AI and the Teaching of Reading Comprehension Part 2

The world’s all atingle with artificial intelligence (AI).

OpenAI reports that ChatGPT serves 700 million users each week. Seventy-eight percent of businesses and other organizations depend upon it, too. Private investment in AI is more than $100 billion in the U.S. alone.

Given that, it shouldn’t be surprising that the nose of this technology camel is poking under the Education tent. Even when educators eschew it, students themselves find ways to bring it in.  

In my last blog entry, I decried some uses of AI in the teaching of reading comprehension. My concern was that these would do more to undermine kids’ progress than to facilitate it.

Much of my criticism was based on a fundamental truism: Nobody can learn for you. We all must do our own learning.

When students read a text and summarize it, they may or may not learn something. But when AI does this summarizing, a lack of learning is guaranteed.

Think of some analogous non-tech examples:

Mom finds out about Janey’s homework assignment at bedtime. She insists her little darling sleep, but has an older sibling complete the work so Janey won’t get in trouble. I don’t know if the big sister will learn something, but it’s certain it will be a zero for Janey.

Butch, the hulking football star, bullies a nerdy classmate into completing his book report. You don’t need an fMRI scan of their brains to recognize that in the game of learning, Butch fumbled.

Even when students seemingly do the work, AI may manage to prevent learning.

Reading comprehension instruction tends to focus too much on how to answer certain kinds of questions and not enough on how to understand text. Question types don’t explain variance in reading achievement, but differences in text complexity do (ACT, 2006). If kids can make sense of text, they can answer any type of question.

Comprehension instruction should focus on world knowledge, complexity and extensiveness of content, vocabulary, syntax, cohesion, discourse structure, literary devices, data presentation devices, tone, bias, genre differences, and so on. Readers need to know how to negotiate these text features (Shanahan, 2025).

That’s why the use of AI to revise texts to facilitate comprehension is such a bad idea. The ability of AI to reliably transform a fifth-grade text into a third grade one without losing or damaging content has not yet been proven. But even when it manages to accomplish that, it strips away or transforms those text features students should be learning to deal with.

Altering the readability of a text – when done well – may ease reading comprehension. But such alterations reduce the possibility that the reading will contribute to improved comprehension.

That might make me sound like a technophobe, striving to keep those nasty modern technologies out of our schools; a charter member of the Luddites Are Us Club.

Nothing of the kind.

Students must do certain things if they are to become good comprehenders. They need to read lots of texts and use information gained from that reading: discussing it with others, writing about it, applying it to problems. The texts they read for this purpose need to be challenging – not much is gained from focusing on texts that can already be read reasonably well. Teaching should help students learn how to surmount text features that may block understanding, and as they gain purchase of those barriers, they become better readers.

That, I believe, is a good place for artificial intelligence to come in. Instead of helping kids to evade and avoid those challenges, it should help teachers to guide kids to deal with them.

When I speak to teachers about teaching reading with complex text, I offer examples of some of those barriers and the kinds of teaching that would help students to solve the texts.

Almost always, I’m asked by teachers (some displaying considerable embarrassment), how can I know which sentences (or cohesive links) might trip up students’ comprehension? I usually give an answer like: “Look for especially long sentences, sentences with a lot of internal punctuation, sentences with multiple clauses or phrases, sentences in passive voice, and so on.” They often nod skeptically, revealing uncertainty, a lack of confidence.

I’ve decided that a better answer is to use artificial intelligence. Teachers don’t have the time to do a thorough analysis of these kinds of text features, and AI can do a reasonably good job of identifying them. With that assistance, teachers can choose which ones they want to address, considering their students and their time constraints.

For instance, I entered into ChatGPT the text, “How the Camel Got His Hump” by Rudyard Kipling, and asked, “What are the five most complex sentences – sentences that might confuse young readers?”

It spit out five sentences, along with explanations as to why they were chosen. Here are three of them:

“In the beginning of years, when the world was so new and all, and the Animals were just beginning to work for Man, there was a Camel, and he lived in the middle of a Howling Desert because he did not want to work; and besides, he was a Howler himself.”

“So he ate sticks and thorns and tamarisks and milkweed and prickles, most ‘scruciating idle; and when anybody spoke to him he said ‘Humph!’ Just ‘Humph!’ and no more.”

“At the end of the day the Man called the Horse and the Dog and the Ox together, and said, ‘Three, O Three, I’m very sorry for you (with the world so new-and-all); but that Humph-thing in the Desert can’t work, or he would have been here by now, so I am going to leave him alone, and you must work double-time to make up for it.’”

