Your Quick Summary, Before the Coffee Gets Cold
Prompt engineering is the practice of giving an AI system clear instructions, useful context and a definition of a successful output. You do not need complicated formulas. You need to describe the task well, inspect the response and improve the instructions when necessary.
- A useful prompt defines the task, context, constraints, output format and review criteria.
- Reusable templates are starting points, not magic formulas, and the result still needs verification.
- The twelve examples below cover writing, research, meetings, planning and everyday decisions.
The twelve templates below are starting points. Replace the bracketed sections with your own information and remove any confidential or sensitive material unless your organization has approved the tool and account.
A simple prompt framework
Use five building blocks:
- Context: What situation should the AI understand?
- Task: What should it do?
- Input: What information should it use?
- Constraints: What rules, audience, tone or limits apply?
- Output: What format should it return?
OpenAI recommends clear and specific instructions followed by iterative refinement. Anthropic also emphasizes direct instructions, context and examples. The objective is not to produce one enormous prompt. It is to make the task understandable and testable.
1. Turn rough notes into a clear email
Turn the notes below into a concise email for [audience]. The objective is [objective]. Use a [professional/friendly/direct] tone. Keep it under [length]. Preserve every date, number and commitment exactly. If information is missing, insert [QUESTION] instead of inventing it. Notes: [paste notes].
2. Improve writing without changing the facts
Edit the text below for clarity and flow. Keep the meaning, facts and point of view unchanged. Remove repetition and unnecessary jargon. Show the revised version first, then list the three most important edits you made. Text: [paste text].
3. Summarize a document with traceability
Using only the document below, create a summary for [audience]. Include: the main conclusion, five supporting points, decisions or deadlines, and unresolved questions. Cite the relevant section or page for every important claim. If the document does not answer something, say so. Document: [paste or attach].
4. Prepare for a meeting
I am meeting [person/team] about [topic]. My objective is [objective]. Based on the context below, create a 30-minute agenda, five questions I should ask, three likely objections and the information I should bring. Do not assume facts beyond the context. Context: [details].
5. Create a first presentation outline
Create a [number]-slide narrative for [audience] about [topic]. The audience currently believes [starting point] and should leave ready to [desired action]. For each slide, provide one message, the supporting evidence needed and a visual idea. Do not write full slide copy yet.
6. Compare options using criteria
Compare [option A], [option B] and [option C] for [situation]. Use these criteria: [criteria]. Create a table with strengths, limitations, assumptions and questions to verify. Do not choose a winner until you identify which criteria matter most.
7. Learn a difficult concept
Teach me [concept] at a beginner level. Start with a plain-English explanation and one concrete analogy. Then give me an example, a common misconception and three questions that test whether I understood. Do not reveal the answers until I respond.
8. Critique an idea constructively
Act as a critical reviewer of the proposal below. Identify the strongest element, the three weakest assumptions, evidence that would change the conclusion and one lower-risk experiment. Be specific and do not rewrite the proposal. Proposal: [paste].
9. Build a research plan
Create a research plan for this question: [question]. Break it into subquestions. Prioritize primary and recent sources. Explain what evidence would be considered strong, what could create bias and how to verify conflicting claims. Do not answer the main question yet.
10. Extract structured information
Extract the following fields from the text: [fields]. Return a table with one row per [item]. Use “not stated” when a field is absent. Add a final column containing the exact passage that supports each row. Text: [paste].
11. Plan a personal project
Help me plan [project]. My deadline is [date], available time is [time] and constraints are [constraints]. Break the project into milestones, weekly actions and the smallest useful first step. Flag assumptions and ask up to three questions before finalizing the plan.
12. Verify an AI-generated answer
Review the answer below as a fact-checker. Separate verifiable claims, opinions and recommendations. For each factual claim, state what source would be needed to verify it. Identify unsupported certainty, missing context and possible outdated information. Do not assume the answer is correct. Answer: [paste].
How to improve a weak result
When the first answer is disappointing, diagnose it before starting over:
- Too generic: add your audience, purpose and examples.
- Wrong tone: provide a short sample of the desired voice.
- Missing details: specify required fields or a checklist.
- Too long: set a length and a hierarchy of importance.
- Unsupported claims: restrict the answer to supplied sources and request citations.
- Wrong format: give a simple example of the expected output.
What prompt engineering cannot solve
A better prompt cannot make a model know information it does not have, turn an unsuitable tool into a secure system or replace professional judgment. It also cannot guarantee truth. The quality of the output depends on the model, the available context, the task and the review process.
For medical, legal, financial, employment or safety decisions, use AI to organize questions and information, not as the final authority.
Practice prompting with feedback
Cavern AI teaches prompting as part of its generative AI masterclasses and private sessions in NYC. Participants practice on realistic tasks, compare outputs and learn how to correct weak or unreliable responses.
Frequently asked questions
Do I need a different prompt framework for every AI tool?
No. Clear tasks, context, constraints and examples transfer well. Individual tools may respond differently, so test and adapt.
Are long prompts always better?
No. Include information that changes the answer. Extra instructions can create conflicts and make the task harder to follow.
Should I tell the AI to “act as an expert”?
A role can help establish perspective, but it does not create verified expertise. Source quality and human review still matter.
Can I save prompts for repeated use?
Yes. Save the prompt together with its purpose, allowed inputs, output format and review checklist. Re-test it when the tool or workflow changes.
