Academic Integrity and Assignment Design

Should we change the way we assign writing this fall given what we know about current AI text generation capabilities? The task force is cautioning educators about the use of unreliable and biased AI text detection software (see our Working Paper: Overview of the Issues, Statement of Principles, and Recommendations). How else can we discourage unacknowledged use of AI and encourage academic integrity? Many educators have expressed an urgent desire for immediate guidance from the MLA/CCCC Task Force on Writing and AI on these matters. Below are some sample questions received during our July 26th webinar.

  • “I’d welcome suggestions for specific writing assignments beginning this Fall. I cannot overestimate how much I would like specific, practical examples. How would folks now consider their approach to the standard 3-5 pp. paper, for instance?” 
  • “I want to see some practical application of how to approach this in my classes that start in a few weeks.”
  • “How do the speakers respond to the idea of having students write in class, pencil to paper? Why would they discourage instructors from adopting practices different from those used in different classrooms?”
  • “I find in our Writing Across the Curriculum (WAC) program that we (writing studies folks) are increasingly becoming the “go to” AI folks on campus. I’m wondering if the group here has any thoughts about how to talk with our colleagues across disciplines about AI, when they might not understand and view writing the way we do in our collective fields? Who might be prone to simply “cut” writing if it’s to be a problem?”
  • “It appears that LLMs can generate a thesis, a proposal, an outline, a draft, and a revision in short order. Does it seem possible to use that to complement our attempts to teach writing as a process? Or is it more likely to to make the idea of the writing process irrelevant?”
  • “Much of the discussion about process-based instruction and a “renaissance” of writing instruction ignores the many (and valid) ways in which writing is used for assessment in classes where writing instruction is not the (or a main) focus of the class.”

Overall approach

Our short answer is that we’re not advocating discarding or shortening essay assignments or only working on them only in class or other proctored environments. Rather, we’re looking at adapting our existing assignments to mitigate dishonesty and learning loss by discouraging misuse as much as possible through recognized best practices in assigning writing.  

We follow scholarship on academic integrity which suggests that in making it easier to do the right thing and harder to do the wrong thing can significantly reduce dishonesty (See Research Agenda for Academic Integrity, edited by Tracey Bretag, and Defending Assessment Security in a Digital World by Phill Dawson).  We have never been able to prevent all cheating, and that will not change. However now that generative AI makes cheating easier and cheaper than ever, it does seem important to adjust assignments.

Certainly, it helps to consider any possible changes to our assignments in the context of our overall policies. Making policies about acceptable AI use clear, specific, and grounded in learning goals is key.  Policies can be formed, discussed, and adapted in collaboration with students to increase trust. We will discuss the question of forming policies in a separate post. In the meantime, see the Quick Start Guide to AI and Writing section on policies.

When it comes to assignment design, we advocate looking for overlap between practices that support learning and those that deter misuse of AI. Often, well-established best practices in writing instruction make it less desirable and more difficult to pass off AI text as student-written.  Such practices include the following.

Design assignments to support intrinsic motivation

It’s always worth underscoring that clarifying the purpose of each assignment and introducing student choice or personal connection to the topic can help make dishonesty less likely. We likely try to do this anyway, but there are always ways to improve. So many resources on writing pedagogy provide guidance here–one that comes to mind is the The Meaningful Writing Project.

Emphasize teacher, peer, and tutor relationships in the writing process 

Situating the writing process in the context of a genuine teacher-student relationship and peer-to-peer relationships makes writing assignments more meaningful. Conferencing with students about their writing and ideas, either informally in class or more formally in scheduled meetings may serve both to support student engagement and accountability. 

Assign steps in the writing process 

Here we mean annotation of readings, brainstorming, outlining, drafting, peer review, revision, and reflection on the process. While AI systems could be used to produce artifacts simulating artifacts of a writing process and metacognitive reflection, assigning them supports and encourages student engagement and makes dishonesty more cumbersome. 

Ask for documentation of and reflection on the writing process

Asking students to share and comment on the version history for their draft can help make it transparent when text was copied and pasted from another source. We are exploring supplementing Google Docs version history with apps such as Draftback that provide more information about the history of the document. Ideally, sharing version history could support students’ own awareness of their process, especially if coupled with assigned metacognitive reflection. Of course, a student could retype language model text and language-model-suggested editing, but this would be cumbersome. If we do allow some use of AI, we might ask students to document such use by linking to and commenting on the chat sessions involved and/or by writing a reflection or filling out a questionnaire on their use of AI.

Test our assignments on language models

We do recommend that teachers test writing assignments using the more sophisticated systems available like ChatGPT running the GPT-4 or, for a free alternative, Bing AI, which also builds on GPT-4. Claude 2 (available on is also similarly sophisticated and can be used for free. We should supply any needed information to the system such as reading materials, assignment directions, rubrics, sample essays, or anything else available to students. For a good resource on how to do this, see Stress Testing Writing Assignments: Evaluating the Exposure of an Assignment’s Tasks to AI by Annette Vee and Tim Laquintano, University of Pittsburgh Writing Institute Workshop on AI and the Teaching of Writing, June 1, 2023. 

The purpose of such testing would be to give instructors familiarity with the quality of outputs that might be expected. Note that it would not necessarily enable us to detect when or if a student has used such a system since the systems can be prompted to generate output in various styles and can give different outputs each time. We should be cautious about trying to recognize AI text intuitively as our judgments may be influenced by unconscious bias as well as inaccurate assumptions about AI output. 

Approaches we are skeptical about

Trying to create assignments AI can’t do

Some instructors have attempted to create assignments they don’t think AI systems can plausibly complete. However, we are not aware of evidence that this is effective. Students may try a variety of prompting strategies to elicit at least C or B level work on a wide range of assignments. We know that language models can mimic types of writing that seem very human, such as metacognitive reflection and personal narratives (these kinds of assignments might, however, be excellent for other reasons such as encouraging intrinsic motivation). Recent updates allow some systems to describe and analyze images (Bing), synthesize information from multiple documents (Claude), and search the internet (Bard, Bing, PerplexityAI). Video transcripts can be fed to AI systems as well. In most cases class discussion is not recorded, so some teachers ask students to incorporate points from class discussion into their essays. Still, given class discussion notes, an AI system can generate an essay that refers to specific points from discussion.

Assigning only in-class writing and/or writing by hand

Some instructors are moving toward assigning writing only by hand and/or only in class. We are concerned that such a shift would remove many opportunities for sustained reflection and revision and development of more complex arguments. We also are aware of concerns about the psychological pressures of in-class writing and the physical barriers for many students with disabilities. There are also concerns about how equitably different handwriting styles will be graded. However, as noted, we do see some opportunity for in-class writing process assignments, especially formative assignments. These might serve as a complement to out-of-class writing and a way to get to know student voice.

Long-term strategies: shaping policy and user interfaces to support writing for learning

Even as we explore pragmatic responses to current AI capabilities, we need not see ourselves as passive and forced only to react to these technological changes. We can also consider how we might participate in civic conversations that will shape the digital writing environment going forward. Is it a societal goal to facilitate human communication and learning by promoting labeling of AI text as such, in keeping with academic values around source citation? Where does policy that supports academic integrity intersect with policy that supports other societal interests like preventing dissemination of disinformation? The task force has made a public comment to the Office of Science and Technology Policy. We suggestions for our future engagement with policy and hope other educators and organizations will take this on as well.