How AI Helps Teachers Differentiate ESL Speaking Questions by Level

My Thursday class has 18 students spanning A2 to B2. Same room, same teacher, three very different proficiency levels. When I ask a B2-level discussion question, my A2 students stare at their desks. When I simplify for beginners, my advanced students check their phones. AI-generated leveled questions changed this dynamic completely.

What Differentiated Speaking Questions Look Like

Same topic -- let's say "food and cooking." Different questions for different levels:

• A2: "What's your favorite food? Do you cook at home?" -- Simple present tense, familiar vocabulary, yes/no or short answer
• B1: "Describe a meal you recently cooked. What ingredients did you use and how did it turn out?" -- Past tense, sequencing, more specific vocabulary
• B2: "Some people argue that cooking skills are becoming obsolete because of food delivery apps. Do you agree? What might we lose as a society if fewer people cook?" -- Abstract thinking, argumentation, conditional structures

All three groups discuss food. But the cognitive and linguistic demands match their abilities.

How AI Makes This Practical

Without AI, creating three versions of every discussion takes triple the prep time. Nobody has time for that. With ChalkLab, I generate the same topic at three levels in about two minutes total. Three separate prompts, three outputs, done.

The tool automatically adjusts vocabulary complexity, sentence structure expectations, and cognitive demand. An A2 question won't include subjunctive mood or abstract reasoning. A B2 question won't insult students with "What color is your favorite food?"

Managing It in the Classroom

I print three colored sheets -- green for A2, yellow for B1, blue for B2. Students know their color. During pair work, I pair within levels so both partners can engage fully. For whole-class discussion, I call on students with questions appropriate to their level. Nobody knows the questions are different unless they compare sheets.

Some teachers prefer to project one question at a time, starting with the easier version and building up. Advanced students answer the harder follow-up while beginners discuss the initial question in pairs. Both approaches work.

What AI Gets Wrong About Differentiation

Sometimes the "A2" questions are actually too easy -- factual questions that don't generate real conversation. "Do you like pizza?" isn't a discussion question; it's a survey. You still need to check that even the simplest questions require more than a one-word answer.

The B2 questions occasionally lean too academic. "Analyze the socioeconomic implications of food deserts" isn't conversation -- it's a thesis prompt. Scan for this and dial it back when you see it.

The Bigger Picture

Differentiated speaking questions aren't just about difficulty. They're about dignity. Every student deserves to feel challenged but not overwhelmed. When an A2 student successfully discusses their cooking routine and a B2 student debates food policy -- in the same classroom, at the same time -- that's good teaching. AI just makes it logistically possible. For more differentiation strategies, see my full differentiation guide.