94% of higher education professionals surveyed in late 2025 reported that they had used AI in their work in the past six months. Among the top tasks they reported using AI to complete: drafting emails, summarizing long documents or meetings, proofreading or copyediting, and creating presentations.
We want AI to save us time.
But in my conversations — both formal, as part of my academic research into operationalizing AI on marketing teams, and informal — I hear two common threads: 1) tell me exactly what AI tools will save me time, and on what tasks; and 2) tell me exactly what is ok and not ok to use AI to do.
Unfortunately, institution-wide policy is not designed to operate at that level of specificity across the full range of work we do. Even highly detailed guidance within a single division can struggle to keep pace with how quickly the technology changes.
So what is the AI-curious higher ed professional to do?
When we set out to complete a task, we have many tools at our disposal: for writing this blog post, For this blog post, I’m using Google Docs in Chrome, with an external keyboard, mouse, and second monitor. I don’t think about those choices. Experience made them automatic.
However, generative AI is a new tool that many of us have very little experience with. Deciding between a mouse and a trackpad happens instantly because I’ve done it hundreds of times. Deciding whether AI will help, and which tool to use, takes more effort. We don’t realize the many steps we take during the automatic tool selections we make every day. So before we can make these decisions quickly and automatically, we must practice making them carefully and intentionally.
Instead of interpreting policy language in real time, I built a simple decision tool that walks through five questions:
1. What is the task?
If you cannot clearly define what you are producing and who it is for, the output will drift. The scorecard forces that clarity up front and helps us think carefully about the specific skills we are assigning to a machine.
2. What is the risk level?
Not all work carries the same consequences. The scorecard separates:
Low risk: internal work, low visibility, easy to fix
Medium risk: external content that is visible but not sensitive
High risk: institutional, legal, or reputational work
Each category sets a different ceiling for AI involvement. High-risk work usually means limiting or avoiding AI altogether.
3. Is AI useful here?
If AI is not helping you start, generate options, improve something, or handle repetitive work, skip it. If it will take more time to generate the result you’re looking for from AI than it will to do the thing yourself, why waste time (and environmental resources)? This is common for tasks that are too complex or nuanced for AI, or tasks that the technology is just not as good or fast at (like generating full images). If, however, it’s a task you do daily, it might be worth spending more time training the large language model once for time saved over multiple sessions.
4. Can I use AI safely?
Three conditions need to hold:
No sensitive or private data will be used.
You, a human, will review and edit everything.
You, a human, are capable of effectively evaluating the quality and accuracy of the output for this task.
If any of those fail, the task is not a fit.
5. Am I willing to own this?
Would you put your name on the result? Do you understand what it produced? Would this hold up if someone asked how it was created?
If not, stop.
You can access the scorecard here.
Moving from Permission to Judgment
There is a clear desire for simple rules that map directly to daily tasks. That would make decisions easier. In practice, the range of roles, tools, and scenarios makes that difficult to maintain. Instead, the more reliable approach is building judgment that holds across situations.
It is incredibly important to note that this framework does not replace Miami’s policies, divisional guidance, or what your boss thinks. But, hopefully, it can help you translate all that into decisions you can make while you work.