Advice for MBA/MPA/MPP Students on How to Use Generative AI (GAI) in Your Classes

Note: This guidance applies specifically to my courses. Check with other faculty before using it elsewhere. Some examples and exercises below require a paid subscription to a frontier AI version, and may not work with free versions.

Embracing Generative AI: A New Era

The transition from slide rules to calculators, or typewriters to word processors, mirrors the rise of Generative AI (GAI). The tools you use now are the most basic GAI you will ever encounter. Mastering these tools is a vital skill for your careers, and is something I want you to learn in this course.

While my hypothesis is that GAI can help you learn more deeply and effectively, we also know that frequent AI use can encourage cognitive offloading, potentially hindering critical thinking. The key is to use GAI in ways that challenge you to improve your thinking, not as a crutch to avoid it.

Recommended Applications of GAI for Student Use

1. Writing, Research and Content Creation

• At a minimum, use GAI for proofreading and copyediting. Personal note: I grew up in a home where English was a second language, neither of my parents attended high school (never mind college), and for various complicated reasons, the first year I completed a full year of school was in 3rd grade (not an ideal learning situation). So basically, my written English can sometimes be a bit chaotic and yes, I find GAI to be a very helpful proofreader and copyeditor.
• Draft and edit research content collaboratively. GAI can make a great thought partner — share your ideas, and get feedback. You can even do research live while editing — the Canvas feature in ChatGPT can be very helpful for this.
• Transcribe handwritten notes or equations and format them professionally. For instance, you can upload a picture of handwritten notes and have the AI transcribe and edit them (this works with ChatGPT; functionality may vary with other implementations). But remember to get value from your notes — you will need to read and revise them.
• Assist in literature reviews and gather background research (note the highlighted word “assist”; also use conventional approaches). For literature reviews and research, I suggest trying Consensus (a GPT available in ChatGPT) as well as Gemini 1.5 Pro with Deep Research, if you have access. This is a comparison of what these research engines found when given the same prompt: link. However, note that content behind paywalls (e.g., recent peer reviewed academic work in paywalled journals) will often not be accessible by these AI tools, hence the need to also apply conventional approaches.
• Experiment with the podcast and summary features in Google’s NotebookLM. But this is best used for optional material, when you are considering if you need to invest the time to learn it in detail and/or take a deeper dive. Or, you can use it to make a podcast about something you are working on (e.g., your final paper, a presentation you are making in class, and so on). For a sample podcast that describes my spring course on implementing urban economic development, try this link: link
• GAI is excellent at translation, making it particularly helpful for foreign language sources. Also, ChatGPT Advanced Voice can act as a simultaneous translator.

2. Persuasion

• GAI is very effective at persuasion. For presentations, practice by presenting to the GAI and asking for critiques about how you can be more persuasive. Share your slides, talking points, or use advanced models like ChatGPT’s voice mode for simulated conversations. Ask for feedback and advice.
• Example exercise: Use GAI for persuasion in this sample exercise — Communications and Persuasion Exercise: LLMs as Communicators, Persuaders, and Simulated Stakeholders (link).

3. Data Analysis, Coding, and Creating No-Code Apps

• Utilize tools like ChatGPT for coding and debugging assistance. One of my students used ChatGPT to learn coding during my Fall 2025 course.
• Employ advanced data analysis features for visualization and insights. Classify and analyze textual data, such as sentiment extraction. Sample assignment: Using ChatGPT to analyze Boston 311 data — link. (this also serves as introduction to writing more complex prompts)
• Building a no code app, one of the interesting new features of GAI, is the ability to program with natural language. This is a student exercise to create a no code app for data analysis: link. For your course, think about how you can create a no code app to solve some type of problem or challenge. Remember, it does not need to be perfect or a production version. A Minimum Viable Product will do.

4. Mathematical Work

• Use tools like OpenAI’s o1 series for derivations and proofs.
• Verify and refine mathematical solutions collaboratively with GAI.

