Last Updated: Summer 2024
Generative AI (GenAI or GAI) tools such as Gemini and ChatGPT are already widely available, and the technology is improving rapidly. With that in mind, all faculty need to consider the ways that students’ access to GenAI might affect the way they complete and learn from the writing assignments in your course.
The exercises below will help you think through the ways that students might use Generative AI—appropriately or inappropriately—in your class and how you might adjust your assignments or teaching accordingly.
Step 1: Consider Your Assignment
To evaluate how GenAI might undermine or mesh with the goals of your assignment, you first need to fully understand what your goals are. If you haven’t already, think through the questions below and write out some quick answers below each.
What skills is the assignment meant to develop or teach?
In other words, what do you want students to learn or practice through their writing for this assignment?
What aspects of the assignment are meant to challenge students?
In other words, what aspects of the assignment should be difficult for students to do well?
(How) will students use these skills later?
It’s certainly acceptable to create a writing assignment that serves the goals of your course alone, but if there are specific skills embedded in the assignment that later instructors will expect your students to learn in your course, it’s important to consider that.
Step 2: Try to Complete the Assignment Using AI
Capabilities of GenAI
With the goals of your assignment firmly in mind, consider the ways that students might use GenAI to assist them. It’s important to keep in mind that AI tools can do a great deal more than write entire essays. They can also:
- Offer topic and thesis suggestions
- Break down the assignment prompt into a list of discrete tasks
- Write outlines (either from scratch or based on ideas fed to it by the user)
- Suggest sources or specific quotes (though some of these may be invented by the AI itself)
- Create bibliographies
- Offer suggestions for revision
- Revise existing text based on prompts from the user (again, this could be text written by the user or by the AI itself)
- Create charts and tables
This is not to say that all GenAI tools do all of these tasks particularly well, but they all can and will do them.
And, of course, a GenAI tool can also write a whole document if the user asks it to.
Limitations of GenAI
Fortunately for us as college faculty, most GenAI tools can’t do the above tasks particularly well. Thus, any student who attempts to use AI for the above tasks will quickly run into some well-documented limitations:
- GenAI tools are prone to “hallucinate,” the industry term for making up results and presenting them as true (i.e. lying). This could include fabricating facts, sources, quotes, or data.
- As of this writing, GenAI has difficulty writing anything longer than about 1000 words. Longer texts tend to ramble or even become entirely incoherent.
- GenAI cannot process or respond effectively to specific details and quotes. AI-written arguments tend to generalize and make sweeping statements but are rarely able to support those statements with details and analysis in the way academic arguments require.
- Since the AI’s model for academic writing is drawn from the internet at large, its writing tends to be rather formulaic. For example, most GenAI tools are extremely attached to the 5-paragraph essay formula, and they tend to assume that all thesis statements must be one sentence long. The major GenAI tools are also bizarrely fond of section breaks for some reason, even in very short essays.
- Getting the best output from GenAI tools requires the ability to write effective prompts and a willingness to experiment and iterate with the results until they come as close as possible to what you want. While this isn’t a particularly difficult skill to acquire, it does require a certain amount of time, intellectual effort, and patience to both acquire and use. This means that students looking to use GenAI to quickly do their homework for them will generally produce underwhelming results.
Shaking Down Your Assignment
With these capabilities in mind, try logging into a GenAI platform (ChatGPT, Gemini, Claude, MS Copilot, etc.) and putting yourself in the role of a student. Think through each step in the writing process and ask yourself how you might get the AI to help you complete it.
As you do this, it’s best not to assume that only cheaters use AI. Rather, try to imagine yourself as an otherwise honest student who sees the AI platform as just another study aid.
Exactly how you approach this step will depend on your assignment, but here are some suggestions:
General Tips:
- Always set the context when you prompt an AI for writing or assistance. So, for example, try telling the AI “I’m a college sophomore in a 300-level Biology class and I’m writing a lab report in response to the following assignment.” Then cut and paste the prompt into the tool.
- Use follow-up prompts to revise or refine the AI’s output. So, for example, you might say “Revise thesis #4 to address the wider implications of the argument” or “The tone of that draft is too informal. Rewrite it to use more academic language.” This doesn’t necessarily improve things (see the limitations noted above), but as a general rule you’ll get better results by tweaking and iterating the AI’s output a few times.
- If the AI balks at an instruction, try to get around its reluctance. For example, Gemini might say that it can’t complete a lab report assignment because it doesn’t have the necessary data. In that case, you can either give it sample data (“Complete the assignment using these results…”) or ask it to fabricate data (“Complete the assignment, fabricating appropriate results as needed.”). This obviously isn’t an ethical use of the tool in a classroom setting, but it will help you understand its capabilities.
Prewriting with AI:
- Give the AI the prompt and context for your assignment (see above). Then, ask it to write ten (or more) sample thesis statements or topic ideas for that assignment.
- Read over the AI’s suggestions. If you see a general pattern or limitation, ask the AI to go outside that pattern. For example: “Write ten more theses that focus on characters other than Romeo or Juliet,” or “Write ten more theses, but this time address how these results confirm or contradict previous experiments.”
- Choose one of the AI’s theses or topic statements and ask it to write an outline for the completed assignment. Again, you might try to refine your results a bit by asking it to add specific points, citations to specific sources, or to come up with counterarguments.
