For the life of me, I really can’t see AI providing any real learning benefit to English-proficient “mainstream” students, and I’m not seeing any research that suggests otherwise. It can, however, help all teachers in selective ways, including in material preparation, writing letters of recommendation, and in data analysis.

There’s a slight caveat to what I said about student use, however. Though I haven’t seen any research on it for non-EL students, I think AI can be helpful to students and teachers alike when it’s used in low-or-no-stakes adaptive-learning practice platforms like Quill or in playing games like Wayground, Blooket, or Kahoot. I think those are excellent formative-assessment tools.

AI with English learners, however, is an entirely different story. In addition to their being beneficial for those low-stakes activities I just mentioned, AI tools can be extremely useful in providing opportunities for ELs to practice speaking—especially at home where there are often no other English speakers. Where there are no peer tutors present (even though I really don’t understand why any school wouldn’t use them), these same AI speaking tools can be helpful in class. AI tools like Google Translate can be essential for newcomers as communication tools, and something like Gemini Storybook can be helpful in creating high-interest reading materials. And when a newcomer is parachuted into an English-proficient class, AI-powered simultaneous translation tools can be a lifeline.

You might also be interested in The Best Posts About Using Artificial Intelligence With ELLs.

Today’s post is the first in a series in which educators will share their own experiences of the good and the bad about AI in the English-learner classroom.

‘AI Literacy Is the Answer’

Irina McGrath, Ph.D., currently serves as an assistant principal at Newcomer Academy in the Jefferson County public schools (Louisville, Ky.) and as an adjunct professor at the University of Louisville, Indiana University Southeast, and Bellarmine University.

Michelle Shory, Ed.D., is the multilingual engagement specialist at Seneca High School in the Jefferson County schools and an adjunct professor at Eastern Kentucky University and Indiana University Southeast.

Multilingual teachers began using generative AI to translate texts, adapt materials, and create unique experiences for their students as soon as ChatGPT came on the scene in November 2022. As tech enthusiasts, we also began experimenting with AI in our schools and classrooms, and on our website, ELL2.0. While our experiences have been overwhelmingly positive, we definitely have some concerns about using AI with multilingual learners.

Scaffolding

Using tools like Claude and ChatGPT has allowed us to translate materials in less common languages like Wolof and Mai Mai. Additionally, we’ve found that tools like Claude can provide translations with greater cultural nuance. When students can get a summary of a text in their native language or use parallel-text (side-by-side) translations, they are better able to focus on learning the content.

In addition to its translation capabilities, AI is an excellent tool for building background knowledge for multilingual learners. Students who are not familiar with certain science, math, art, or other academic content can use ChatGPT to gain a basic understanding of challenging or new concepts in their native language, as AI has the ability to explain content in many languages. Once that foundation is built, MLs are more successful in expanding their knowledge through English input from teachers, peers, and textbooks.

Image creators like Adobe Firefly and Nano Banana have been lifesavers—allowing educators to create custom images to support comprehension. AI-created videos, like those available in Notebook LM, are also extremely useful in the classroom for supporting comprehension and vocabulary development. They take a few minutes to create and make any lesson more engaging and accessible for students, especially multilingual learners, by combining visuals, audio, and key vocabulary in a clear, meaningful way. Notebook LM also creates impressive infographics and podcasts that can deepen background knowledge and pique interest in a topic.

Diffit and Brisk have enabled teachers to modify materials based on students’ proficiency levels and cultural backgrounds. When teachers use carefully crafted prompts, these AI-powered platforms can produce culturally relevant texts, as well as comprehension questions and writing tasks. This not only allows students to relate to the text more easily but also helps them learn about their peers’ cultures as they learn new content.

Student Engagement

While many educators are more comfortable implementing AI than allowing their students to use AI, we have become fans of teacher-created chatbots using School AI or Google Gems. These tools make lessons more engaging and draw students into the learning process. For example, multilingual learners can interact with historical figures such as Amelia Earhart, asking questions about her life, achievements, and challenges while practicing their reading and writing skills, expanding vocabulary, and developing inquiry and critical-thinking skills.

We have also seen that providing immediate and precise feedback has benefited our students. We recently used an AI formula in Google Sheets to quickly give writing feedback to over 300 MLs based on WIDA criteria and standards. This feedback gave students a current WIDA level score (1-6), a glow (strength), and a grow (opportunity) to further develop their writing. Additionally, we used Gemini to provide individual feedback for each English-language development teacher and to suggest a mini-lesson for each teacher’s students. Providing timely and prescriptive feedback motivates students and teachers.

