Workshop 2: Workflow Enhancement
Workshop 2: Workflow Enhancement
Use ChatGPT Edu, Copilot Chat, and Microsoft 365 Copilot to speed up the work you already do
Tuesday, July 7, 2026 | 1:00–2:00 PM | Bring a laptop and a real (non-sensitive) task
Didn’t bring a task? Download a synthetic demo file (all content fictional and non-sensitive): Study description (Word) · Committee email thread (Word) · Seminar attendance data (Excel) · Three abstracts (Word)
What You’ll Leave With
By the end of this hour, you will be able to:
- Spot which of your recurring tasks are good candidates for AI help
- Use Microsoft 365 Copilot inside Word, Outlook, and Excel for everyday tasks
- Use ChatGPT Edu with GPT-5.5, and Copilot Chat with Enterprise Data Protection, to summarize, rewrite, and analyze faster
- Build a small “prompt recipe” you can reuse
- Apply the workflow ethics: verify before it counts, disclose AI use appropriately, and avoid over-reliance
Run of Show
| Time | Segment |
|---|---|
| 0:00–0:08 | Overview: where AI fits in a workflow; Copilot Chat, Microsoft 365 Copilot, and ChatGPT Edu |
| 0:08–0:23 | Ethics & Governance: verification, disclosure, and over-reliance |
| 0:23–0:42 | Activity 1: Microsoft 365 Copilot in Office |
| 0:42–0:55 | Activity 2: ChatGPT Edu for your real task |
| 0:55–1:00 | Wrap-up and save your prompt recipe |
Overview (8 min)
Where AI fits. The biggest time savings come from the repetitive middle of a task—the first draft, the summary, the reformat, the “make this clearer.” AI handles that draft; you bring the judgment and the final word.
Copilot vs. ChatGPT Edu—when to use which.
- Microsoft 365 Copilot shines when the work is already in a Microsoft file: a Word doc to rewrite, an Outlook thread to summarize, an Excel sheet to analyze. It sees the document you’re in and respects Microsoft 365 permissions.
- Copilot Chat with Enterprise Data Protection is useful for web-based drafting, summarizing, and brainstorming when you do not need a reusable GPT.
- ChatGPT Edu shines for open-ended drafting, summarizing pasted text, file-based analysis, and building reusable assistants. Use GPT-5.5 when it appears in your model picker; otherwise use the current default/latest GPT model in the Drexel workspace.
Use each tool according to Drexel’s data classifications. See Tools & Access.
A task is a good AI candidate if it’s (1) text-heavy, (2) something you do repeatedly, and (3) low-stakes enough that a draft-plus-review is fine. Think: status updates, meeting recaps, first drafts, summaries, reformatting.
Ethics & Governance: Verification, Disclosure, and Over-Reliance (15 min)
The thread continues. Workshop 1 covered how AI fails and who’s accountable. Today: what that accountability means once AI is in your daily work.
Verify before it counts. Speed is the selling point and the trap. AI can fabricate facts, statistics, and citations that read as authoritative. For anything that goes into a real document or decision, check it against the source—especially numbers and references. A quick rule from the intensive course: if you can’t open it, you can’t cite it. Treat every AI-supplied figure as unverified until you’ve confirmed it.
Disclose appropriately. Norms are still forming, but the safe defaults are clear: for official work products, external communications, and anything research-related, be transparent that AI assisted, following Drexel’s AI policy and your unit’s expectations. Disclosure protects your credibility and models the behavior we want across the school. When unsure, ask your supervisor or PI what level of disclosure they expect.
Don’t over-rely. If AI does all the drafting and thinking, two things erode: your own skill and judgment, and your voice. Outputs also drift toward a generic sameness. Use AI to get past the blank page and to pressure-test your thinking—not to replace it. Read what it produces critically, keep your own voice, and stay the expert in the loop.
Confidentiality in everyday workflows. Even approved tools have limits. Microsoft 365 Copilot can see the document and files you have access to—make sure that’s appropriate before you ask it to summarize across them. Match every task to Drexel’s data classification, and never paste regulated data into a tool not cleared for it. See Tools & Access.
Three checks on anything AI helped write: Verified? (facts and citations confirmed). Disclosed? (AI use noted where expected). Yours? (it says what you mean, in your voice). If not all three, it’s not ready.
Activity 1: Microsoft 365 Copilot in Office
Activity 1: Microsoft 365 Copilot in Office
Goal: Use Microsoft 365 Copilot where your work already lives. Pick the track that fits you. Use non-sensitive or appropriately classified content only. Each track has a worked example with a demo file — use it as-is, or substitute your own material and run the same three prompts.
