AI in Mentoring & Advising

Moderate · Governance, Ethics & Equity track · ~35 min hands-on + readings and quiz

← Back to the Governance, Ethics & Equity track

Where AI can help with advising tasks — and where the human relationship is the whole point.

What you’ll be able to do

  • Identify which advising tasks benefit from AI assistance and which don’t
  • Use AI to draft recommendation letters, advising notes, and resource compilations while keeping student data safe
  • Set boundaries on AI use in mentoring conversations and sensitive advising situations

Overview

Mentoring and advising are relationship-based. An AI can draft a recommendation letter, summarize a student’s milestones, or compile resources for a career question — but it cannot replace the trust, judgment, and personal knowledge that make advising work.

The data rules matter here more than in most topics: student records are FERPA-protected, advising conversations are often sensitive, and recommendation letters are confidential. Never paste identifiable student information into any tool unless it’s approved for that classification.

Use AI for the administrative parts of advising (drafting, formatting, compiling) and keep the relational parts (listening, judging readiness, delivering difficult feedback) human.

Practice activities

Activity 1 · Moderate — Draft a recommendation letter with guardrails

Time ~18 min · Tools ChatGPT Edu

Goal. Use AI to draft a recommendation letter while protecting student privacy and keeping your voice authentic.

Setup. Use this fabricated student profile (do not use a real student’s information):

Student: Jordan Rivera (fabricated), 2nd-year MPH in Epidemiology. Applying for a CDC/CSTE Applied Epidemiology Fellowship. Strengths: strong quantitative skills (completed advanced biostatistics, proficient in R and SAS), led a community-based participatory research project on lead exposure in North Philadelphia, excellent at communicating with community partners. Areas of growth: still developing confidence presenting to large audiences. You have supervised Jordan for 14 months as their practicum advisor. Jordan analyzed 3 years of childhood blood lead level data for the Philadelphia Department of Public Health and produced a policy brief that was shared with City Council.

Steps.

  1. Draft the letter:

    Using this student profile, draft a recommendation letter for the CDC/CSTE Applied Epi Fellowship. Tone: warm but professional. Length: ~400 words. Include specific examples from the profile. Do not invent accomplishments not listed here.

  2. Review the draft critically:

    • Does it sound like you, or does it sound like generic AI?
    • Did it invent any accomplishments not in the profile?
    • Would this letter distinguish Jordan from other applicants?

  3. Rewrite the opening paragraph in your own voice — this is the part that most needs to sound like a person who actually knows the student.

Expected result. A usable first draft that saves time on structure and phrasing, but requires your edits to sound authentic and to add the personal observations only you can provide.

Check your work. Read the letter aloud. If it could be about any student, it’s too generic. The specific details (lead exposure project, City Council brief, community partner skills) should be front and center.

Common pitfalls. AI-drafted recommendation letters tend toward inflation (“exceptional,” “outstanding,” “one of the best”) and sameness. Tone them down and add the specific, personal details that only a mentor can provide. Never paste real student records — use fabricated profiles or de-identified details.

Stretch (optional). Draft a second letter for a different opportunity (e.g., a PhD program) and notice how the emphasis should shift.

Activity 2 · Moderate — Sort advising tasks: AI-appropriate vs. human-only

Time ~15 min · Tools ChatGPT Edu

Goal. Draw a clear line between advising tasks where AI helps and situations where it has no place.

Setup. Consider these 10 advising scenarios:

  1. Compiling a list of fellowship and job opportunities for a graduating student
  2. Drafting talking points for a meeting with a student about their academic performance
  3. A student discloses a mental health crisis during an advising meeting
  4. Summarizing a student’s coursework and practicum history for an annual review
  5. Helping a student brainstorm dissertation topic ideas
  6. Writing a student’s individualized development plan (IDP)
  7. A student asks whether they should report a lab safety concern
  8. Formatting a student’s CV for a specific opportunity
  9. Deciding whether a student is ready to defend their thesis
  10. A student asks you to help them think through a career decision

Steps.

  1. For each scenario, classify it: AI-appropriate (AI can do the bulk of the work), AI-assisted (AI helps with drafting/research but human judgment is central), or Human-only (AI has no place here).

  2. Check your classifications:

    Here are 10 advising scenarios and my classifications. Do you agree, and are there any I’ve classified as AI-appropriate that actually need more human involvement? [your list]

  3. For each “human-only” scenario, write one sentence explaining why AI involvement would be inappropriate or harmful.

Expected result. A classified list that distinguishes administrative tasks (AI-appropriate), judgment-heavy tasks (AI-assisted), and relationship/crisis tasks (human-only). Scenarios 3, 7, 9, and 10 should likely be human-only or heavily human-led.

Check your work. Would you be comfortable telling the student that AI helped with each task you marked as AI-appropriate? If not, reconsider.

