How to Write Better AI Prompts for Work

Current as of May 6, 2026

AI can draft, summarize, analyze, brainstorm, compare, edit, and automate parts of knowledge work. But the quality of what you get depends heavily on the quality of what you ask. A weak prompt gives the AI too little direction, so it fills in the gaps with assumptions. A strong prompt gives the AI the goal, context, constraints, examples, and output format it needs to produce something useful.

The best workplace prompts are not magic phrases. They are clear work instructions. Think of AI as a capable colleague who is fast, tireless, and knowledgeable, but who does not automatically know your audience, standards, business context, risk tolerance, or definition of “good.” Your job is to make those things explicit.

Recent guidance from OpenAI emphasizes “outcome-first” prompting: define the target result, success criteria, constraints, available context, and desired final answer, then let the model choose an efficient path. OpenAI’s current GPT-5.5 prompt guidance also notes that shorter, outcome-focused prompts often work better than overly process-heavy prompt stacks. [1]

The 80/20 Rule of Better Prompting

Most workplace prompt quality comes from five things:

1. Goal — what result do you want?
2. Context — what background should the AI consider?
3. Audience — who is the output for?
4. Constraints — what must be included, excluded, checked, or avoided?
5. Format — what should the final answer look like?

A practical starter formula:

Act as [role].
Help me [task].
Context: [background, audience, situation, source material].
Success criteria: [what good looks like].
Constraints: [length, tone, facts only, avoid X, cite sources, ask if missing].
Format: [table, memo, email, bullets, markdown, JSON, etc.].

This is similar to several widely used frameworks. Atlassian recommends focusing on persona, task, context, and format. [2] The Journal of Accountancy’s 2026 prompting article recommends task, context, expectations, and output. [3] MIT Sloan’s essentials guide highlights three core strategies: provide context, be specific, and build on the conversation. [4]

1. Start with the Business Outcome

Bad prompt:

Write something about our customer onboarding.

Better prompt:

Create a one-page customer onboarding checklist for new B2B SaaS customers.
The goal is to reduce first-week confusion and help account managers standardize kickoff calls.
Audience: customer success managers.
Include: account setup, access permissions, kickoff meeting, success metrics, training, first 30-day review.
Format: markdown checklist with section headings.
Tone: practical and concise.

The second prompt works because it defines the real business outcome: reduce confusion and standardize onboarding. It does not merely ask for content; it describes the job the content must do.

Prompting principle

Before asking AI to generate anything, finish this sentence:

“The output is successful if it helps me…”

That sentence becomes your success criteria.

2. Give the AI a Role, But Do Not Stop There

Role prompting helps shape the answer. For example:

Act as a senior product marketer.

This is useful, but incomplete. A role alone does not tell the AI what task to perform, what your company sells, who the audience is, or what constraints apply.

Better:

Act as a senior product marketer for a B2B cybersecurity company.
Create three positioning angles for a new endpoint detection feature.
Audience: CISOs at mid-market financial services firms.
Context: they care about audit readiness, alert fatigue, and fast deployment.
Format: table with columns for angle, core message, proof point, and risk of overclaiming.

Atlassian’s guide explains that persona gives the model useful framing, but the prompt still needs the task, context, and format to produce a strong result. [2]

3. Add the Context the AI Cannot Guess

AI is powerful, but it is not a mind reader. It does not automatically know:

• your company’s strategy,
• your customer segment,
• your internal terminology,
• the decision being made,
• the political sensitivity of the topic,
• what has already been tried,
• what “good” looks like in your organization.

Bad prompt:

Summarize this report.

Better prompt:

Summarize this quarterly sales report for the VP of Sales.
Focus on:
1. revenue compared with last quarter,
2. pipeline risks,
3. three largest enterprise deals,
4. actions needed before the board meeting.
Ignore minor regional details unless they affect total revenue.
Format: executive brief, max 400 words.

MIT Sloan’s guide notes that providing context and specificity helps produce more targeted, useful results. [4]

4. Be Specific About the Task

Vague prompts create generic outputs. Specific prompts create usable outputs.

