Skip to content
Try Edumo!

Resources > AI Guide > Chapter 1

Chapter 1: AI and Language Teaching

Before we get into prompts and tools, it's worth spending a few minutes on what AI actually is, what it can genuinely help you with, and where you should keep your expectations firmly grounded. Getting this mental model right will save you a lot of frustration later.

What AI is

When we say "AI" in this guide, we're mostly talking about large language models as experienced in tools like ChatGPT, Claude, and Gemini. These are programs that have been trained on enormous amounts of text. They're very good at predicting what words should come next in a sequence, which makes them surprisingly good at generating text that reads like it was written by a human.

Although the providers are building in capabilities that resemble thinking, they don't understand language the way you do. They don't understand your student the way you do. They are very sophisticated text generators, and that's both their power and their limitation.

What AI is good at

For language teachers, AI excels at tasks that involve generating or transforming text. Things that may eat your time but don't require deep knowledge of your specific student or your judgement:

  • Generating content at scale. Need a reading text about renewable energy at B1 level? A set of vocabulary exercises for legal professionals? A role-play dialog between a hotel receptionist and a guest? AI can produce these in seconds.

  • Adapting existing materials. You have a great article but it's C1 and your student is A2? AI can simplify it while keeping the core ideas. You have a vocabulary list but need it turned into fill-in-the-blank exercises? AI can do that too.

  • Handling repetitive creation tasks. Making five variations of the same exercise type for different topics. Generating vocabulary lists for different professions. Creating comprehension questions for each chapter of a reader.

  • Working across languages. AI can generate content in dozens of languages, create bilingual materials, explain grammar concepts in a learner's native language, and identify cognates and false friends between language pairs.

What AI is bad at

AI is generally weak at things that require understanding people, making judgments, or being reliably correct:

  • Knowing your student. AI doesn't know that Maria gets anxious when she doesn't understand something, that Thomas only practices when the material is about football, or that Yuki has been struggling with articles for six months.

  • Making pedagogical judgments. Should you push this student harder or slow down? Is this the right moment to introduce the subjunctive? Is the student ready for authentic materials or do they need more scaffolding?

  • Being consistently accurate. AI can make things up. The technical term is "hallucination," which means it generates text that sounds confident and correct but is factually wrong. In language teaching, this can mean: answer keys with incorrect answers, grammar explanations with subtle errors, vocabulary definitions that are slightly off, and CEFR level judgments that miss the mark.

  • Understanding culture and nuance. Language is deeply cultural. AI can generate a formally correct business email in Japanese, but it might miss the subtleties of keigo (honorific language) that a native speaker would catch. It can produce a dialog for a German doctor's visit, but the social dynamics might feel off.

These are the things that make you a teacher, not just a content source. AI generates text. You understand the person in front of you, make decisions based on that understanding, and create the conditions for learning to actually happen.

The elephant in the room: "Will AI replace me?

After talking to over 100 language teachers, I can tell you this: not a single one has been replaced by AI. But a significant number are worried about it. And if you've been teaching for a while, you've probably heard similar promises before: interactive whiteboards, MOOCs, language apps. Each one was going to transform everything, but the profession is still here.

AI is different in scale. That's worth acknowledging honestly. AI conversation tools are getting better. Apps like Gliglish and Talkpal can now hold reasonable conversations in many languages. Duolingo has added AI-powered speaking practice. These tools give learners something that was hard to get before: unlimited conversation practice at any hour. That's genuinely valuable, and it goes beyond what traditional language apps offered.

But they can't build a genuine relationship with (most) learners. Understand why someone keeps making the same mistake and find a new way to explain it. Know when to be encouraging versus when to challenge. Adapt a lesson on the fly because the student walked in stressed about a job interview tomorrow. Celebrate genuine progress. Be a real person who the learner doesn't want to disappoint.

Learning a language is a deeply human process. It involves vulnerability, trust, motivation, and connection. A learner doesn't persist through the frustrating intermediate plateau because an algorithm told them to. They persist because they have a teacher who knows them, believes in them, and holds them accountable.

The teachers who are thriving with AI are using it as an assistant, to do the repetitive, time-consuming parts of their work faster. They create materials in minutes instead of hours. They generate customized content for each student's profession and interests. They spend less time on prep and more time on the part of teaching that requires a human: the teaching itself. If anything, teachers who learn to use AI effectively become more competitive, not less.

That's the mental model I'd encourage: AI is your prep assistant. A very fast, occasionally wrong, never tired prep assistant that you still need to supervise.

Things to watch out for

A few specific things to be aware of as you start using AI for teaching materials:

  • Hallucination in answer keys. This is the biggest practical risk. AI-generated exercises sometimes come with incorrect answer keys. A fill-in-the-blank exercise might accept a wrong preposition. A grammar exercise might mark a correct answer as wrong.

  • CEFR level drift. You ask for B1 and get B2. Or you ask for A2 and the vocabulary is clearly more advanced. AI has a rough sense of CEFR levels but it's not calibrated the way a trained teacher would be.

  • Inconsistency between runs. The same prompt can produce different quality outputs each time you use it. Monday's vocabulary exercise might be excellent; Tuesday's might be mediocre. This is normal. It's how these models work. If the output isn't good enough, try again or refine your prompt.

  • Cultural assumptions. AI tends to default to certain cultural contexts (often American English, Western European social norms). If your student's context is different, you may need to specify this explicitly in your prompts.

  • Data privacy. Be thoughtful about what information you share with AI tools. Some services use your input to train their models. Avoid putting personally identifiable learner information into prompts. "My B1 student who is a lawyer interested in contract law" is fine. "My student Maria Rodriguez at legal firm X" is not necessary and not wise.

Given these limitations, it's a good habit to always skim and review AI-generated content before sharing it with learners. Check answer keys, verify the difficulty level matches what you asked for, and make sure no personal information has slipped into your prompts. Think of it like reviewing the work of a fast but occasionally careless assistant.