Best Practices for Effective Use

• The best way to learn is by using a GAI — though it can be frustrating initially. Plan on spending at least 20 hours with the AI before you think of yourself as “onboarded”.
• Always verify AI outputs, particularly for critical tasks. Current implementations still have issues around hallucinations, guardrails, bias and company policies.
• Be mindful of data confidentiality when using GAI.
• “AI is not good software. It is pretty good people” (E. Mollick, 2023). GAI is designed to produce varied responses. Asking the same question in separate sessions may yield different answers.
• If the model refuses to do something that you think it should be able to do, try these strategies: rephrase your prompt, small changes in a prompt can have a big impact, reassure it that it can perform the task and tell it to try again (yes, this sometimes works), encourage persistence, explain the task’s importance, or restart the session and begin again. :-)

Implementation Tips

Always remember, you are responsible for what you submit. It is never acceptable to attribute errors to GAI or claim ignorance about its outputs. You must understand and be able to explain the content you use. Particularly in my classes, if you submit work in writing I am assuming it is your work, and if I call on you in-class you will be able to explain it.

GAI is not a search engine. While it excels at synthesis, it exercises editorial control, which can sometimes lead to bias or errors. For an interesting example of this problem, please see:

• Beres, Damon (2024) Why does ChatGPT refuse to say certain names? The strange scenario is a reminder that despite the mystique tech companies cultivate around their AI products, firms still have an awful lot of direct control. Atlantic Intelligence. link
• and for my view on this, please see: Strauss, Steven. (2024). “Ok, this is weird: ChatGPT will not say certain names.” link

Start small, using GAI for simpler tasks, and gradually expand its application. Use it to complement — not replace — your analytical thinking.

And, for my students — I look forward to seeing how you integrate these tools into your learning process and use them to challenge yourselves academically!

Best regards,

Steven Strauss, Ph.D.
John L. Weinberg/Goldman Sachs & Co. Visiting Professor
Princeton University School of Public and International Affairs
Follow me on LinkedIn: link

Bibliography

Carrasco-Farre, Carlos. (2024). “Large Language Models Are as Persuasive as Humans, But How? About the Cognitive Effort and Moral-Emotional Language of LLM Arguments.” arXiv preprint arXiv:2404.09329.

Costello, T. H., Pennycook, G., & Rand, D. G. (2024). “Durably reducing conspiracy beliefs through dialogues with AI.” Science.

Gerlich, M. (2025). “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking.” Societies, 15(6). https://doi.org/10.3390/soc15010006.

Korinek, Anton. (2024). “Generative AI for Economic Research: LLMs Learn to Collaborate and Reason.” NBER Working Paper 33198, National Bureau of Economic Research, Cambridge, MA.

Lopatto, E. (2024). Stop using generative AI as a search engine: A fake presidential pardon explains why you can’t trust robots with the news. The Verge. https://www.theverge.com/2024/12/5/24313222/chatgpt-pardon-biden-bush-esquire

Mollick, E. (2023). “On-boarding your AI Intern.” https://www.oneusefulthing.org/p/on-boarding-your-ai-intern.

Mollick, E. (2024). “Getting started with AI: Good enough prompting.” https://www.oneusefulthing.org/p/getting-started-with-ai-good-enough.

Mollick, Ethan. (2024). “Today’s AI is the worst AI you will ever use.” https://www.linkedin.com/posts/emollick\\\\_todays-ai-is-the-worst-ai-you-will-ever-activity-7106305750431322112-Xr7n/.

Rettenberger, L., Reischl, M., & Schutera, M. (2024). “Assessing Political Bias in Large Language Models.” arXiv preprint arXiv:2405.13041.

Salvi, Francesco, Horta Ribeiro, Manoel, Gallotti, Riccardo, & West, Robert. (2024). “On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial.” https://arxiv.org/abs/2403.14380.

Strauss, Steven. (2023). “Why I’m encouraging my students to use Generative AI.” https://link.medium.com/AqtfINTe6Pb.

Strauss, Steven. (2024). “Gentle Reminder: LLMs/AIs like ChatGPT and Perplexity are excellent for generating and synthesizing information but unreliable as true search engines.” Link https://www.linkedin.com/posts/ssstrauss_search-llm-ai-activity-7271302655103176705-Pklc?utm_source=share&utm_medium=member_desktop

Strauss, Steven. (2024). “Ok, this is weird: ChatGPT will not say certain names.” Link

Wolfram, S. (2023). “What Is ChatGPT Doing … and Why Does It Work?”, Wolfram