AI for College Teachers: 10 Tasks to Hand Over, 5 to Keep
By Eduvy Editorial · 25 September 2026 · AI & Education
Give AI the first draft of anything you will read, correct and sign yourself: question papers, rubrics, lesson plans, feedback and accreditation write-ups. Keep for yourself anything that judges a student, exposes their personal data, or could harm a patient if wrong. Below: 10 tasks with prompts to copy, 5 to keep, a checking table, and a department policy your HOD can adopt.
What the evidence says about time
The most rigorous test so far is a randomised trial in England, reported in December 2024 by the Education Endowment Foundation. 259 science teachers in 68 secondary schools used ChatGPT with a short guide and cut weekly lesson planning from 81.5 to 56.2 minutes, about 31%, with no drop in quality judged by blinded experts. They were school teachers, and we found no Indian college study, so our checking table has no "time saved" column.
Three rules before you start
- Prefer an institutional account. OpenAI may use chats from personal ChatGPT accounts for training unless you opt out; its Business, Enterprise and Edu plans are not used for training by default. Google says some Gemini chats are read by human reviewers and asks users not to enter confidential information.
- No student names, USNs, marks or photos in any prompt. Use placeholders or anonymised IDs.
- You are the author. Read every line before it reaches a student or a file.
10 tasks to hand over this semester
1. Question paper drafts mapped to COs and Bloom's levels
AICTE's Examination Reform Policy (2018) recommends questions aligned to course outcomes and the revised Bloom's taxonomy, and gives "no more than 40% weightage for knowledge-oriented questions" as an example cap.
Draft a [50]-mark internal test for [course name and code], [semester], [university and scheme]. Course outcomes: CO1 [paste] to CO5 [paste]. Syllabus for this test: [paste]. Write [5] questions, each showing marks, the CO assessed and the Bloom's level (L1 Remember to L6 Create). Rules: no more than 40% of marks at L1 and L2; at least one question at L4 or above; every CO covered. End with a table of marks by CO and by level. No answers.
2. A first-draft CO-PO mapping
NBA's Self Assessment Report format rates CO-PO correlation as 1 (slight), 2 (moderate) or 3 (substantial).
Here are the course outcomes for [course]: [paste]. Here are our POs and PSOs, word for word from our NBA documents: [paste]. Propose a CO-PO and CO-PSO matrix using 1, 2, 3 or blank. Justify every filled cell in one line, linked to the CO's action verb. Flag any CO mapped to more than [six] POs, and any PO no CO addresses.
3. Session plans
Write a one-page plan for a [55]-minute class on [topic] for [second-year B.Pharm] students who know [prior topics]. Include the outcome and its CO, an opening question, the explanation sequence with board work, a ten-minute pair activity, three exit questions at different Bloom's levels, and homework. Use the notation in [prescribed textbook].
4. Rubrics
The same AICTE policy describes a rubric as criteria, descriptors and a scale of performance levels.
Build an analytic rubric for [a mini-project report / lab record / case presentation] worth [20] marks: 4 to 5 criteria, four levels (Excellent, Good, Satisfactory, Needs improvement) with marks for each. Every cell must describe something visible in the work; no phrases like "good understanding". Map each criterion to [CO numbers].
5. Case studies for nursing and pharmacy
Write a fictional case study for [third-year B.Sc Nursing, Medical-Surgical Nursing] on [an adult with type 2 diabetes admitted with a foot ulcer]. Invent every detail. Include history, vital signs, lab values and current medicines by name only, no doses. Add five discussion questions on assessment, nursing diagnosis, priorities, patient education and one ethical issue. Tag every clinical value [VERIFY].
For pharmacy, ask for a prescription to audit instead. Add doses yourself from a verified source.
6. Feedback on student drafts
Below is an anonymised student draft [paste, name and USN removed] and the rubric [paste]. Do not give a mark. For each criterion, quote one sentence that works and one that needs work, then give two specific suggestions. Address the student as "you", in plain English, under 200 words.
7. Simpler lab manuals
Rewrite this lab procedure for first-year students: [paste]. Keep every step, value, safety precaution and their order exactly; simplify only the language. Numbered steps, one action each, under 20 words. Add an apparatus checklist and three common mistakes. List anything ambiguous instead of guessing.