The Journal of Accountancy’s 2026 guidance warns that being too vague or too wordy leads to weaker AI results, while concise specificity improves output quality. [3]

5. Define the Format Before the AI Answers

The right format can save more time than the right wording.

Ask for formats such as:

• executive summary,
• memo,
• email,
• markdown article,
• checklist,
• table,
• comparison matrix,
• decision brief,
• JSON,
• slide outline,
• FAQ,
• step-by-step SOP,
• risk register,
• meeting agenda,
• action plan.

Example:

Compare these three vendors for our procurement team.
Use a table with columns:
Vendor, strengths, weaknesses, implementation risk, security concerns, pricing uncertainty, recommended use case.
End with a 5-sentence recommendation.

OpenAI’s prompt engineering docs recommend being explicit about the output structure, especially for tasks that require consistent or machine-readable responses. [5]

6. State Constraints and “Do Not” Rules

Workplace outputs often fail because the AI includes things you did not want: jargon, risky claims, invented numbers, excessive detail, legal-sounding language, or a tone that does not fit your brand.

Add constraints like:

Do not invent facts.
Do not mention features not listed in the source material.
Do not use hype words such as “revolutionary,” “game-changing,” or “cutting-edge.”
Do not exceed 300 words.
Do not provide legal advice; flag legal questions for counsel.
Use only the provided transcript.
If information is missing, write “Not available in the source.”

This is especially important in regulated or high-stakes work. MIT Sloan warns that AI can produce inaccurate or fabricated outputs, so users should review results critically. [4]

7. Give Examples When Style Matters

If you want the AI to match a style, show the style.

Bad prompt:

Write this in our brand voice.

Better prompt:

Rewrite the announcement in our brand voice.

Brand voice examples:
- “We keep security practical, not scary.”
- “Simple controls beat complicated promises.”
- “Your team should know what to do next, not decode a policy manual.”

Rules:
- short sentences,
- plain English,
- confident but not dramatic,
- no buzzwords.

Text to rewrite:
[paste text]

This is called few-shot prompting: giving examples so the model can infer the pattern. The Journal of Accountancy notes that example-based prompting is useful when you want AI to match your tone, style, or structure. [3]

8. Ask for Options Before Asking for the Final

For creative and strategic tasks, do not ask for one answer too soon. Ask for options first.

Generate 12 possible titles for this article.
Group them into:
- practical,
- bold,
- SEO-friendly,
- executive audience.
Then recommend the best 3 and explain why.

This gives you range, not just a single guess. It is useful for:

• headlines,
• campaign ideas,
• product names,
• presentation angles,
• negotiation scripts,
• customer email variants,
• strategy options.

9. Use AI as a Thinking Partner, Not Just a Writer

Many people use AI only to draft text. That leaves value on the table.

Better workplace uses include:

• identifying missing assumptions,
• finding risks in a plan,
• turning messy notes into decisions,
• creating alternatives,
• comparing tradeoffs,
• preparing meeting questions,
• pressure-testing an argument,
• creating checklists,
• converting expert knowledge into SOPs.

Example:

Review this project plan like a skeptical operations leader.
Find:
1. hidden assumptions,
2. timeline risks,
3. unclear ownership,
4. dependencies that may block delivery,
5. questions I should ask before approving it.
Format as a risk table.

10. Ask the AI What It Needs

When a task is complex, the best prompt may be:

I want you to help me create a board-ready update on our AI adoption program.
Before drafting, ask me up to 7 questions that would materially improve the quality of the update.
Do not ask for information that would only make a small difference.

This prevents you from guessing all the missing context upfront. It also helps the model surface the information it needs to produce better work.

11. Use Step-by-Step Reasoning for Analysis, But Keep the Final Clean

For complex analysis, ask the AI to work carefully and verify its result, but do not always require a long reasoning transcript. A useful pattern:

Analyze the options carefully before answering.
In the final answer, provide:
- recommendation,
- key reasons,
- risks,
- assumptions,
- next steps.
Keep the reasoning concise and decision-oriented.

