Higher order writing tasks can support students’ abilities to engage critically with subject matter, requiring not just a grasp of existing knowledge but an ability to evaluate, synthesize, and generate new insights. These tasks are foundational in fostering critical thinking and analytical skills, which are pivotal in academic and professional settings. In our study, the data indicate a significant dependence on AI software among students, especially in the realms of “Finding Evidence or Examples,” “Finding Information,” and “Understanding Complex Topics.” This reliance on AI for sourcing and digesting complex material aligns with recent findings by Wambsganss et al.5, who found that university students particularly value AI tools that support autonomous learning and complex task completion. However, while Wambsganss et al. focused on a structured writing tutor system, our findings suggest students apply similar strategies even in open-ended contexts.
Our findings both complement and extend previous research on AI-assisted learning in higher education. For example, Banihashem et al.6 found that students effectively used AI-generated feedback to improve their essays, but primarily focused on technical aspects of writing. In contrast, our study reveals that when given freedom to use AI as they choose, many students leverage it for deeper conceptual understanding and evidence gathering. This difference suggests that structured AI systems might be unnecessarily constraining student engagement with AI tools. Similarly, while Escalante et al.8 documented student preference for AI feedback in specific writing tasks, our findings show students actively discriminating between tasks they consider appropriate for AI assistance versus those requiring independent thought.
The utilization of AI for “Independent Research” suggests a proactive approach by students, using AI to dig deeper into subjects and find relevant scholarly articles, datasets, and other resources that are beyond the superficial layers of information accessible through conventional search engines. This aspect of AI use may highlight a shift towards more autonomous learning, where students harness technology to drive their inquiry and extend their research boundaries.
Similarly, the frequency with which students reported using AI for “Finding Information” reflects the pivotal role of AI in sifting through information, verifying facts, and identifying the most relevant data. This process is crucial in building a solid foundation for any higher order writing task, where substantiated arguments and evidence-based claims form the backbone of quality academic writing.
Furthermore, the recurrent use of AI for “Finding Evidence or Examples” and “Understanding Complex Topics” underscores its significance in enhancing students’ research capabilities. “Finding Evidence or Examples” is vital for supporting arguments, illustrating points, and grounding discussions in concrete instances or findings, thereby enriching the academic discourse. “Understanding Complex Topics”, on the other hand, involves breaking down content into comprehensible segments, a process where AI can provide explanations, simplifications, or different perspectives that aid in grasping challenging concepts.
Collectively, these findings demonstrate the integral role that AI can play in facilitating higher order writing tasks by enabling access to, and interpretation of, complex information. Students are not merely passive recipients of AI-generated content; instead, they are engaging actively with AI-based tools to augment their research processes, enhance comprehension, and construct well-informed, analytically-robust academic texts. This blending of student effort and the capabilities of AI for tackling higher order writing tasks is emblematic of a shifting educational paradigm, where AI can act as a “more expert other” that allows students to work within the zone of proximal development23. Specifically, students use the AI to help them build on what they already know, ideally engaging in critical thinking and deep learning in the process, as they evaluate the AI-generated content against what they have learned in class and decide how best to integrate such content with their pre-existing knowledge in ways that meet the requirements of their assignments.
Lower order writing tasks, though often perceived as merely the final polishing steps in the writing process, are crucial for conveying ideas clearly and professionally. These tasks encompass the mechanics of writing, including grammar, punctuation, and spelling, which are fundamental for enhancing the readability and credibility of academic writing. Our analysis sheds light on the potential that AI offers for supporting students in these essential components of the writing process, with “Revising and Editing” standing out as the most prevalent code in this category. This finding aligns with Gombert et al.‘s19 research showing students’ strong preference for AI assistance with technical writing aspects. However, while Gombert et al.‘s study focused on structured essay assessment, our findings reveal students independently choosing to use AI primarily for these lower-order tasks even when given complete freedom in their AI use.
The prevalence of AI use for editing and revision in our study both confirms and extends previous research. For example, Banihashem et al.6 found that students valued AI-generated feedback for technical improvements, similar to our findings. However, our results suggest a more nuanced approach—students actively distinguished between using AI for mechanical improvements versus maintaining their original voice and ideas. This selective approach to AI assistance wasn’t observed in previous studies of structured AI writing systems, suggesting that when given agency, students develop more sophisticated strategies for integrating AI into their writing process.
The act of “Revising” involves more than just making minor edits; it requires students to review the text carefully to improve its flow, clarity, and overall structure. Students’ use of AI for revision suggests a strategic approach to refining their arguments, reorganizing information for better coherence, and ensuring that their writing meets the expected academic standards. AI tools can offer suggestions for restructuring sentences, enhancing word choice, and identifying logical inconsistencies, thereby providing support for students aiming to improve their compositions.
“Proofreading,” on the other hand, focuses on correcting surface errors in writing, such as typos, misspellings, and grammatical mistakes. The frequent use of AI for proofreading points to its effectiveness in catching errors that might elude the human eye, ensuring that documents are polished and error-free. This is an important task, as even minor mistakes can distract readers and detract from the writer’s credibility. By automating the laborious process of proofreading, AI allows students to focus more on the content and substance of their work.
These findings illustrate the potential role of AI in students’ writing process. AI does not replace human judgment, but rather complements students’ efforts in crafting well-written, technically sound documents. By using AI for lower order writing tasks, students are able to produce texts that are not only conceptually strong but also flawless in their execution. This dual focus on content and form is essential for achieving excellence in academic writing.
