Harvard doesn't publish a single answer to "can I use AI?" It requires every instructor to give you one. That means the policy for your Expos section and your econ problem sets can be completely different, and both are binding.
Here's what Harvard's official guidance says, how to read the policies you'll see on Canvas, and which models fit the work Harvard students do.
What Harvard actually says
We checked these pages in October 2026.
The Office of Undergraduate Education. OUE's generative AI guidance says all faculty must tell students the policies governing generative AI in their courses, and should post them on Canvas. It gives instructors three sample policies:
- Maximally restrictive: no generative AI at any stage of the work, including early brainstorming.
- Fully encouraging: AI is welcome on all assignments, with acknowledgment and citation.
- Mixed: AI is allowed for some things and not others. In that sample, "the default is that such use is disallowed unless otherwise stated."
All three samples say breaking the course policy counts as academic misconduct.
Privacy. The same page is firm: "No confidential information can be loaded into GAI systems," because users have no expectation of privacy. It also notes that some AI companies' terms let them access what you enter.
University-wide guidelines. Harvard's initial guidelines on generative AI tell students to follow existing academic integrity policies, review their school's handbook, and ask instructors when a policy isn't clear.
For the university-supported AI tools available to you, check HUIT's AI pages with your HarvardKey. Availability differs by school and changes often.
What that means in practice
- Read the Canvas policy for every course. It's required to be there. If you can't find it, ask.
- Mixed policies are the trickiest. If a course allows AI for some tasks, anything not listed is likely off limits.
- Acknowledge and cite. Even in courses that encourage AI, the sample policy expects you to say how you used it.
- Don't paste confidential material into public tools. That includes research data, other students' work and anything from a job or internship.
Which models suit Harvard coursework
Expos and writing-heavy courses
Harvard's first-year writing requirement, Expos, is about learning to build an argument yourself. Each section sets its own AI policy, so read yours closely. If it allows any use, keep it to feedback on your own draft.
Try: "Here's my thesis and the topic sentences of each paragraph. Don't rewrite anything. Tell me which paragraph doesn't support the thesis and why."
Claude is a careful reader for this kind of feedback. Our ChatGPT vs Claude for students post compares the two on writing tasks.
Economics
For econ, use AI to understand a model before you apply it on a problem set.
Try: "Explain the intuition behind why a price ceiling below equilibrium creates a shortage, then show how deadweight loss appears on the graph. Use words, not just formulas."
ChatGPT and Gemini are both strong on quantitative reasoning. Check any numbers against your lecture notes.
Government, history and social studies
These concentrations live on reading and sources. Use a search model that links to real pages, like Perplexity, to find starting points, then go to the primary sources. Never cite a source you haven't opened.
Try: "I'm arguing that the 1965 Voting Rights Act changed Southern party alignment. What are the main scholarly positions on this, and which authors should I read first?"
For reading long documents, our Gemini vs ChatGPT for research students post goes through which handles big PDFs better.
Pre-med and sciences
Use AI to explain and quiz you, not to write lab reports.
Try: "Quiz me on the steps of the citric acid cycle. Ask one question at a time and tell me what I got wrong before moving on."
How to use AI at Harvard without breaking the rules
- Find each course's AI policy on Canvas in week one. Save a copy.
- Figure out which type it is. Restrictive, encouraging or mixed. If it's mixed, list exactly what's allowed.
- Ask your TF or instructor when unsure. Harvard's guidelines tell you to ask, and a short email is enough.
- Acknowledge every permitted use. Tool, task and how you used the output.
- Verify facts and citations yourself. Models invent references, and a fake citation is hard to explain.
- Keep confidential material out. OUE's rule on this is not a suggestion.
Where LazySusan fits
For course material, start with the Harvard-supported tools listed by HUIT. LazySusan is for the rest: personal projects, club work, side businesses or comparing models before you trust an answer.
It brings ChatGPT, Claude, Gemini, Grok, Perplexity and DeepSeek into one workspace with one bill, along with image, video and voice models. Multi-Chat sends one prompt to several models and shows the answers side by side, which is the quickest way to see where models disagree and a fact needs checking.
The Student plan is $99 a year with a .edu email, with 5,000 tokens a month that roll over. Prompt School has free lessons on prompting, and you can sign up here.
LazySusan isn't affiliated with or endorsed by Harvard University. Your course's Canvas policy comes first.
The short version
At Harvard, every course must tell you its AI policy, and they range from no AI at all to AI everywhere. Read each one, ask when it's unclear, acknowledge what you use and keep confidential information out of AI tools.