Even with good prompting habits—and even with frameworks like CARE or RICCE—some tasks are still difficult for GenAI to do well in one shot. This happens when the task requires reasoning and decision-making, not only nice writing. Typical examples at a university are: writing a recommendation for management, comparing tools, proposing a policy, designing an assessment, or preparing a structured argument where accuracy matters.
In these cases, a helpful communication style is CoT “lite”. It is not about long “thinking out loud”. It is a short, practical routine that changes the rhythm of the interaction:
- Plan (3–5 steps)
- Draft (produce the output following the plan)
- Check (quick quality control + what is missing / what to verify)
This approach reduces errors, makes the output more consistent, and saves time because you fix fewer problems later.
Step 1: PLAN (short and concrete)
Ask GenAI to propose a brief plan first. The plan forces structure and prevents the model from jumping to a polished text too early.
Beginner prompt for PLAN:
- “First, give a plan in 3–5 bullets. Only the plan.”
Example (recommendation task):
- “First, create a 4-step plan for recommending a new student voting tool to management. Include: criteria, comparison, risks, and final recommendation. Only the plan.”
Typical plan bullets might look like:
- define goal and constraints
- list evaluation criteria
- compare options against criteria
- recommend one option + explain why + mention what to verify
Step 2: DRAFT (write the actual output)
Now ask GenAI to write the output using the plan. This keeps the result aligned with the intended logic.
Beginner prompt for DRAFT:
- “Now write the draft following the plan above.”
Example (draft request):
- “Now write the recommendation for management based on the plan. Use the structure: Goal → Criteria → Option A vs Option B → Recommendation → Risks/assumptions. Keep it under 250 words.”
Step 3: CHECK (quick quality control)
The final step is a short review. This is where you ask GenAI to act like a checker: verify the requirements, flag missing facts, and highlight uncertainty.
Beginner prompt for CHECK:
- “Now do a quick check: what is missing, what should be verified, and does it meet the constraints?”
Example (check request):
- “Check the draft: (1) does it include criteria and a clear recommendation, (2) does it avoid invented facts, (3) what key info should be verified (price, GDPR, support), (4) suggest one improvement.”
This step is especially useful for preventing “confident nonsense”. A good check section will admit uncertainty and point to what needs confirmation.
CoT “Lite” examples you can reuse
Example 1: Compare two tools (decision support)
“Plan in 4 bullets → Draft comparison → Check for missing data and risks.”
Example 2: Policy draft (GenAI use in a course)
“Plan (sections) → Draft policy → Check for clarity, fairness, and transparency rules.”
Example 3: Assessment redesign (GenAI era)
“Plan assessment strategy → Draft task + rubric → Check if it is resistant to copy-paste answers.”
A ready-to-copy CoT “Lite” template
Use this as a universal prompt pattern:
PLAN: Create a plan in 3–5 bullets for [task].
DRAFT: Write the output following the plan. Use [structure].
CHECK: Verify constraints and list what must be checked or added.