ChatGPT in College Assignments: Study Exposes Missing Rules in Library Education

The research examined undergraduate course documents at Sultan Qaboos University in Oman, exploring how clearly they addressed AI-supported assessment, responsible use, and the skills students need to judge information. The findings reveal a gap between AI’s potential in education and the written instructions students receive, with the weakest guidance covering some of the responsibilities most important to future librarians.

A student opens ChatGPT to brainstorm an assignment, tidy up a paragraph, or make sense of a difficult topic. The course handbook says little about AI, leaving the student to guess where useful support ends and academic misconduct begins.

That uncertainty sits at the centre of Noura Al Hosni's study, "ChatGPT as an AI-based assessment method in library information science education: investigative study in the information studies department," published in Frontiers in Education. The research examined undergraduate course documents at Sultan Qaboos University in Oman, exploring how clearly they addressed AI-supported assessment, responsible use, and the skills students need to judge information.

The findings reveal a gap between AI's potential in education and the written instructions students receive, with the weakest guidance covering some of the responsibilities most important to future librarians.

What 30 Course Outlines Revealed

Al Hosni reviewed 30 undergraduate course outlines from the Information Studies Department for Fall 2024 and Spring 2025. These represented 62.5% of the department's 48 undergraduate courses; exclusions included courses with non-responsive instructors, practical training modules, and graduation projects.

A ten-part checklist, developed from 16 selected research publications and reviewed by two experts, examined AI policies, assessment design, academic honesty, disclosure, information checking, transparency, feedback reliability, ethical safeguards, and connections to learning outcomes.

The study treated AI-based assessment broadly, covering rules about student use, assignments that teach AI literacy and ethics, tasks requiring documented AI assistance, and instructor use for feedback or evaluation.

Each course received an overall classification based on its most frequent checklist rating, with tie-breaking rules giving priority to verification and disclosure:

• 20 courses, or 67%, fell into the category covering absent, unclear, or prohibited AI use.

• Six courses, or 20%, recognised limited use, such as brainstorming, drafting, searching, or language editing.

• Four courses, or 13%, provided clearer permission or more developed integration into coursework and assessment.

The strongest integration appeared mainly in courses with practical components, where searching databases, organising information, handling data, and developing projects offer opportunities to use AI.

The Biggest Gaps Concern Trust and Responsibility

Checking AI-generated information was the least developed area: 83.3% of course outlines received the lowest rating for verification requirements. Students rarely received detailed instructions to check factual claims, examine bias, or confirm that suggested references existed.

For a discipline built around information credibility, that omission carries particular weight. A convincing answer with an invented citation can look useful until someone checks the source, and learning to recognise that problem belongs at the heart of professional preparation.

Ethical safeguards were similarly weak: 80% of outlines received the lowest rating for fairness, bias, privacy, and accessibility, leaving limited guidance about sensitive information, unequal access to paid tools, or alternatives for students unable to use them.

AI-related academic integrity guidance received the lowest rating in 76.7% of outlines. Disclosure requirements and checks on AI-generated feedback or scoring each received that rating in 73.3%. Comprehensive guidance on disclosure, verification, feedback reliability, and ethical safeguards appeared in only 3.3% of courses for each area.

The gaps extended to basic communication: 70% received the lowest rating for explicitly naming AI tools, 63.3% for AI-use policies and assessment integration, and 60% for transparency about AI's role. Connections to learning outcomes were comparatively stronger, with 40% showing basic alignment and 6.7% showing robust alignment. Even here, more than half received the lowest rating, suggesting that recognising AI's usefulness often stopped short of explaining how it would support or be assessed against specific skills.

Where AI Could Help Learning and Where It Could Undermine It

The paper's review of earlier research describes several potential benefits, including faster feedback, more flexible learning support, redesigned assignments, and reduced time spent on repetitive assessment tasks. These possibilities were not tested through a classroom experiment in this study.

In library and information science, AI could support search strategies, summarisation, metadata creation, digital content organisation, and research assistance. Language and revision support could make learning more accessible, and analysis of student performance could help instructors identify people needing additional guidance.

Assignments that ask students to compare AI responses with scholarly sources, identify errors, and explain their corrections could develop critical thinking alongside practical AI literacy. Those skills connect directly to work in libraries adopting automated classification, smarter search systems, and other AI-supported services.

The risks come from unreliable outputs and poorly designed assessment. Automated feedback can miss context, complex reasoning, creativity, or disciplinary knowledge; excessive dependence can reduce opportunities for students to practise the skills an assignment is meant to develop.

Unacknowledged AI-generated work creates authorship and integrity problems. Unexplained grading can damage trust, paid tools can create unequal opportunities, and uploading student work or confidential information can expose sensitive data. Keeping AI education concentrated in practical courses can leave ethical and evaluative gaps elsewhere in the curriculum.

Clear Rules Need to Reach Every Assignment

Al Hosni recommends a departmental policy and syllabus statements defining permitted, restricted, and prohibited AI use, with examples separating assistance from substitution. Practical courses would introduce AI activities tied to learning outcomes, using rubrics covering prompt quality, originality, accuracy, source evaluation, ethics, disclosure, and reflection.

Disclosure templates would record tools, versions, dates, prompts, contributions, verification steps, and student responsibility. Logs at outline, draft, and final stages would show how assignments developed. Students would earn marks for checking at least one AI-assisted claim, citation, or source against credible material and documenting corrections.

Training for teachers and students, confidentiality safeguards, bias awareness, and alternatives to paid tools would support responsible participation. AI-supported grading would require human review before release and a clear appeals process. Instructor interviews and student surveys would explore classroom practice and learning outcomes, helping refine guidance that teaches students to question AI and remain accountable for what they submit.