Concerns

While there are many reasons to use AI with MLs, we have also seen a few downsides.

Cheating, of course, is one of the main concerns. We suggest creating AI contracts or having students disclose if and/or how they used AI on an assignment.

Another downside is that although AI can generate correct information, it can also produce false content. Multilingual learners who are still developing their content knowledge may not always detect these inaccuracies and may think the information is correct. This can lead to misunderstandings and require teachers to spend additional time clarifying and correcting the information.

In some cases, we have noticed that students become overly dependent on AI. Instead of engaging in the learning process themselves, they turn to artificial intelligence for quick answers. This can impede their academic success because many concepts build on one another. When a concept is not fully understood due to AI use, MLs may continue to struggle with future assignments.

Finally, AI literacy is the answer. The answer isn’t to ban AI or use AI detectors (which have proved time and again to be unreliable and to flag MLs more often). AI is here to stay, and when used responsibly, it can be a great ally. When students understand both the benefits and limitations of AI, they are more likely to use it to support their learning rather than rely on it for quick answers.

We Can’t ‘Out-Surveil These New tools’

Sarah Carr is a LIEP (language instruction educational program) department chair for the Henrico County public schools in Virginia, WIDA fellow, and an advocate dedicated to equitable instructional design for multilingual learners:

Another day in the ESL classroom. A student submits their assignment—a carefully designed assessment of language production and applied content knowledge—with a smile and an apology for being late. I quickly scan it, noting the flawless grammar and university-level vocabulary. “What in the ChatGPT is this?” I say, somewhat jokingly. “Nooooooo, miss! No ChatGPT!” I smile and shake my head.

Another day, another attempt by a desperate Tier A student to circumvent assignments designed specifically to meet them at their level. How can we, as educators, expect students to engage in a productive struggle as they continue to deepen their dependence on large language models?

These students’ drive is not laziness but a resourceful response. My students are typically recently immigrated, with low English proficiency. I cannot speak to the level of scaffolding in their other seven high school-level classes and understand why they may not trust that my assignment is one they will have the language to complete. Enter AI—a life vest in the stormy sea of language development.

It’s no wonder that students rely on it so heavily, just as they have relied on AI translation services for the past 15 years. But now the stakes are higher, as LLMs not only translate one’s thoughts but can completely replace the thinking process—a danger likely to go over the head of a teenager simply trying to survive their first or second year of high school in the U.S.

How do we mitigate this issue as educators? Currently, the response is largely reactive: AI detectors, verbal quizzes, and academic-dishonesty referrals. However, what if we used AI to our advantage by teaching students how to use it as a thought partner?

My state, Virginia, already has technology standards in place for integration in practice, including the use of AI. Additionally, my district has been proactive in developing guidance for educators to actively teach appropriate AI usage to students. All of this may be easily navigable for an English-proficient student, but what of our entering, emergent, and developing students? How can we actively design to support these principles in the ESL classroom?

Here is how I have been reshaping the way I use AI in my own classroom practice:

Design for proficiency levels based on comprehension : If a student believes the linguistic output is far out of their range, they will default to LLMs. This strategy relies on differentiation for various levels within the classroom.

Ground writing with oral-language production: Students must be able to answer questions about, describe, and defend their written work. By regularly incorporating speaking into assignments, students will implicitly understand that they must be able to spontaneously speak on a subject, encouraging real learning rather than avoidance. Furthermore, oracy can fill the gaps and help students demonstrate knowledge they don’t yet have the language to write.

Assessment as learning: For writing assignments in particular, the process is the product. Multistep research, peer reviews, frequent check-ins with small tasks not only help chunk a large task but encourage language production in reporting, evaluation, and analysis. Encouraging a culture of self-development and an emphasis on progress rather than perfection helps allay the fear that drives many students to misuse AI in the first place.

Explicitly teach and practice AI literacy: Rather than avoiding AI, teach students the skills to use it as what it is—a tool. AI can be used to provide examples of grammatical structures or to parse the nuances of various synonyms. It can provide an example of a specific type of writing structure or used as a thought partner to answer questions and assess answers. Evaluating the output of an LLM aligns with the Virginia standards of assessment and my district’s adopted Learner Profile and enhances critical thought.

Additional classroom uses could include using AI as a conversation or thought partner, simplifying complex texts to a student’s proficiency level for comprehensible input, viewing example sentences of vocabulary in context prior to writing their own, providing sentence starters or frames to support student writing tasks, or teaching students how to prompt AI for specific types of feedback on grammar, organization, or logic.