OpenAI’s current prompt guidance advises focusing on the desired outcome, success criteria, constraints, and final answer rather than over-specifying every internal step. [1]

12. Use the Right Framework for the Job

Different prompt frameworks work better for different use cases.

Persona–Task–Context–Format

Best for everyday workplace tasks.

Persona: You are a senior HR business partner.
Task: Draft a manager email about the new performance review cycle.
Context: Managers are confused about deadlines and calibration meetings.
Format: 200-word email with bullets for key dates.

Task–Context–Expectations–Output

Best for concise, repeatable business prompts.

Task: Analyze customer churn survey responses.
Context: B2B SaaS customers who cancelled in Q1 2026.
Expectations: Identify themes, quote representative comments, separate product issues from pricing issues.
Output: table plus 5-bullet executive summary.

COSTAR

Best for polished professional outputs.

COSTAR stands for:

• Context
• Objective
• Style
• Tone
• Audience
• Response

This framework is useful because it forces you to define what the model should know, what it should achieve, how it should sound, who it is for, and how it should respond. [6]

Example:

Context: We are launching a new expense policy for a 500-person company.
Objective: Explain the policy clearly and reduce employee confusion.
Style: Internal announcement.
Tone: Direct, helpful, not punitive.
Audience: All employees.
Response: Markdown email with subject line, 3 key changes, FAQ, and next steps.

CRISPE

Best for prompts where experimentation and tone matter.

CRISPE stands for:

• Capacity/Role
• Insight
• Statement
• Personality
• Experiment

It encourages you to define the AI’s role, provide background, state the task, set the tone, and request multiple versions or examples. [7]

Example:

Capacity/Role: Act as a customer success communication specialist.
Insight: The customer is frustrated because implementation is two weeks late.
Statement: Draft a status update that acknowledges the delay and rebuilds confidence.
Personality: Calm, accountable, specific.
Experiment: Provide 3 versions: concise, empathetic, and executive-level.

13. Build a Reusable Prompt Library

If a prompt works, save it. Good prompts become reusable work assets.

Create a shared prompt library with:

• use case,
• owner,
• last updated date,
• prompt text,
• example input,
• example output,
• required review step,
• privacy warning,
• known limitations.

The Journal of Accountancy recommends capturing and reusing successful prompts to improve team productivity and standardize workflows. [3]

14. Manage Prompts Like Workflows

For teams, prompt quality is not just about individual skill. It is about repeatable systems.

For important prompts, use:

• peer review,
• version control,
• test cases,
• output scoring,
• change logs,
• approval workflows,
• privacy checks,
• escalation rules,
• source requirements.

Parloa’s 2025 guide argues that prompt frameworks help move prompting from one-off trial and error into a repeatable discipline with shared vocabulary, templates, versioning, evaluation, and review. [6]

15. Match the AI Tool to the Task

Do not assume one AI tool is best for everything. Some tools are stronger for:

• writing,
• coding,
• long-document analysis,
• spreadsheet work,
• research,
• image understanding,
• enterprise search,
• automation,
• customer support workflows.

The Journal of Accountancy’s 2026 article recommends matching the tool to the task and considering privacy and security when choosing AI systems. [3]

16. Protect Confidential and Sensitive Data

A great prompt is not great if it creates risk.

Before pasting information into an AI tool, ask:

• Is this personal data?
• Is this customer confidential information?
• Is this financial, legal, medical, HR, or security-sensitive?
• Is this covered by company policy?
• Does the AI tool retain or train on inputs?
• Should this be anonymized or summarized first?
• Should this be handled only in an approved enterprise AI environment?

Useful prompt constraint:

Use only the anonymized information below.
Do not infer identities.
Do not include personal data in the output.
Flag any missing information instead of guessing.

17. Require Evidence for Factual Work

For factual, legal, financial, technical, or market research tasks, ask for sources and uncertainty.