All five would be potentially useful for syntax teaching, and Kipling tales are particularly good selections for a heavy emphasis on syntax. In this case, I wanted to emphasize the third sentence. ChatGPT’s explanation for the choice is that it includes dialogue embedded in a long sentence, several clauses joined by commas and semi-colons, and the figurative phrase, “Humph-thing.”

I’d ask the kids, “Why was the Man angry at the Humph-thing?” or “Why did the Man apologize to the Horse, Dog and Ox?” or “Why might the Horse, Dog, and Ox dislike the Camel?” Those questions can be answered with information from that sentence. If they can answer, then the sentence was not a barrier to comprehension. If they could not, then I’d teach them how to make sense of it, focusing on the quotation marks, figuring out how to break it into parts, and so on.

I wasn’t surprised that AI could identify complex sentences and uncover what made them complicated and potentially confusing. I didn’t ask it to suggest comprehension questions for those sentences. Experience tells me it can do that well.

What about potential barriers that may be harder to determine, like cohesive links? These are important because they connect the ideas in a text.

I entered an episode of Antoine de Saint-Exupéry’s “The Little Prince,” and asked, “Can you show me some of the cohesive ties in this text?” It provided several so I followed up with, “Do any of these appear to be especially difficult for readers to recognize?”

It nominated several possibilities. For instance, consider this complex linkage of ideas:

“I cannot play with you,” the fox said. “I am not tamed.”

“Ah! Please excuse me,” said the little prince.

But, after some thought, he added:

“What does that mean — ‘tame’?”…[8 intervening sentences]…

“It is an act too often neglected,” said the fox. “It means to establish ties.”…[48 intervening sentences]…

If you want a friend, tame me…”

To make sense of this exchange, readers must connect tame, establish ties, and friendship, something hard to do given the distance between their mentions in the text.

My AI helper also pointed out some reappearance of concepts in this text without exact repetition: unique-important, men-hunters, and ellipsis (omissions of words the reader must fill in): “‘On another planet?’ ‘Yes.’” (A full sentence response would be: “Yes, it is on another planet.”)

Those were helpful, but I was especially pleased with the identification of parallel structure or pattern repetition in this text. In this short episode, the terms “to me” and “to you” are repeated 9 times. ChatGPT pointed out that this reinforced text cohesion throughout through rhythm and structure.

For me, this structure emphasizes that this conversation is a meeting of the minds of two strangers who come together from very different perspectives. They are trying to communicate and connect, but they also want to maintain their view and to make certain the other recognizes this. By reiterating that what is being stated is my opinion or my point of view (“to me”) makes this clear. This is a wonderful example of the structure of the text conveying or reinforcing the information that is being stated in the words themselves. Kids need to learn that authors repeat themselves for a reason and it is worth noting those repetitions and thinking about the reason for them.

I made some other requests of AI, too, asking it to identify multiple themes or tone (and what words or structures the author used to convey tone). Tone refers to the atmosphere of a text and reveals an author’s attitude towards the subject at hand or the characters. A text may be funny, sad, playful, nostalgic, frightening, authoritative, and so on.

In all these cases, AI provided more than I could possibly teach in one set of lessons. I liked having the options, and, of course, by querying AI more specifically, I may end up with fewer choices but ones particularly relevant to my curricular goals. Even with that, a teacher will need to make the choices. Perhaps the items that AI identifies will be too easy for your students or irrelevant to your goals.

As you can see from these examples, artificial intelligence can be a useful tool for helping develop more powerful comprehension lessons. Instead of using it to remove barriers to comprehension, use it to identify the barriers and then show students how to surmount them.

When it comes to reading comprehension, our job isn’t to smooth their way, to avoid and evade difficulty. No, we are responsible for making students powerful and independent, teaching them how to recognize and transcend linguistic and conceptual obstacles. 

AI has a role to play in the teaching of reading comprehension. But that role should involve it as an assistant to the teacher, not to the reader.

References

ACT, Inc. (2006). Reading between the lines: What the ACT reveals about college readiness in reading. Iowa City: Author.

ChatGPT, personal communication, May 5, 2026.

de Saint-Exupéry, A. (2018). The little prince (I. Testot-Ferry, trans.). Ware, England: Wordsworth Editions.

Kipling, R. (1902/2001). How the camel got his hump. New York: NorthSouth Books.

Shanahan, T. (2025). Leveled reading, leveled lives. Cambridge, MA: Harvard Education Press.