8. Kannada and English explanations
NEP 2020 says more higher education institutions will teach in the mother tongue or local language, or bilingually. AICTE runs its own AI translation platform, Anuvadini.
Explain [how a transformer works] for first-year students from Kannada-medium schools. Use simple, everyday Kannada, with technical terms in English in brackets the first time. Then give a three-line English summary with the same terms, and one example from daily life in Bengaluru.
Have a Kannada-medium colleague read it first: machine-translated technical terms can be wrong.
9. Attendance and marks emails, without the data
AI writes the template; your ERP or mail merge fills in the numbers, so no student data reaches the chatbot.
Draft a short email to parents of first-year [department] students explaining [the attendance rule, e.g. VTU's 2022 regulations require 85% attendance in each course] and [the eligibility rule, e.g. at least 40% of CIE marks to sit the SEE], what happens if a student falls short, and whom to contact. Use placeholders [Student name], [Course], [Attendance %], [CIE marks]. Factual, respectful, no threats. English and Kannada versions.
10. First drafts of NAAC write-ups
NAAC is moving from its seven-criteria framework to Binary Accreditation with Maturity-Based Graded Levels, so paste the exact metric wording from the manual you are assessed under.
Draft the response to NAAC metric [number and exact wording] using only these verified facts: [bullets with numbers, dates and source documents]. [300] words. Add no number, event or claim not in my list. Where something is missing, write [MISSING: what is needed]. End with the evidence document for each claim.
5 things not to hand to AI
1. Final marks and grades
A mark is a decision you must defend in a revaluation. Once in force, section 8(3) of the Digital Personal Data Protection Act, 2023 requires personal data used for decisions about a person to be complete, accurate and consistent. AI can draft comments; you decide the mark.
2. Student personal data in public tools
The DPDP Rules, 2025 were notified on 14 November 2025; most duties take effect on 13 May 2027. Your institution will answer for processing done on its behalf, may use an outside processor only under a valid contract (section 8(2)), and must take reasonable security safeguards (section 8(5)). Failing those safeguards can cost up to ₹250 crore; breaching duties on children's data (under 18, which includes some first-years) up to ₹200 crore. A class list pasted into a personal chatbot goes to a third party with no contract.
3. Misconduct verdicts from "AI detectors"
- OpenAI withdrew its own AI-text classifier on 20 July 2023 for low accuracy: it caught 26% of AI-written text and wrongly flagged 9% of human writing.
- A 2023 study in Patterns ran seven detectors on 91 TOEFL essays by non-native English writers. They wrongly labelled 61% as AI-generated on average, and 89 of the 91 were flagged by at least one. Many Indian students write in a second language.
- Turnitin says its AI score should not be the sole basis for adverse action against a student.
Treat a score as a reason to talk, not as proof: hold a short viva and look at drafts and version history. The UGC 2018 regulations require plagiarism panels to follow natural justice; do the same for AI cases.
4. Live exam papers in public tools
Given those data practices, a paper pasted into a personal chatbot has left your control. Draft from the syllabus, finalise offline, and never paste a live paper, answer key or scheme of evaluation into an external tool.
5. Unchecked clinical and drug-dose facts
WHO's January 2024 guidance on large multi-modal models in health notes "documented risks of producing false, inaccurate, biased, or incomplete statements" and warns of "automation bias", where users miss errors they would otherwise catch. Check every dose, interaction and contraindication against the prescribed textbook, the current National Formulary of India or the product's prescribing information.
How to check the output
| Task | Main risk | How to check |
|---|---|---|
| Question paper | Out-of-syllabus questions; wrong Bloom's labels | Solve every question in the time allowed; a colleague re-labels levels blind |
| CO-PO mapping | Everything marked 3; PO wording altered | Compare POs word for word with your NBA file; justify every 3 |
| Session plan | Unrealistic timing; different notation | Time it against your last class; match the textbook's symbols |
| Rubric | Vague or overlapping levels | Two colleagues mark three old scripts with it; rewrite rows where they disagree |
| Case study | Wrong clinical values | Check every [VERIFY] tag; a clinical colleague reads it |
| Feedback | Generic comments; invented quotes | Confirm each quote is in the draft; edit the tone |
| Lab manual | A step or safety note dropped | Compare line by line with the original |
| Kannada explanation | Wrong technical terms | Kannada-medium colleague reads it |
| Parent email | Rule misquoted; harsh tone | Check against the current regulation; send yourself a test |
| NAAC draft | Invented numbers | Trace every number to a record |
A department AI policy you can adopt
Fill in the brackets and add it to the department handbook.