This shift in how we view AI use helps foster a healthier relationship with this technology, all while ensuring students are building agency rather than falling victim to the learned helplessness trap. When students learn the skills to actively engage with AI, they develop both language and critical thought.

There is no way to stop changes in technology, nor to out-surveil these new tools, so we must deliberately design learning environments centered around ethical, thoughtful use and make a place for our students to practice with intentionality.

‘I Had to Do Something’

Lisa Egle is an associate professor of English-language studies at Passaic County Community College in New Jersey and the founder of Mayafuna Learning, where she offers AI integration coaching and professional development for writing instructors and consulting for college administrators.:

The first semester after ChatGPT’s release, I saw my students’ writing change—and I knew what was happening. They were using AI. Some were just getting a little grammar help, cleaning up their writing, sounding more fluent. Others were misusing it as a ghostwriter. Across the board, none really knew how to use it well.

What I also knew was that I had to do something.

I teach writing and ESL at Passaic County Community College in New Jersey and I was already an early adopter, using AI in my own work before it hit my classroom. My students are multilingual adults, many of them immigrants juggling work, family, and school; some have just come out of high school.

Knowing my students and knowing the tool, I’ve spent the past two years redesigning my assignments and assessments around AI. The question isn’t whether. It’s how.

10 Tips for Teaching ELL Students to Use AI

Whether you’re teaching high school or college, AI is already in your classroom. Here’s what’s working in mine. Some of it might work for you, too.

Try a “go ahead and cheat” exercise. Tell students to take their draft to any AI and ask it to rewrite the entire thing to make it “perfect” or “sound like a native speaker wrote it.” Then share the results in class. They’ll hear right away that it isn’t them. That moment of recognition, of seeing their voice disappear, is more powerful.

Define cheating versus learning, together. Have the conversation with your students. One definition that works: mindless copy-pasting with no evaluation of AI’s suggested revisions = cheating. Using AI to identify errors, get feedback, and make informed decisions is learning. Let students help build that definition because then they’re more likely to respect the rules if they had a hand in writing them.

Show the difference between a bad prompt and a good one. Model various prompts live on screen or in a short video. Demonstrate how “fix my essay” and “make me sound fluent” lead AI to take over and ruin their voice while a good prompt gets targeted help that is their own writing but better.

Teach them to identify themselves as an ELL and add constraints. Have students start with: “I’m an English learner. My first language is _____.” Then add specific instructions:

“Show me where my coherence could improve, but don’t rewrite for me.”
“Where do I need to develop my ideas? Give me a few suggestions.”
“Does this sentence make sense? If not, explain why.”
“Are any of my errors being caused by my first language? Show me and explain.”

Have students ask for a visual display. Tables, bolding, and clear formatting make corrections and explanations easier to read when interacting with AI. When errors and suggestions are laid out visually, students have a better chance of learning from them.

Teach them to question everything. Tell students: “If the AI replaces a word, ask why. If it restructures a sentence, make sure the new version still says what you meant.”

Teach them to push back. “Why did you change that? Was it incorrect?” “That doesn’t sound like me. Can you speak to me on my level and in my voice?” Model this yourself by taking a student paragraph (with permission), running it through AI, live.

Build AI into your assignments. Make AI use part of the assignment structure, not an add-on. Clearly lay out where, when AI is allowed or required.

Require visibility and accountability. Ask students to record a short video showing their AI chat and explaining what they accepted, what they rejected, and why. They should also submit a link to the chat. Make both count toward the grade.

Also, remind students who is in charge. I tell mine: “At this company/in this classroom, I’m the CEO, and you’re a project director. The project: your writing assignment. AI is just an assistant, and he/she likes to take over. Don’t let that happen. Teach him/her how to work with you, and you’ll succeed.”

AI isn’t going away. The students I saw in that first ChatGPT semester didn’t know how to use it well. The ones I teach now do. That’s the difference teaching makes.

Thanks to Irina, Michelle, Sarah, and Lisa for contributing their thoughts.

They responded to this question:

What is the good and the bad about using artificial intelligence with English learners?

Consider contributing a question to be answered in a future post. You can send one to me at lferlazzo@epe.org. When you send it in, let me know if I can use your real name if it’s selected or if you’d prefer remaining anonymous and have a pseudonym in mind.

You can also contact me on X at @Larryferlazzo or on Bluesky at @larryferlazzo.bsky.social

Just a reminder; you can subscribe and receive updates from this blog via email. And if you missed any of the highlights from the first 13 years of this blog, you can see a categorized list here.