Reading comprehension is one of those things that looks simple to teach but is surprisingly hard to do well. The standard approach — find a text, write some questions, mark the answers — works, but it scales poorly, doesn't adapt well to different learner levels, and eats up more prep time than most teachers have.
AI changes that equation. But not in the way a lot of articles suggest.
This guide isn't about prompting ChatGPT to spit out a generic worksheet. It's about using AI thoughtfully to build reading comprehension exercises that are genuinely level-appropriate, rooted in real texts your students actually want to read, and designed to measure understanding rather than just recognition.
Before getting into the how, it's worth understanding what makes comprehension exercises fail — because AI can replicate those failures just as easily as a tired teacher at 10pm.
The most common problems are:
Questions that test reading, not understanding. "According to the text, what did Maria do on Tuesday?" is a retrieval question. It checks whether a student looked at the right sentence, not whether they understood the passage. True comprehension exercises ask students to infer, synthesize, or evaluate — not just locate.
Texts that have nothing to do with students' lives. A B2 English learner in Hungary doesn't necessarily find a passage about New England autumn particularly engaging. The more disconnected the text is from a student's world, the more cognitive load goes into just caring enough to read it carefully.
One-size-fits-all difficulty. A single worksheet sent to a class of 25 students hits some perfectly, bores others, and loses the rest. Differentiation is almost always an afterthought.
AI, used well, can fix all three of these. Used carelessly, it reproduces them.
Teachers talk about CEFR levels or Lexile scores, but in practice, appropriate difficulty has three layers:
Most readability tools only measure the first two. AI can help you address the third, which is often the real barrier for students — especially in language learning contexts.
A text about quantum physics written in simple sentences is still hard. A text about family dinner written with some complex vocabulary is often manageable. When you're generating or adapting texts with AI, keep all three dimensions in mind, not just word frequency.
You have two options here: adapt an existing text, or generate one from scratch. Both work; the right choice depends on your context.
Adapting a real text tends to produce more authentic reading material. Students benefit from encountering real-world language — idioms, genre conventions, register — rather than AI-smoothed prose that is technically correct but somehow lifeless. A short news article, a product review, an excerpt from a book, or even a social media thread can all work.
To adapt, prompt your AI tool with something like:
"Here is a text [paste text]. Rewrite it at a [B1/grade 5/age 11] reading level. Keep the core meaning and main ideas. Maintain a natural, authentic tone — don't oversimplify to the point where it sounds unnatural. Aim for sentences averaging 15-18 words."
Generating a text from scratch is more useful when you need something very specific — a topic that connects to a current unit, a particular genre you're teaching, or a scenario relevant to your students. The key is to give the AI a detailed brief:
"Write a 250-word informational text about [topic] for [age group / level]. Use vocabulary appropriate for [level]. The text should include at least one inference-able conclusion that isn't stated directly. Write in the style of a [short magazine article / blog post / letter / interview]."
That last line matters more than it seems. Genre shapes how students read, and exposing students to different genres — rather than always using the same "neutral passage" format — builds more transferable comprehension skills.
This is where most AI-generated exercises fall down. Left to its own devices, an AI will generate mostly retrieval questions because they're the easiest to verify. Push back on that.
Structure your questions across comprehension levels. A useful framework is the classic three tiers:
Literal (retrieval) — "What does the article say about X?" These are necessary but shouldn't dominate.
Inferential — "Why do you think the author chose to mention X at this point?" or "What can we conclude about Y, even though it isn't directly stated?" These require students to read between the lines.
Evaluative / applied — "Do you agree with the author's conclusion? Why or why not?" or "How does this connect to what we discussed about [related topic]?" These push students to think beyond the text.
When prompting for questions, be explicit:
"Write 6 comprehension questions for the following text. Include 2 literal questions, 2 inferential questions, and 2 evaluative questions. For the inferential questions, make sure the answers cannot be directly quoted from the text — students should have to reason to get there."
Then review what you get. AI will sometimes label an inferential question that is actually just retrieval. Read each question against the text yourself before using it.
This is where AI genuinely saves time. Once you have a core text and question set, you can generate variations in minutes rather than hours.
For lower-level learners:
For higher-level learners:
With AI, you can generate all three versions — simplified, core, extended — from a single prompt cycle in under 15 minutes. That's genuinely transformative for a subject teacher managing mixed-ability groups.
Reading comprehension and vocabulary knowledge are deeply intertwined. A student who doesn't know 8-10% of the words in a text will struggle to comprehend it, regardless of how good the questions are.
Consider adding a vocabulary task before the reading (not after). Pre-teaching 5-6 key words primes students to notice them in context, which deepens both vocabulary acquisition and comprehension. AI can generate matching exercises, gap-fills, or definition-writing tasks from your word list in seconds.
After the reading, a vocabulary-in-context task — "What does the word 'reluctant' mean in paragraph 2? How do you know?" — reinforces inference skills while consolidating new language.
A reading exercise that ends with question 6 misses an opportunity. Comprehension deepens when students have to do something with what they've read — talk about it, write about it, or connect it to something else.
Short, low-stakes output tasks work well:
These can be as short or as extended as your lesson time allows. The point is to close the loop: read → process → respond.
Redmenta's AIhandles much of this workflow in one place — generating the text, the questions, and the vocabulary tasks, and letting you edit everything before sharing with students. Because exercises are interactive and auto-graded, you get immediate data on which questions students found hard, which is useful feedback for your next lesson.
The differentiation step is particularly well-suited to the platform: you can create two or three versions of the same exercise, assign different versions to different student groups, and review results side by side without managing multiple separate documents.
AI-generated reading exercises are only as good as the review you put in. An AI tool can produce fluent, plausible-sounding content that is subtly wrong — a text with a factual error, a question with two defensible answers, or an "inferential" task that is actually just retrieval in disguise.
The time saved in generation should be reinvested in review. Read the text as your student would. Try to answer each question yourself. Check that the harder questions genuinely require thinking rather than scanning.
That review step is what separates an AI-assisted exercise from a good one.
