The narrow claim is that Copilot helped with an initial synthesis. It did not write the plan. That distinction matters for everyone who runs Copilot in an organization.
What JMU says it did
According to The Breeze, JMU used Copilot, a generative AI chatbot, to help develop the five pillars of President Jim Schmidt's strategic plan. The plan is now finalized as The Madison Promise. The paper says the university collected more than 200 pages of online feedback from faculty, students and staff.
Chris Orem, JMU's associate vice president for Planning, Analytics, and Institutional Research, described the reasoning. He said AI could give an initial synthesis that might otherwise take longer, but the team did not take it at its word.
Orem said his team gave Copilot multiple prompts to boil the feedback down to common concepts. The reported workflow was:
- Two teams. A lead team included the plan's three co-chairs. A three-person logistics team included Orem.
- Human presence. Orem said team members attended the feedback events. When the AI produced responses from specific feedback, staff reviewed them until they were confident in the results.
- Leadership iteration. The vice presidents and co-chairs then worked through what the pillars meant in the broader context.
- Stated rationale. The work ran on a tight timeline. Orem called AI a starting point, not the final product.
All of this is Orem's account as reported by The Breeze. The paper gives no prompt logs, no error rate and no comparison against a human-only summary. Nothing in the reporting shows whether the output was accurate, complete or free of bias. I can't say whether the safeguards worked. I can only say what JMU says they were.
The plan and its timeline
JMU's own pages corroborate the broader process, but not the Copilot details. JMU's Planning, Analytics and Institutional Research office describes an eight-month process that began with President Schmidt's August 2025 announcement. It was built on listening sessions and idea forums.
JMU's Board of Visitors endorsed "The Madison Promise: JMU 2032 Strategic Plan" on April 24, 2026, with long-term targets reaching to 2040. The five pillars are:
- Enhance the purpose, value and impact of a JMU education.
- Invest in our people.
- Chart a new path as a national research university.
- Become a preferred partner for the Commonwealth and beyond.
- Create an innovative campus community.
The Board also approved a 2027 Campus Master Plan. JMU describes it as a physical roadmap for the strategic plan. It warns that its conceptual maps and drawings do not mean construction is imminent. The Breeze reports the plan received $3.37M for its initial implementation phase.
Which Copilot was it?
This is the detail IT readers will care about, and the source doesn't settle it. The Breeze says "Microsoft Copilot." JMU's AI platforms page lists three Microsoft offerings:
- Copilot Chat. It is free to the JMU community. JMU says that when users are logged in, Microsoft does not monitor their access and no data is ingested to train the underlying models.
- M365 Copilot. It is an extra-cost subscription for employees, used inside apps such as Teams, Outlook, Word, PowerPoint and Excel.
- Copilot Studio. It is an extra-cost employee subscription for building custom agents, and it needs approval through a Technology Solution Request.
Nothing in the reporting says which of these the planning team used, or which account type. Treat any assumption about that as speculation.
JMU's AI data page adds that Copilot is the only AI tool sanctioned for use with JMU data as input, and only under the university's data-protection policies. The page also notes that Copilot keeps an activity history that users can delete. I could not verify the published standard's wording that prohibits "Highly Confidential Data" in Copilot Chat. That claim came from background notes, not the pages I could read. It is also unknown how JMU classified the planning feedback.
What Microsoft's documentation says
JMU's claims line up with Microsoft's general statements about work and school accounts. Microsoft's support documentation says Copilot Chat provides enterprise data protection when you sign in with a work or school account. Prompts, including any work content you add, and Copilot's responses aren't used to train foundation models.
There is a catch for administrators. The same Microsoft page says prompts and responses are logged, and an IT admin can view that logged information with Microsoft's search and audit tools. Microsoft's Learn documentation also says logged prompts and responses can fall under the same retention policies used for Microsoft Copilot.
Two practical points follow:
- "Not used for training" does not mean "not stored." Logs can be searched, retained and audited.
- A university that feeds community comments into Copilot has created records it may need to manage. I'm drawing that inference from Microsoft's logging documentation. JMU has not said anything about retention for this project.
The campus reaction
The Breeze reports three skeptical voices, all attributed opinions rather than findings that anything went wrong:
- Melanie Shoffner, Faculty Senate Speaker and professor of English education, called overreliance the biggest problem. She said technology can turn from a tool into a default.
- Andrea Miller, a sophomore political science major, worried about inaccuracies because AI is known to hallucinate. She wants a person to vet the feedback.
- Maddy Probst, a freshman architectural design major, said a planning process should be human. She argued that using AI makes it harder to push critical thinking among students.
Miller also pointed to a perceived contradiction. Her department heavily discourages AI use, yet the university's plan used it.
The Breeze says JMU doesn't have, and can't have, a single university-wide AI policy because departments differ. It says faculty are expected to set their own policies in syllabi. That framing deserves care. JMU now publishes university-level AI pages covering approved tools and data handling. The working-group page shows that Orem is on an AI Strategy Working Group whose remit includes governance recommendations, guiding principles, and opportunities and risks. The sources don't say the planning episode prompted that group.
What the process does and doesn't prove
Clustering recurring themes in hundreds of comments is a sensible job for a language model. It is also where subtle failure can hide. A summary can overweight what is repeated and bury what is rare. The rare comment is sometimes the important one. Synthesis is not the same as deciding whose input counts. That judgment stayed with people, according to JMU.
The reported safeguards are the right ones in principle: attendance at the events, review of outputs, and leadership deliberation on top. An outside reader can't verify how rigorously they were applied.
Takeaways for IT teams
If you are piloting Copilot for similar work, a few habits follow from this case:
- Disclose early. Much of the friction here is about the AI use becoming known after the fact. The Breeze's piece cites an April article, but the debate is only now surfacing.
- Name the product and the account. "We used Copilot" isn't a governance statement. Say whether it was Copilot Chat with a work account or a licensed M365 Copilot.
- Classify the data first. Know whether the input is confidential before it goes into a prompt.
- Keep humans in the loop. Document who reviewed what, and how disagreements with the AI summary were resolved.
- Plan for logs. Prompts and responses may be retained and discoverable.
JMU's case shows AI as an accelerator for a first pass. Whether students and faculty accept that depends less on the tool than on how openly the university explains its use.