A new kind of story has made its way into elementary school classrooms, joining the books in classroom libraries and the passages in early reading curricula: AI-generated text.
Tools designed for educators to use can rewrite articles to different reading levels, or generate decodable text—passages designed to give students practice with the specific letter-sound connections they’re learning in phonics lessons.
In a recent Education Week article, researchers and teachers discussed the potential pros and cons of adding AI-generated text to children’s reading diets.
The technology could give teachers a way to better differentiate for individual student needs. But text written by artificial intelligence has some marked differences from stories written by humans—and not all tools are equally skilled at crafting texts for instructional purposes, studies have shown.
AI can write stories and accompanying comprehension questions in minutes, said Jean Gunderson, a Title I reading interventionist in Elkton, S.D. It’s a task that used to take her hours.
Still, she said, she has to be careful, prompting the technology precisely and sorting through all the output. “The thinking about what I want takes longer than the creation of it,” she said.
For school leaders and teachers who use AI-written passages in lessons, researchers urged similar caution. Read on for a few common pitfalls to watch for.
1. AI writes in a particular voice
Studies of AI writing show the technology tends to take a middle-ground opinion on most topics, said Thomas Juzek, an assistant professor of computational linguistics at Florida State University.
This shift in tone can happen when AI has been asked to rewrite a passage, too, he said, “and without humans noticing.” Given this phenomenon, it’s possible that tools teachers use to rewrite text to a different reading level could also shift the text’s tone or meaning, he said.
AI agents also tend to give similar answers to the same prompt, Juzek said. This could pose a problem if teachers are hoping to receive a variety of texts that meet the same parameters, he said.
“One thing I would ask the companies: How do you ensure diversity in what your models produce?” Juzek said.
2. Some tools have trouble estimating text level
Most general-purpose AI tools struggle to write text at a requested grade level, a 2025 study found.
The researchers on that study asked several tools—including ChatGPT, Google’s Gemini, and Meta’s Llama—to create passages written at a 3rd and 9th grade level. At both grade levels, the text the AI produced was multiple grade levels higher than requested, as evaluated by Lexile levels and the Coh-Metrix, another measure of text complexity.
Linling Shen, a doctoral student at the University of Texas at Austin and the lead author on the study, has continued to research AI’s capability in this area, and said the models are more accurate now.
But getting them to return something a classroom teacher could easily use requires detailed prompting, she said. She usually includes a laundry list of constraints, specifying grade level, the frequency of specific words, and asking it to include some word definitions.
“It takes a much longer time for AI to process all the information, but the outcome is decent,” Shen said. “It’s much better than giving AI one prompt, ‘Give me a grade-level text.’”
3. Not all tools have a deep understanding of academic standards and curriculum quirks
Some general-purpose AI tools struggle to produce usable decodable text, passages that early readers use to practice their phonics skills, said Julie Highsmith, a 1st grade teacher in Florida’s Palm Beach County district.
When Highsmith and her colleagues asked the chatbots to write passages constrained to specific phonics patterns, the tools often got them wrong, including words with patterns students hadn’t yet mastered, she said.
Some developers of education-focused tools say their products are designed specifically to address this problem. They say they’re trained on high-quality decodable texts and taught to align to different curricula’s scopes and sequences.
Leaders who plan to use any kind of AI text generator for early reading practice should probe companies about this alignment, said Anne Murphy Karabell, a researcher at American University’s Baker School of Education who studies early-childhood literacy.
Karabell is also a member of Schools Beyond Screens DC, part of a national coalition that advocates for stricter guardrails around technology in the classroom.
“I’ve worked with Core Knowledge Language Arts, UFLI, Fundations—they’re different,” she said, naming several reading programs.
Regardless of assurances from companies, educators should always double-check AI output to ensure it works with their curriculum and addresses student needs, she said.
“We really should be looking at it, and making sure it’s meaningful, anticipating the words that children will have trouble with.”