Track A — Word (drafting & rewriting)
Worked example: from study aims to lay summary. Open Workshop2_Demo_StudyDescription.docx (a synthetic pedestrian-safety study description) or your own document, then run:
- “Summarize this document in 5 bullet points for a project status update.”
- “Rewrite the first paragraph for a community advisory board — no jargon, 8th-grade reading level, keep it to 4 sentences.”
- “Draft a short ‘Next Steps’ section: one paragraph, three concrete steps, active voice.”
Watch for: does the lay version keep the facts accurate? Did anything get invented? That’s your verify step.
Track B — Outlook (email triage)
Worked example: find the decisions, draft the reply. Open a long thread of your own, or use the mock 6-message curriculum-committee thread in Workshop2_Demo_EmailThread.docx (paste it into Copilot Chat if you’re not working in Outlook), then run:
- “Summarize this thread. List every decision made and every open question.”
- “List the action items as a table: owner, task, due date.”
- “Draft a reply confirming I’ll have the syllabus draft ready by July 20 — concise and friendly, two sentences.”
Watch for: did it invent a decision that was never actually made? Threads are where AI over-summarizes — in the demo thread, is the accreditation crosswalk a decision, an action item, or an open question?
Track C — Excel (quick analysis)
Worked example: seminar attendance and spend. Open Workshop2_Demo_SeminarAttendance.xlsx (synthetic monthly attendance and catering cost for three departments) or your own sheet, then run:
- “Summarize this data. What stands out?”
- “Add a column that flags rows where cost per attendee is over $12.”
- “Suggest a chart showing the attendance trend by department across the year.”
Watch for: check the flag formula against one row by hand before you trust the column.
Reflection
- Did Copilot save time versus doing it by hand? Where did you still need to correct it?
- Which of your weekly tasks could this replace the first-draft step for?
Activity 2: ChatGPT Edu for Your Real Task
Activity 2: ChatGPT Edu for Your Real Task
Goal: Run one task you actually need to do this week, start to finish.
Instructions
- Pick a real task that is non-sensitive or appropriate for ChatGPT Edu under Drexel’s Terms of Use: a report section, a summary of a long document, a set of meeting notes to clean up, a draft announcement, a literature blurb on a public topic.
- In ChatGPT Edu, select GPT-5.5 if it is available; otherwise use the current default/latest GPT model. Write a PTCF prompt (Persona, Task, Context, Format—from Workshop 1). Be specific about the format you want.
- Run it. Then iterate: shorten, change tone, add a section, ask for a table. Push it 2–3 rounds until it’s close to usable.
- Do the final edit yourself. Note what you changed—that’s the part that needs your expertise.
Worked Example: Summarize-Then-Act
No task of your own? Use Workshop2_Demo_Abstracts.docx — three short synthetic abstracts on pedestrian safety. Paste them into ChatGPT Edu with:
“Summarize each abstract as one row of a table: population, method, key finding. Then draft a 3-sentence email to my research team on what these mean for our streetlight and injury analysis.”
Then iterate:
- Round 2: “Cut it to half the length. Make the findings column quantitative — effect sizes, not adjectives.”
- Round 3: “Add a column rating each study’s relevance to our project, 1–3, with a one-phrase justification.”
Verify: check every number in the table against the abstracts — confidence intervals are the easiest thing for AI to mangle.
Two Patterns Worth Knowing
Summarize-then-act. Paste a long document and ask: “Summarize the key points, then draft a 3-sentence email I can send to my team about what they need to do.” One step gets you from raw material to a usable product.
Make-it-a-table. For anything with structure—options, pros and cons, a comparison, a schedule—add: “Put this in a table.” It’s faster to scan and easier to drop into a doc.
Build a Reusable Recipe
Once a prompt works, save it. Strip out the specifics and leave placeholders:
You are a [role]. [Task]. Context: [details]. Format: [structure, length, tone].
Keep these in a running note. Next week you fill in the blanks instead of starting over.
AI can invent facts, citations, and numbers. For anything that goes into a real document or decision, check it against the source. Never paste sensitive or regulated data into any tool unless that tool and use case are approved for it.
Wrap-Up (7 min)
- Microsoft 365 Copilot for work inside Office files; Copilot Chat for protected web chat; ChatGPT Edu for open drafting and reusable assistants.
- The win is the first draft and the summary—you keep the judgment.
- Save the prompts that work as reusable recipes.
Try this before Workshop 3: save two prompt recipes you’d actually reuse. We’ll turn shared recipes into team tools next session.
Looking Ahead
Workshop 3: Collaboration on Projects — moving from solo use to team use: AI for meeting notes and follow-ups, project planning, and the team tools that standardize how a group works—shared Projects in ChatGPT Edu (one workspace for a team’s files, instructions, and chats), custom GPTs, and a shared prompt library.