From AI-Supported ESL Writing to Co-Researchers: Empowering Refugee Youth Through Youth Participatory Action Research (YPAR)

By Muhammad Ahmad, University at Buffalo

DOI: https://www.doi.org/10.69732/KZKP4254

She had barely spoken during our first session together. When I introduced the project, she sat quietly, nodding along, but saying little. I didn’t push. I had worked long enough with refugee youth to know that silence often has a story behind it.

Eight weeks later, that same student was standing at the front of a university conference room, showing slides, walking a group of professors through her research findings. Her voice was steady. Her argument was clear. When she finished, she looked out at the room, not to check for approval, but with the quiet confidence of someone who had just discovered something important about herself.

I had started this project with a simple goal: help refugee youth improve their writing using AI tools, and empower them in the process. What unfolded was something much bigger.

Why This? Why Now?

Refugee youth in ESL classrooms carry a particular kind of weight that is easy to miss from the outside. They are learning a new language while simultaneously navigating displacement, cultural adjustment, and the pressure to keep up academically in a system that was rarely designed with them in mind. I have spent over fifteen years teaching English as a foreign language across the Middle East and South Asia, and most recently in the United States, where I work as an academic coach supporting refugee and immigrant students. In my work as an academic coach, I saw this every day: students who were bright, motivated, and full of things to say, but who struggled to get any of it onto the page.

Writing is especially challenging for this population, and not only because of grammar or vocabulary. Writing demands a sense of voice, a belief that what you have to say is worth saying. For many of my students, that belief had been quietly worn down long before they arrived in my classroom.

Traditional writing instruction, even when well-intentioned, often falls short. Class sizes are large, teacher time is limited, and meaningful feedback cycles are rare. This is where AI tools like ChatGPT and Gemini opened a door I hadn’t anticipated, not as a replacement for the teacher, but as a patient, always-available tool that could respond to student writing at any moment. The question driving this project was simple: could these tools, used intentionally, not only improve their writing, but also change how they saw themselves as learners?

Two Phases, One Goal

I worked with ten senior high school students on a two-month pilot project that was developed as part of the second stage of “The Impossible Project,” an initiative funded by Mozilla and led by another instructor within my department. The purpose of my part of the project was two-fold: to improve their writing skills, and to empower them as learners. The project followed the principles of YPAR (Youth Participatory Action Research), a framework that positions young people not just as subjects of research, but as its architects. I was not their classroom teacher; I came to this as an independent researcher, but one they already knew from my work at their school. We met as a group every week, and what happened across those eight sessions surprised me in the best way.

Phase 1: Students as Writers. After brainstorming as a group about what kind of research we wanted to do, the students agreed to investigate the role of AI in improving second language learners’ writing. I encouraged them to draft short texts independently and mess around with ChatGPT and Gemini: paste a paragraph in, see what comes back, and bring it to the group.

The students quickly noticed that the feedback they were getting was too generic and didn’t address what they wanted specifically. In our group discussions, they shared this frustration openly. Later, we came up with a plan to work on specific prompts that would help them get more tailored feedback from the AI tools. They specified their context by explicitly stating they were high school ESL learners and instructing the AI not to use overly advanced vocabulary. We then co-developed targeted prompts focusing on specific linguistic elements during our group discussions, such as checking verb tenses, grammar, and punctuation. By the end of Phase 1, students had concluded that using AI with specific prompts helped them improve certain aspects of their writing: grammar, vocabulary choice, and authorial voice. Seeing the students interact, support, and push each other during these sessions made me so happy. That energy was a key factor in moving onto the second phase of the project, the empowerment phase!

Phase 2: Students as Co-Researchers. In this phase, I wanted to elevate things and make the students co-researchers and experience what it actually feels like to conduct research. I navigated them through the basic stages: selecting a topic (which they had already played a huge part in), reading about the problem, finding participants, collecting data, analyzing it, and presenting findings. The students were super excited to try the whole process.

They started looking for participants in their communities (families, peers, school friends), all second language learners from different ages. They introduced their research, asked participants to write short paragraphs or sentences, ran those writings through the AI tools, and analyzed the feedback. Some worked on the analysis on their own; others brought it back to our group meetings. In all, the students enjoyed this entire process so much. They felt they were doing something really connected to their people and friends. They felt they were part of something huge. For me, to make them taste the fruit of their great work, I arranged for them to present their findings at the Center for Information Sciences Conference at the University at Buffalo (UB). Before the conference, the students worked together to prepare their presentation, dividing it into sections, deciding who would cover what, and rehearsing as a group. On the day, every single one of them stood up and presented. That was such a huge step in their journey.

In Their Own Words

After conducting individual interviews with each student at the end of the project, I was amazed by how rich and valuable their insights were. What came through most clearly was not just that they had learned; it was how they had learned to think.