- If the assignment involves research, ask the AI to suggest ten (or more) sources you might use for this assignment. If it doesn’t do so automatically, ask it to give you bibliographic citations in a specific style (e.g. MLA or APA) and links to access the sources online. If the sources it suggests are inappropriate for the assignment, ask it to try again. For example: “These sources were written for a general audience. Suggest ten scholarly sources I could use for this assignment.”
Composing with AI:
- Again, give the AI a context and prompt and then ask it to write a complete response to the assignment. You can also ask it to turn a topic statement or outline from the previous step into a completed assignment. (“Write an essay based on thesis #4.”)
- Get the AI to revise its first draft. You can ask it to address global issues (“Revise the previous draft, adding more detailed analysis throughout.”), improve specific sections (“Revise the second and third paragraphs, explaining how specific details from the text led you to your conclusions.”), or add specific ideas (“Add a new point somewhere in the previous draft addressing why the third data point is an outlier from the others.”).
- Ask the AI to follow its own advice. Some AI platforms will follow up each draft with suggestions for how to revise it. You can ask the AI to explain and address these suggestions itself. For example, “Where would I add more detailed evidence in the second paragraph?” “Okay, revise the second paragraph, adding more detail in the way you suggested.”
Revising with AI:
Note: for this exercise, it’s best to use a real student draft, if possible. There are, of course, ethical considerations around feeding student work into a tool owned by a third party, so we strongly recommend that you get students’ permission if you plan to use their work this way. Short of that, you can also copy an AI-generated draft from the previous exercise, start a new chat, and use that draft as a starting point for this exercise.
- Give the AI your context, the essay prompt, and your draft. Then, ask it to give you suggestions for how to revise the draft. For example: “I’m a college student in a 200-level Sociology class. I’m writing an essay in response to this prompt… Here is my draft… Give me some suggestions for revising my essay.”
- Ask the AI to identify your thesis and to summarize your argument in one paragraph. If the AI can do this effectively, that indicates that the essay has a clear thesis and a clear overall argument; if it can’t, then that indicates a major problem with the thesis or structure.
- Tell the AI to correct all grammar errors in the draft and identify all changes in bold. This won’t accomplish much if you’re using an AI-generated draft, of course, because AI rarely makes grammar errors.
- Ask the AI to assess your tone and writing style. For example, “Is my writing formal enough for this type of assignment?” or “Identify places where I could be more concise.”
- Have the AI implement its suggestions for you. For example, “Okay, rewrite the sections you just identified to make them more concise.”
Step 3: Consider the Results
After each exercise above, assess the AI’s results:
- How well did it complete each task?
- What limitations, misunderstandings, or errors did you see in its output?
- How might using the AI this way help or hinder students completing this assignment?
- Could a student get an A on this assignment using AI in this way?
- If so, how much work would they have to do? What aspects of the AI’s work would they need to edit, revise, fact-check, or supplement?
- If a student did produce an A paper using AI this way, would they still do the intellectual work the assignment is designed for?
Step 4: Consider What’s Next
How you respond to GenAI in your classes is, of course, up to you as the instructor. However, we can offer some general advice:
It’s vital to have a clear and comprehensive AI policy for your course.
This should ideally be written into the syllabus and restated on every major writing assignment. There are several helpful guides for writing AI course policies available online (here’s a pretty good one from Carleton College). AI policies generally break down into four categories:
Total Ban: Students may not use AI in any way on the assignment. If you choose to go this route, you should still specify what range of AI uses you mean–for example, are your students still permitted to use editing tools like Grammarly, as long as the tool doesn’t write any text for them?
Some Use Allowed, with Permission: Students may use AI for some specific purposes, as long as they get your explicit permission in advance. This is a nicely flexible option, since it allows you to consider students’ requests on a case-by-case basis, rather than trying to list every allowed and prohibited use ahead of time. However, it’s not practical for high-enrollment classes, where dozens of student requests would be overwhelming.
Some Use Allowed, with Documentation: Students may use AI in certain specified ways without prior authorization, as long as they document their AI use (by exporting the chat log and submitting it with their assignment, for example).
Total Freedom: Students can use AI in any way they see fit without prior authorization and potentially without documentation. Even if you choose this route, it’s worth explaining to students why you’re not particularly concerned about AI use in your course.
It’s worth noting that, if you have no stated AI policy on your syllabus or assignments, you effectively have a Total Freedom policy by default.
For more suggestions, see our guide to writing AI course policies.
No matter what policy you set, you should spend some time explaining it in class.
Students are more likely to respect the boundaries you draw around AI use if you explain the reasoning behind them. How does the writing AI produces fall short of your assignment goals? Even if AI could be helpful, how would using it hurt their learning? There will, of course, always be students who are willing to cheat, but if you assume that most students are willing to learn, and if you respect them enough to explain how your course policies are designed to help them learn, they’ll listen more often than not.
If you allow some AI use, consider teaching students how to use it well.
As the exercises above illustrate, there’s a definite craft to getting the most out of AI tools, and that craft is just as new to our students as it is to us. So, if you’ve tested AI on your assignments and identified certain techniques that work well or certain pitfalls that students will likely encounter, you should tell them.
Stress that your policies only apply to your class.
Students generally assume that policies around AI use, plagiarism, and academic misconduct are universal, so what counts as fair use in one class will be fair use in all of them. You’ll do your students and your colleagues a favor if you dispel this notion explicitly. This is especially true if you have a fairly open policy on AI use–tell your students that their other instructors are likely to be less open, and it’s their responsibility to know the rules for each class and each assignment!