varied contexts. AI can generate novel sentences using a target word, explain nuance, show usage in different registers, and quiz students in ways that reduce rote memorization.
Comprehensible input at the right level. Teachers often struggle to find reading and listening material pitched at just slightly beyond a student’s level (what Krashen called “i+1”). AI can generate custom texts — a news article about soccer simplified to a novice-mid level in Spanish, for example — that meet students where they are.
The Case for Caution
Despite these genuine benefits, AI presents real risks to SLA when it replaces rather than supports the learner’s cognitive effort.
Cognitive offloading undermines acquisition. Language learning is fundamentally a process that happens inside the learner’s brain through struggle, pattern recognition, and internalization. When a student submits a prompt to ChatGPT and pastes the response as their own writing assignment, no acquisition takes place. What Stephen Krashen called the “affective filter” — the mental state required to process and absorb new language — cannot be bypassed. And, indeed, when a student simply types in his/her native language (L1), then copies and pastes an AI-generated answer, the student usually does this without reading, let alone analysing, what the AI has returned. The student may turn in a polished paragraph, but they have definitely not processed language.
Over-reliance on translation. Even “helpful” uses of Google Translate can interfere with acquisition if students use it as their first resort rather than their last. Processing language through one’s native tongue does not build the direct L2 neural pathways that fluency requires. Students who habitually translate are building a crutch that becomes harder to abandon over time.
False confidence and grade inflation. When AI-assisted work earns good grades, students receive inaccurate feedback about their actual level of proficiency. This is particularly problematic in world language courses, where proficiency is cumulative — a student who hasn’t truly internalized present tense structures will struggle when the curriculum moves to the subjunctive.
Adolescent development and executive function. Middle and high school students are still developing the self-regulation and long-term thinking needed to use AI as a tool rather than a shortcut. Research on adolescent cognition suggests that when an easy route to a grade exists, many students — especially those under stress — will take it, even if they genuinely want to learn. This is not a moral failing; it is developmental reality, and it places the responsibility on us teachers to design tasks where the shortcut doesn’t lead to the destination.
The Nuanced Reality
The honest answer is that AI is neither the salvation nor the downfall of language learning — it depends entirely on how it is used. A student who uses an AI chatbot to practice ordering food in Italian, gets corrected on their errors (immediate feedback), and tries again has done something genuinely valuable. A student who has ChatGPT write a journal entry in Italian and submits it as their own has done nothing at all; in my opinion, doing that is actually an act of plagiarism. Our job, then, is to create conditions in which the former is more likely than the latter. But, how? Especially when it seems that we as tech users have not had any input on whether or not we can choose to use AI – they are already embedded in online search tools, in our email apps…they’re embedded everywhere it seems!
Designing AI-Resistant Lessons and Assessments
The goal here is not to wage a futile war against technology. AI detectors are unreliable, and any arms race between students and software is one teachers will lose (we are already overworked as it is!). Instead, the strategies below focus on designing tasks where authentic language use is the only path to success — where AI simply cannot do the work in a way that benefits the student.
Please let me know in the comments if you either have used these strategies in the past (and, if you’ve abandoned them, please share why!), or once you do implement any of these – let me know how it goes!
Strategy 1: Prioritize Spoken and Oral Assessment
This is the single most powerful tool in the world language teacher’s arsenal. AI cannot speak for a student in real time. I felt very vindicated about my take on oral assessments when I heard Neil deGrasse Tyson tell Stephen Colbert that “maybe, maybe education involves an oral exam.” (The Late Show with Stephen Colbert, October 2023)
Structured speaking assessments. Instead of written paragraph assignments, conduct brief oral interviews (even 3–5 minutes per student) using a proficiency-based rubric. Ask students spontaneous follow-up questions based on their answers — questions they could not have prepared with AI. “You said you went to the beach last summer — who did you go with, and was anyone in your family afraid of the water?” No AI preparation can anticipate that.
Timed in-class recorded responses. Have students record a 60- to 90-second video response to a prompt they see for the first time in class. The time constraint and spontaneous nature make AI use impossible.
Socratic seminars and fishbowl discussions in the target language. Even at the novice level, students can be asked to respond to each other’s statements. The interactive, unpredictable nature of conversation is the ultimate AI barrier.
Participation structures that assess process, not just product. Build into daily instruction speaking routines — interpersonal tasks, partner interviews, role-plays — that are assessed formatively over time. A student’s grade emerges from many small authentic interactions rather than a single high-stakes written product. One of my middle school colleagues uses TALK rubrics a lot, for formative as well as summative tasks. Another middle school colleague included this grade under “Habits” and asked students to keep track of their verbal interactions with peers. Is it behavioral? Yes, but it is also based on a student’s progress in producing language spontaneously – for me, this is valuable because it is a less performative use of Italian.
