Ground Yourself in Principles
Rule of thumb: Let AI be your assistant, not your author. Keep asking: “Am I still doing the intellectual work this assignment is designed to help me learn?”
“The key is knowing when to trust, when to question, and when to verify. This comes best from experience.”
David Wood, Rewiring Your Mind for AI (Wood, 2025)
How These Tools Work
Chat tools such as ChatGPT and Microsoft Copilot Chat run on large language models. A model learns patterns from enormous amounts of text and writes a response by predicting what text is likely to come next. Unless the tool also searches the web or your files, it is not looking facts up, and even when it searches, it can misread what it finds. That is why AI can give a fluent, confident answer that is wrong, cite a real source that doesn’t say what the AI claims, or invent a citation. Models also carry the biases of the text they learned from.
Newer tools called agents go further. They can open files, browse websites, run code, and take actions on your behalf. Step 4 covers what that means for you.
Principles to Work By
Lead with Your Thinking
Your ideas, reasoning, and judgment should drive the work. Sounding like you is necessary but not enough. You can rewrite AI output in your own voice and still skip the thinking the assignment was meant to build. Try the problem yourself first, then bring in AI, and use it in ways that build skills you will need after UVA.
Be Transparent
Disclose when and how you use AI (Step 6). Unpermitted or undisclosed AI use can be an Honor offense (Step 1).
Own the Work
You are responsible for the accuracy, quality, originality, citations, and integrity of everything you submit, including material developed with AI (UVA FAQs). Rewording AI output does not make the ideas yours. Owning the work means you can explain it, defend it, and correct it.
Work Iteratively
Use follow-up prompts to question, refine, and stretch ideas. Push back on answers the way you would with a study partner.
Protect your Privacy
Use University-supported tools. UVA’s licensed versions of Microsoft Copilot Chat, Google Gemini, and Notebook LM keep your prompts out of model training (UVA Library, 2026), and UVA’s agreement for its ChatGPT workspace does the same. In Copilot Chat, a green shield with a check mark shows you are signed in with UVA’s protections. In ChatGPT, confirm you are in the UVA workspace and not a personal account.
Chats in a project or custom GPT you share with others, including one an instructor shares with a class, can be visible to the other members (OpenAI, 2026). Keep private material out of shared spaces.
Even in a supported tool, some information needs approval before it goes in. Do not enter confidential, sensitive, regulated, or proprietary information unless UVA has approved the tool for it. That includes the following. The FAQ “What information should I not enter into an AI tool?” has the current list.
- personal or health information, yours or anyone else’s;
- student records;
- unpublished work and research data;
- another person’s private or copyrighted material; and
- your NetBadge password, one-time codes, or any other credential.
Your own class notes are generally fine to use. Upload an instructor’s slides, recordings, or exam materials only with permission (adapted from Monash University). With any tool outside UVA’s agreements, treat what you type as if you were posting in a semi-public forum. See the ITS AI tools list, Using AI at UVA: Guidelines & Terms of Use, and the University Data Protection Standards.
Recording is covered too. Get your instructor’s permission before you record a class or run an AI note-taker, as UVA’s policy on recording classroom lectures requires, and ask before you record meetings or office hours.
Think Critically
AI output can be inaccurate, incomplete, biased, or fabricated, including citations that do not exist. More often, the source is real but unreliable, the AI summarized it wrongly, or it doesn’t address your question. Don’t trust an answer, quotation, calculation, or citation because it sounds confident (UVA FAQs). Lo’s CARE approach gives you a routine for reading AI answers (Lo, 2026a):
- Classify the answer. Is it factual, interpretive, constructive, or strategic? Each kind needs a different check.
- Assess it. Does it hold up? What is missing, assumed, or one-sided?
- Review the evidence. Trace claims, quotes, and data back to their original sources, and read those sources yourself (Caulfield, 2019).
- Enhance it with your own knowledge, course materials, and better sources.
Before you rely on an answer, ask three questions. Does accuracy matter here? Do I know enough to verify it? Am I willing to take responsibility if it is wrong (adapted from University of Wisconsin–Madison Career Services)? Never cite an AI tool’s description of a source. Find the original, confirm it says what you claim, and cite it (UVA FAQs).
Consider the Wider Impact
AI affects more than your assignment. Data centers that run these tools consume large amounts of electricity and water. Access and quality are uneven across people and languages. The tools are changing what employers look for in graduates. Part of AI literacy is weighing those effects when you decide whether and how to use AI (Lo, 2025b).
Quick Self-Check
Before you use AI on coursework, ask yourself:
- Am I still doing the intellectual work this assignment is designed to help me learn? (UVA FAQs)
- Is this helping me learn, or just saving time? (Duke University Academic Resource Center)
- Is it enabling or hindering my mastery of the course objectives? (University of Michigan)
- Guidance for Students (University of Florida)
- Have I checked whether AI is allowed for this task? (University of Toronto)
When AI Doesn’t Help
Sometimes AI responses will be off-target, shallow, or wrong. When that happens, adjust your approach.
- Revise your prompt. Add context, constraints, or an example.
- Try a different angle. Ask for an example instead of a definition, or ask it to question you.
- Switch tools. Different University-supported tools have different strengths.
- Ask for the slower path. Most AI tools are designed to hand you answers. For learning, ask for hints, questions, or feedback instead.
- Talk to a human. Your instructor, a TA, classmates, the Writing Center, or a UVA librarian may be a better next step than another prompt.
Choosing a Tool
Tools change quickly, so learn the categories rather than the brands. The table lists current examples.
Type of Tool |
Good for |
University-Supported Examples |
Watch for |
|---|---|---|---|
|
General-Purpose Chat Assistant |
Explaining concepts, brainstorming, feedback on your drafts, tutoring |
ChatGPT (UVA workspace); Microsoft Copilot Chat; Google Gemini |
Confident errors; follow the data rules above |
|
Study tools grounded in your materials |
Organizing course materials you are allowed to upload, study guides, practice questions |
Google NotebookLM |
Upload only materials you have permission to use; still check answers |
|
Research search tools |
Finding scholarly sources and summaries of findings |
Consensus; ChatGPT search and deep research |
Open and read every source; cite the source, not the tool |
|
Coding assistants and agents |
Explaining, writing, and testing code |
ChatGPT can explain and write code; dedicated coding agents are not licensed for students yet, so check the ITS list |
See the coding agent guidance in Step 4 |
|
Writing and editing assistants |
Grammar, clarity, and style |
Chat assistants above can do this; tools like Grammarly are outside UVA agreements |
Often restricted in writing and language courses |
|
Image generators |
Visuals and design concepts |
ChatGPT, Copilot Chat, and Gemini can generate images |
Copyright; label AI-generated images |
Sources: UVA Library, 2026; ITS AI tools list. For any tool not on the ITS list, such as Perplexity, a personal Claude account, or a personal ChatGPT account (free or paid), do not enter protected information or your own unpublished work.
The sources cited on this page are listed in the Toolkit References.