Common pitfalls. The temptation is to use AI for everything because it’s faster. But advising a student through a crisis, evaluating readiness, or navigating an ethical dilemma requires presence, judgment, and trust that AI cannot provide.

Stretch (optional). Draft a one-paragraph “AI in advising” norm for your department that distinguishes permissible from impermissible uses.

Activity 3 · Moderate — Catch the sameness problem across multiple letters with Claude Cowork

Time ~25 min · Tools Claude Cowork (personal/consumer account — not Drexel-supported)

Goal. Use Cowork’s ability to work across a whole folder of files to draft several recommendation letters at once — then make the AI expose the “sameness” problem that a one-letter-at-a-time workflow (like Activity 1) can never reveal.

Data classification warning. Claude and Claude Cowork are consumer/PI-purchased tools, not Drexel-supported, and are approved for Low Risk Data only. Student records are FERPA-protected and must never touch this tool. This activity uses only the fabricated profiles below — never substitute a real student, and never connect Cowork to a folder containing real advising files.

Setup. Create a new folder on your computer (e.g., rec-letter-practice) containing three text files, one fabricated student profile per file (do not use any real student’s information):

File 1 — jordan-rivera.txt. Student: Jordan Rivera (fabricated), 2nd-year MPH in Epidemiology. Applying for a CDC/CSTE Applied Epidemiology Fellowship. Strengths: strong quantitative skills (completed advanced biostatistics, proficient in R and SAS), led a community-based participatory research project on lead exposure in North Philadelphia, excellent at communicating with community partners. Areas of growth: still developing confidence presenting to large audiences. You have supervised Jordan for 14 months as their practicum advisor. Jordan analyzed 3 years of childhood blood lead level data for the Philadelphia Department of Public Health and produced a policy brief that was shared with City Council.

File 2 — sam-okafor.txt. Student: Sam Okafor (fabricated), graduating MPH in Community Health and Prevention. Applying to a PhD program in public health with a qualitative research focus. Strengths: exceptional qualitative skills (designed and conducted 30 semi-structured interviews with immigrant caregivers about vaccine hesitancy, coded them in NVivo, co-authored a manuscript under review), thoughtful writer, generous peer mentor in the department’s writing group. Areas of growth: sometimes over-scopes projects and needs help narrowing research questions. You have taught Sam in two seminars and supervised their thesis for a year.

File 3 — priya-shah.txt. Student: Priya Shah (fabricated), 1st-year MPH in Environmental and Occupational Health. Applying for a summer internship with a state health department’s environmental epidemiology unit. Strengths: strong geospatial skills (built interactive GIS maps of heat-island exposure and asthma ED visits for a class project, proficient in QGIS and ArcGIS), fast learner, reliable under deadline. Areas of growth: newer to formal coursework in statistics; early in professional development. You have known Priya for 6 months as her academic advisor and course instructor.

Steps.

  1. Open Claude Cowork and connect it to your rec-letter-practice folder. Ask it to draft all three letters in one pass:

Read each of the three student profile files in this folder. For each student, draft a separate recommendation letter (~400 words, warm but professional) saved as a new file, tailored to that student’s specific target opportunity. Use only the examples in each profile — do not invent accomplishments. Each letter should be distinctly about that student.

  1. Now make Cowork audit its own work for sameness:

Compare the three letters you just drafted against each other. List every sentence, phrase, or structural pattern that repeats nearly verbatim across letters for different students — including identical openings, identical closing endorsements, and reused adjectives. Present the repeats side by side so I can see them.

  1. Read the flagged repeats. For each one, decide: is this harmless letter convention, or interchangeable boilerplate that would make a reviewer reading all three letters discount them?

  2. Rewrite the flagged boilerplate yourself, per student — as in Activity 1, in your own voice, drawing on a specific observation only you could make about that student (Jordan’s City Council brief, Sam’s interview coding, Priya’s heat-island maps).

  3. Delete the practice folder when finished, and disconnect Cowork from it.

Expected result. Three personalized letter drafts plus a concrete list of repeated language across them — direct, checkable evidence of the “sameness” risk that reviewing one letter at a time cannot surface.

Check your work. Swap the student names across the three letters and reread. Any paragraph that still works under the wrong name is boilerplate you need to rewrite.

Common pitfalls. Fellowship and admissions reviewers often read multiple letters from the same recommender — identical phrasing quietly devalues every student it describes. Never connect Cowork (or any unapproved tool) to folders containing real student records; Low Risk Data only.

Stretch (optional). Ask Cowork to produce a one-page comparison table of which concrete, student-specific evidence appears in each letter — any letter whose column is mostly adjectives rather than evidence needs another pass from you.

Check your readiness

Answer these, then check — your score suggests whether to dive in or skim the readings first.

Useful resources