- Scope. Applies to all faculty, staff and students of the Department of [name] from [date]; reviewed every semester.
- Approved tools. [List.] Personal accounts only for tasks with no student or institutional data.
- Personal data. No student names, USNs, marks, attendance, photos, health or contact details in any AI tool without an institutional contract with the provider.
- Examinations. No live question paper, answer key or scheme of evaluation in any external AI tool.
- Authorship. Faculty answer for the accuracy of AI-assisted material. Clinical and drug information is checked against [named sources].
- Assessment. Faculty decide marks. AI may draft feedback against a published rubric.
- Student use. Each assessment is labelled: AI not allowed; allowed with disclosure; or allowed freely. Unlabelled means AI for learning only, not for submitted text.
- Disclosure. Students state the tool, date, purpose and what they checked.
- Suspected misuse. Detector scores are never the sole evidence. [Committee] hears the student, reviews drafts and may hold a viva, under university malpractice rules.
- Training and review. One session per semester for faculty and for students; the HOD logs incidents and updates this policy each semester.
IIT Delhi's guidelines take a similar line, including course policies that do not penalise students for disclosing AI use.
Related reading: How to Use AI to Study in College Without Hurting Your Marks · Which Careers Will AI Change? Skills to Add During Your Degree
Frequently asked questions
Has UGC or AICTE issued rules on teachers using AI?
We found none specific to generative AI as of September 2026, so university and institution rules apply. For publications, the UGC 2018 plagiarism regulations remain in force: similarity above 40% can cost an annual increment and two years of supervising new research students.
Can I use ChatGPT to set question papers?
For drafts built from the syllabus, yes, if your institution permits it. Solve every question yourself and finalise offline.
Are AI detectors accurate enough to penalise a student?
No. OpenAI withdrew its own detector for low accuracy, detectors wrongly flag non-native English writers at high rates, and Turnitin says its score should not be the sole basis for action.
Can I paste student marks into ChatGPT?
Don't. Under the DPDP Act your institution answers for how student data is processed, including by outside services. Most duties apply from 13 May 2027, but there is no reason to wait. Anonymise, or use a contracted tool.
Task 9 is easiest when attendance and marks live in one system. Eduvy's college ERP keeps admissions, students, attendance, fees and hostel on one data layer. See what Eduvy offers colleges.
Sources
- EEF, ChatGPT in lesson preparation trial (educationendowmentfoundation.org.uk)
- NFER evaluation report, 12 Dec 2024 (nfer.ac.uk)
- OpenAI, how data is used to improve models (help.openai.com)
- Gemini Apps Privacy Hub (support.google.com)
- AICTE Examination Reform Policy, November 2018 (aicte-india.org)
- NBA Self Assessment Report format, CO-PO correlation levels (nbaind.org)
- UGC, Salient features of NEP 2020: higher education (ugc.gov.in)
- AICTE Anuvadini AI translation platform (anuvadini.aicte.gov.in)
- VTU B.E./B.Tech Regulations 2022, attendance and CIE eligibility (vtu.ac.in)
- NAAC notices on Binary Accreditation and MBGL (naac.gov.in)
- Digital Personal Data Protection Act, 2023, full text (meity.gov.in)
- PIB, DPDP Rules 2025 notified, penalties and phased compliance (static.pib.gov.in)
- OpenAI, AI classifier withdrawn 20 July 2023 (openai.com)
- GPT detectors are biased against non-native English writers, published in Patterns, 2023 (arxiv.org)
- Turnitin guidance on its AI writing detection model (guides.turnitin.com)
- UGC plagiarism regulations 2018, natural justice and faculty penalties (ugc.gov.in)
- WHO guidance on large multi-modal models, 18 Jan 2024 (who.int)
- Indian Pharmacopoeia Commission, National Formulary of India (ipc.gov.in)
- IIT Delhi, On the use of Generative AI at IIT Delhi (academics.iitd.ac.in)