Strategy 2: Embed Personalization and Specificity
AI-generated content is generic by nature. The more personal and specific a task requires a student to be, the less useful AI becomes.
Assign essays, journal entries, or short-answer responses that require students to reference specific things from their own lives: their own family members’ names, a specific memory from a recent school event, an opinion about something the class discussed that day. Ask students to describe a photograph they took themselves. Have them write about their response to a piece of music played in class that morning.
These details are not in any AI’s training data. A student might still have AI write the grammar structures, but the product will be visibly generic and disconnected, which itself provides evidence for a conversation about authentic language use.
Strategy 3: Design Process-Based Assessments
Shift emphasis from the final product to the visible process of creating it.
Drafting in class. Require that a significant portion of any writing assignment be drafted in class, on paper or on a supervised device. The student brings handwritten notes or an in-class draft as the basis for any final polished version. The teacher can see the messy, real thinking of an actual learner.
Portfolio with reflection. Have students maintain a working portfolio of their language use over time. Include audio recordings from early in the year and late in the year. Ask them to reflect — in the target language, at an appropriate level — on what they can do now that they couldn’t do before. This longitudinal view is essentially impossible to fake with AI.Visible error as evidence of learning. Culturally reframe error as proof of authentic work. Celebrate a student’s creative attempt at a complex construction, even if it fails. When students know their imperfections are valued as evidence of real engagement, they are less incentivized to produce polished AI text.
Strategy 4: Use Unpredictable In-Class Assessments
Exit tickets in the target language. At the end of a class, students respond to a prompt on paper — a question about what they learned, a short description of the image just discussed, a brief reaction to the story they read. The in-class, pen-and-paper context makes AI irrelevant.
Rapid-fire written responses with a time limit. Give students 8 minutes to write as much as they can about a given topic. Fluency under pressure reveals authentic language ability and cannot be AI-assisted in real time without a device.
Dictation and transcription tasks. These old-fashioned techniques are AI-proof. Listening to target-language audio and transcribing it accurately requires genuine listening comprehension and spelling knowledge.
Strategy 5: Leverage Interactional and Performance Tasks
Interpretive-to-interpersonal sequences. After reading or listening to something, students immediately discuss it with a partner using a structured format, then respond to a teacher’s live questions. The chain — input, comprehension, social production — cannot be broken by AI.
Task-based language teaching (TBLT) activities. Design tasks with an information gap — where two students each have different pieces of information and must communicate in the target language to complete something together. Neither student can have AI do their half in real time.
Theatrical and presentational performance. Having students perform a short skit, present a project to a real audience, or teach a concept to a younger class creates conditions where embodied, spontaneous language use is unavoidable.
Strategy 6: Build Meta-Awareness and Honest Conversations About AI
Rather than treating AI use as purely forbidden, consider explicitly teaching students how AI can and cannot support genuine learning. Show them what a ChatGPT-written French paragraph looks like compared to an authentic novice’s writing. Discuss why submitting the former doesn’t prepare them for the exam, the travel experience, or the college interview in a foreign language.
Students who understand (and care about) what they are actually trying to build — neural pathways, not a grade — make better choices. In my opinion, after two years teaching middle school for the first time in my 20 year career, this is especially true for students in Grade 7 (more than my high schoolers who suffer more pressure to “get a good grade”), who are capable of genuine metacognitive reasoning when it is invited. The language classroom can be a rich site for teaching digital literacy and intellectual integrity as explicitly as it teaches vocabulary and grammar.
Conclusion
AI is neither the enemy nor the savior of the world language classroom. It is a powerful set of tools that, in the hands of thoughtful teachers and motivated learners, can extend practice, lower anxiety, differentiate instruction, and make the target language more accessible. But, in the hands of a student looking for a shortcut to a grade, it produces fluent-sounding text that represents a total absence of language learning.
As teachers, our response to AI should not be paranoia or prohibition. It should be good pedagogy: design assessments that value the authentic struggle of a real learner, prioritize speaking and interpersonal communication, embed personalization and process, and build a classroom culture in which genuine language use is the whole point. In that classroom, AI becomes a useful companion rather than a substitute for learning — and students leave with something no AI can give them: actual proficiency in another language.
Please share your thoughts in the comments! Have you used any of these strategies (or a version of them) in your classroom before? What are your recommendations? What do you think of AI in the Italian classroom?
Project-Based Italian 