We also note that, for students to benefit the most from using AI for revising and proofreading, they should have guidelines for the sort of prompts to use. Specifically, these should be prompts that generate feedback on the grammaticality of students’ writing rather than rewriting students’ work. Our analysis demonstrated that some students used these types of prompts, such as the student who shifted their prompt to elicit suggestions for slight revisions of their words. This sort of feedback provides students with an opportunity to see grammatical trouble sources and actively engage in correcting them.
Our analyses revealed complex patterns in how students interact with AI technologies in their academic endeavors. The concept of “Efficiency” emerged prominently in our codes, aligning with Strzelecki’s20 findings on student AI adoption patterns. However, while Strzelecki found general acceptance of AI tools, our study revealed a more nuanced approach where students strategically used AI to streamline specific tasks while maintaining their intellectual independence. This selective efficiency mirrors Thi Thuy’s22 findings about students’ careful navigation of AI benefits and limitations, though our results suggest even more sophisticated decision-making about when and how to leverage AI support.
The students in our study went beyond simple time-saving uses, demonstrating what Wambsganss et al.5 described as “adaptive engagement” with AI tools. By using AI to handle mechanical tasks like proofreading and quote integration, students reported being able to focus more on critical thinking and complex analysis. This finding extends previous research by showing how some students, when given agency, naturally develop workflows that preserve their intellectual engagement while leveraging AI for efficiency. Unlike studies of structured AI systems6,8, our results suggest that students can effectively self-regulate their AI use to maintain both productivity and learning quality.
Moreover, the data reveal a substantial emphasis on student independence, indicating a perception of AI not merely as a tool for task completion but as an enabler for students’ own creative and independent thinking and learning. The use of AI to help students generate “Independent Ideas” points to its capacity to provoke thought, offer unique perspectives, or suggest novel solutions that might not have been immediately apparent to the student. This can be particularly valuable in brainstorming sessions, conceptual development phases, or when seeking to explore alternative approaches to problem-solving. Similarly, AI can assist in overcoming writer’s block by suggesting sentence starters or offering different ways to phrase ideas, thereby aiding in the translation of complex thoughts into coherent written form. This aspect of AI use has the potential to foster a sense of autonomy in students by offering them an ever-present “sounding board” and “peer” reviewer for their ideas.
Despite the many ways that students relied on AI for their projects, our data indicate a sophisticated skepticism about these systems. This finding both aligns with and extends previous research on student AI perceptions. While Farhi et al.21 documented general student concerns about AI ethics and reliability, our study reveals more specific and nuanced forms of skepticism. Students in our study demonstrated what Strzelecki20 calls “critical technology acceptance,” actively evaluating AI outputs against their course knowledge and making informed decisions about what to include or exclude. This approach extends beyond the basic digital literacy described by Lamb et al.26 to encompass sophisticated evaluation of AI-generated content.
Particularly noteworthy was students’ careful maintenance of their intellectual independence, a finding that adds new dimensions to existing research on AI adoption in higher education. Unlike studies showing either broad acceptance22 or resistance2 to AI tools, our results suggest students can develop balanced approaches that leverage AI’s benefits while preserving their academic autonomy. This selective engagement may reflect the explicit permission given for AI use in our study, contrasting with contexts where AI use is either mandated or restricted.
The concerns expressed by students indicate that educators and academic institutions need to better address the role of AI technology in educational settings. Teachers should engage students in conversations about AI, providing resources that enable them to critically evaluate these technologies and their applications. By exploring the capabilities and limitations of AI systems, students can learn to make well-informed decisions about how to effectively and ethically incorporate them into their academic work. In addressing students’ concerns head-on, the academic community can work towards integrating AI in a manner that both enhances the learning experience and maintains academic integrity.
While our study suggests that AI may provide effective support for students’ learning in higher education contexts, there are some limitations to consider. First, our data consists of students’ self-reported use of AI in submissions that were not anonymous. Students were asked to document their AI usage as part of their identified final project submissions, which may have influenced how they characterized their interactions with these tools. Social desirability bias could have led students to underreport their use of AI or to emphasize certain types of AI usage (like proofreading and editing) that they perceived as more academically acceptable while downplaying other uses. The lack of anonymity may have amplified this effect, as students might have been more cautious about documenting AI use when their names were attached to their submissions. Some students may have opted not to report their AI use at all, fearing potential academic consequences despite explicit permission to use AI for the assignment. Future research might benefit from alternative methods of documenting AI use, such as automated tracking of AI interactions, anonymous reporting mechanisms, or structured interviews that explicitly address students’ comfort in discussing their AI use.
Second, our study captures a specific moment in time – early 2023, just months after ChatGPT’s release. Student approaches to AI use likely evolved as they gained more familiarity with these tools and as the tools themselves improved. The rapid pace of AI development means that patterns we observed may not reflect current student practices.
Third, our findings come from a single course at one university, focused on sustainability and technology. The nature of the final project – analyzing conceptual networks – may have led to different AI usage patterns than might emerge in other types of courses or assignments. Students in this General Education course also came from diverse academic backgrounds, which could have influenced their comfort with and approaches to using AI tools.
Fourth, our sample size of 39 students who documented their AI use represents only a portion of the course enrollment. While we know these students came from various majors, we lack demographic information that could help us understand how factors such as age, year in school, or prior technology experience might influence AI usage patterns.
Finally, this study’s original purpose was pedagogical rather than research-oriented. Had we designed it primarily as a research study, we might have implemented different data collection methods, such as structured interviews or surveys, to gather more detailed information about students’ decision-making processes regarding AI use.
Future research might address these limitations through larger-scale studies across multiple institutions and course types, longitudinal analyses of how AI use evolves over time, and mixed-methods approaches that combine self-reporting with other data collection strategies.
