Should We Use AI to Generate Mentor Texts?

by Carl Anderson and Matt Glover

AI is evolving quickly, and we’re aware of its growing role in education.  While we’re not against teachers using AI to enhance instruction, we want to be careful about where and how teachers use it, what they gain when they do, and what they give up.

If you’ve spent time working with AI chatbots, you may have discovered that they can generate any kind of text quickly. Ask one, “Create a personal narrative about a day at the beach,” or, “Write a literary essay about why the character of Harry Potter is brave,” and within seconds, almost like magic, one appears on your computer screen, fully formed. You can even ask chatbots to write texts that include particular craft moves, such as transitions in essays or dialogue in realistic fiction.

It would seem that chatbots can help address the problem that teachers face of gathering mentor texts for the writing units of study. Right? After all, it takes time to gather mentor texts, time that busy teachers seem to have less and less of these days. And some genres we teach children are harder to find than others. 

However, for several reasons, we have strong reservations about using AI to generate mentor texts.

First, think about the meaning of the term, “mentor.”  A mentor is a more experienced person who helps us become better at an important skill. When we teach with mentor texts in classrooms, we name the author of the text as we teach with it. (“Let’s take a look at how Jacqueline Woodson wrote the lead to her story, Sweet Sweet Memory.) When possible, we share information about an author, to help students understand why they wrote the text (e.g. Jaqueline Woodson wrote Sweet, Sweet Memory in memory of her grandfather, Ganaar, who was a gardener.)  We do this to demystify writing for children, to help them understand that real people—like Jacqueline Woodson–write texts, which means that they can do so themselves. And we do this to support children’s identities as writers, which means they need to see themselves doing the same things that published authors do.

However, if we were to teach with an AI generated mentor text, there is no person behind the text! We can’t name the author, because there is none. We can’t share details about the author’s life that explain why they wrote the text, because there aren’t any. We can’t make the point that students can be writers like the author, because there is no author and, in fact, the text was written by an AI algorithm.

Second, we teach with mentor texts because we want students to learn to use the same craft techniques the authors use, for similar reasons.

If we were to teach with AI generated mentor texts, it would be difficult to authentically study the craft moves that these texts contain. Not because AI generated mentor texts don’t contain craft moves—the AI chatbot will search its memory for similar texts, and create ones with commonly used craft techniques, or we can tell the AI chatbot to write a text using specific craft techniques that we want to teach. The problem we would have is with the important discussion about why these craft techniques are used in the mentor text.  When we study a craft technique in a mentor text, we ask students, “Why do we think the author used this craft technique?  What effect did they think this craft technique would have on their readers?” 

For example, we might ask students, “In Sweet, Sweet Memory,” why do you think Jacqueline Woodson, used the technique of the “description list” to describe the setting of the story (“Somewhere an owl is hoo-hooing soft and loud. Under the porch there are field mice moving around. Maybe a frog or two.”) In response to this question, students can genuinely speculate that that Woodson used this description list to create a beautiful image of the setting in her readers’ mind.

We can’t have this conversation about an AI generated mentor text! That’s because the text was generated at our command, not because a real person wrote it. We can’t speculate about why the “author” used a craft technique because there was no thinking mind behind the text, and because the AI chatbot didn’t use craft techniques to effect readers, since AI chatboxes aren’t people who write for audiences! 

Third, if we ask an AI chatbox to create mentor texts that contain certain craft moves, then the texts will reflect only what we know already know about the genre. If we rely solely on AI to generate mentor texts, then we are missing the opportunity to learn more about the genre that we would have if we were to look for authentic examples where they are found in the world.  This is especially important for new teachers when they are building their knowledge base about writing.

For example, let’s say we ask an AI chatbox to generate several arguments that each have a claim, three reasons, a counterargument, and a conclusion, because that’s how we think arguments usually go. If instead we were to look for arguments in the world, we would discover that some have two reasons, some four (note that this argument has five!). And in some arguments, the author starts their lead with a counterargument, or writes a counterargument before a reason. By studying real examples of argument, we would not only deepen our knowledge of the genre, but we are then able to show students that there are many ways to write an argument—and that as writers, they need to choose the one they think will best help them argue their point.

Fourth, one of the powerful ways we teach with mentor texts is to point out the ways that students are becoming like the authors of the texts. For example, when we confer with mentor texts, both of us often say to students things like, “Wow, I see that you’re using description lists in your story, just like Jacqueline Woodson does in Sweet Sweet Memory!” This is powerful feedback for students. When we compare them to mentor authors who they love and admire, we help build their confidence as writers.  

If we were to teach with AI chatbot generated mentor texts, how can we have this kind of conversation with students?  What would we say?  “Wow, I see that you’re writing just like the algorithm in the AI chatbot?” In what way would this “compliment” have the powerful effect that comparing students to real, live authors that students admire has?

Finally, it’s important to consider the important role that mentor texts play in increasing student engagement in writing. When we teach with mentor texts, we try to select mentor authors who have things in common with the students we’re teaching, and who write about topics that reflect the identities and interests of our students. As Rudine Sims Bishop has written, texts should be “mirrors” for children that reflect their various identifies, their lives, and their interests, so that they realize that these identities, lives and interests are worth writing about.  

We are at a loss when we think about engagement and AI generated mentor texts.  Since there is no person behind an AI generated text, a student cannot be inspired by the author of the texts in the ways they can be inspired by reading a text written by authors like Jacqueline Woodson. After the careful and important work done over the last decade to include mentor texts in our teaching that are representative of the children in a particular classroom, using AI-generated texts seems like a step backwards to us.

We understand that teachers are pressed for time, and that using AI-generated mentor texts can help mitigate the time crunch. However, while it does take some time to find high quality mentor texts in the world, there are excellent sources for mentor texts that are readily available for teachers, such as their classroom or school library, or children’s magazines such as Cricket, Highlights, or Scholastic Storyworks. And we find that when teachers work together to find high-quality mentor texts, the process takes less time. (We describe ways that you can do this collaboratively in our book, How to Become a Better Writing Teacher.)

We also understand that some genres are hard to find. For example, where can you find high-quality literary essay for third graders? We suggest that in these cases, it’s better for you and your colleagues to write your own mentor texts, so that students are studying texts written by real people (you and your colleagues) who can authentically talk about what they did to write them, and why.  

And we also think that if a genre is hard to find, this may suggest you should consider that it not be a genre that you study with your students. For example, literary essay is almost never written for elementary age children as the audience. Since children rarely see it, then why are we studying it in the first place? 

Works Cited

Bishop, Rudine Sims.  1990.  “Mirrors, Windows and Sliding Glass Doors.”Perspectives:  Vol. 6, No. 3, Summer Issue.

Woodson, Jacqueline.  2007.  Sweet, Sweet Memory. Jump At the Sun: New York City.

Carl Anderson and Matt Glover are teacher educators who work with teachers around the world. They are co-authors of the new book, How to Become a Better Writing Teacher.