In Phase 1, students quickly became critical users of AI, not just passive ones. They noticed when the feedback was too generic, and they pushed back. One of the most powerful things I observed was how they developed a clear sense of ownership over their own writing, including what the AI was and wasn’t allowed to change. One student described it this way:

“I accepted it when it changed the vocabulary, it gave me better words, professional words, I would say. But when it completely changed everything I wrote and gave me different things from what I wanted, I rejected that.” –a student, post-project interview

That sense of authorship, knowing when to accept AI suggestions and when to push back, became a real theme across the group. Another student put it plainly:

“When we write an essay, we always want to have our tone. Everyone has like a different tone of writing. We just want to keep that. If ChatGPT gives you a dramatic difference, you should not accept it.” –a student, post-project interview

In Phase 2, the students took everything they had learned and brought it to their communities. They recruited participants, collected writing samples, ran them through AI tools, and analyzed the feedback. Presenting their findings at UB was the moment it all came together. One student’s words after the presentation still stay with me:

“I feel proud because we all did a very big presentation in front of all professors in college which showed that we can face anything as English is not our first language.” –a student, after the UB presentation

What I Found

After finishing the whole project, I conducted individual interviews with each student to reflect on both phases. Their insights were so amazing and valuable. Across the two phases, I observed consistent growth in three areas of writing: grammar, vocabulary choice, and (most importantly) authorial voice. These gains were observed by qualitatively comparing the students’ initial drafts written before using the AI with their final pieces after the AI intervention, looking specifically for shifts in sentence structure, word choice, and individual expression. In addition to that, during the post-project interviews, students explicitly reflected on these changes, explaining how their prompting choices helped them retain their unique perspective. Most importantly, they reported actively rejecting AI suggestions that altered their original meaning or tone, thereby confirming the observed growth in their individual authorial voice. These gains were most clear among students who had learned to direct the AI rather than just accept whatever it gave back.

But the more significant finding was harder to measure. By the end of the project, every student I interviewed described themselves, in one way or another, as a researcher. That shift in identity matters enormously. In my experience as an academic coach, students who see themselves as capable inquirers approach their writing differently, with more investment, more confidence, and more willingness to revise. One student summed up what that felt like from the inside:

“I mean we did everything that researchers will do, you know. We collected data, we asked questions, and we were trying to find answers to those questions. And we share our results with a group of people, you know.” –a student, post-project interview

The confidence I observed at the UB presentation, in posture, in voice, in how students answered questions, was not something I had explicitly taught. It had grown out of the process itself: the collaboration, the ownership, and the experience of doing real work that mattered to real people in their own communities.

What This Means for Your Classroom

What does this project mean for a teacher sitting in a classroom right now? All of our students have these challenges to one extent or another, even if they are not refugees, so these ideas can benefit everyone. Here are three concrete starting points.

First, start small with AI feedback. You don’t need a full research project to try this. Ask students to write a short paragraph on any topic, then have them paste it into ChatGPT or Gemini and ask: “What can I improve in this paragraph?” Let them bring the feedback back to class and discuss it together. That single step alone opens a rich conversation about what good writing looks like, and who gets to define it.

Second, teach prompt engineering as a language skill. One of the most surprising things I noticed was how much the quality of the AI feedback depended on how students asked for it. Helping students write better prompts is, in itself, a writing and critical thinking exercise. Try asking: “How can you tell the AI to focus only on your word choices?” or “What would you change in your prompt to get more specific feedback?”

Third, look for the empowerment angle. Writing improves when students feel they have something worth saying. Find small ways to make their work visible beyond the classroom. Confidence and writing skill grow together, and this project showed me just how quickly that growth can happen.

Being Honest

This project was a pilot, and like any first attempt, it came with real limitations. The group was small (ten students), and what worked in this specific context, with these specific students, may look very different in another classroom or community.

The two-month timeframe was also tight. There were moments when I felt the students needed more time to sit with a phase before moving forward, especially the transition from writers to co-researchers, which required a significant shift in how they thought about their own role and their right to ask questions.

I also want to be honest about the AI tools themselves. They are not neutral. The feedback they generate reflects certain assumptions about what “good writing” looks like, assumptions rooted in standard academic English. That tension is real and worth naming. It’s something I plan to examine more critically in future work.

In addition, anyone using AI with students needs to keep privacy in mind. During the onboarding and brainstorming phase of my project, I explicitly instructed the students to avoid sharing any personal data or private details with the AI tools. As a result, the students’ writing focused on general, non-sensitive topics such as “What does success mean to you?” or describing a time they tried something new.

I keep coming back to that moment in the university conference room. A student who had barely spoken in our first session, standing, presenting, owning the room. It wasn’t just her writing that had changed. Something in how she carried herself had shifted.

That is what this work is really about. When we give refugee youth the tools, the space, and the genuine belief that their questions are worth investigating, they don’t just become better writers; they begin to see themselves differently. And in my experience, that shift in identity is where real and lasting language learning begins.

 

AI disclosure: Minimal use of AI: Generative artificial intelligence was used in the preparation of this article only for brainstorming ideas or for spelling and grammar suggestions.