Research the current requirements for [topic].
Use authoritative sources only.
Cite each factual claim.
Separate:
- confirmed facts,
- interpretations,
- open questions,
- recommendations.
If sources disagree, explain the disagreement.

AI can sound confident even when wrong. MIT Sloan’s guide specifically warns that AI-generated content can be inaccurate, misleading, or fabricated. [4]

18. Use AI to Improve Your Prompt

When you do not know how to ask, ask AI to improve the question.

Improve this prompt for clarity, specificity, and output quality.
Keep it under 150 words.
Add missing context fields as placeholders.
Here is my rough prompt:
[paste rough prompt]

Or:

Turn this task into a high-quality prompt using:
- role,
- task,
- context,
- constraints,
- output format,
- success criteria.
Task:
[paste task]

This turns prompt writing into a collaborative process.

19. Compare Weak vs. Strong Prompts

Example: Email

Weak:

Write a follow-up email.

Strong:

Write a follow-up email to a prospect who attended our product demo last week but has not replied.
Context: They are VP of Operations at a mid-size manufacturing company.
They were interested in inventory management features but said budget approval may be difficult.
Goal: restart the conversation without sounding pushy.
Offer one useful resource.
Tone: brief, helpful, professional.
Length: under 140 words.

Example: Meeting Notes

Weak:

Summarize these notes.

Strong:

Turn these meeting notes into an action-oriented summary for the project team.
Include:
- decisions made,
- open questions,
- action items,
- owner,
- deadline,
- risks.
If owner or deadline is missing, write “not specified.”
Format as a markdown table plus a 5-bullet recap.

Example: Strategy

Weak:

Give me a marketing strategy.

Strong:

Create a 90-day marketing strategy for a B2B SaaS startup selling compliance automation to fintech companies.
Budget: $25,000.
Team: 1 marketer, 1 founder, freelance designer.
Goal: generate 40 qualified sales calls.
Include channels, weekly actions, KPIs, assumptions, and risks.
Prioritize low-cost channels and founder-led content.
Format as a phased plan.

20. Use This Prompt Quality Checklist

Before sending a prompt, check:

• Did I define the goal?
• Did I specify the audience?
• Did I give enough context?
• Did I define the output format?
• Did I include constraints?
• Did I say what to avoid?
• Did I provide examples if style matters?
• Did I ask for sources if facts matter?
• Did I protect confidential data?
• Did I define what “good” means?
• Did I ask for assumptions or missing information when needed?

21. A Universal Workplace Prompt Template

Copy and adapt this template:

Act as [role or expert perspective].

Task:
[Clearly describe what you want done.]

Context:
[Provide background, audience, business situation, source material, constraints, and why this matters.]

Success criteria:
A good answer should:
- [criterion 1]
- [criterion 2]
- [criterion 3]

Constraints:
- [length]
- [tone]
- [things to avoid]
- [source or evidence requirements]
- [privacy or compliance rules]

Output format:
[Specify table, memo, checklist, email, markdown, JSON, bullets, etc.]

Before answering:
If essential information is missing, ask up to [number] clarifying questions.
If you can proceed with reasonable assumptions, state them briefly and continue.

Common Prompting Mistakes

1. Asking for “better” without defining better

Instead of:

Make this better.

Use:

Rewrite this to be clearer, shorter, and more executive-friendly.
Preserve the key message.
Remove jargon.
Keep under 200 words.

2. Providing context after the AI already answered

Give the most important context upfront. Refinement is useful, but missing context wastes time.

3. Asking for too many unrelated tasks at once

Break complex work into stages:

1. analyze,
2. outline,
3. draft,
4. critique,
5. revise.

4. Accepting the first answer

Ask for improvement:

Critique your answer.
What is missing, risky, vague, or unsupported?
Then produce a stronger version.

5. Forgetting verification

For high-stakes work, AI should assist, not replace expert review.

Final Takeaway

Better prompting is not about memorizing secret phrases. It is about communicating work clearly.

The best prompts:

• define the outcome,
• give relevant context,
• specify the audience,
• set constraints,
• show examples,
• request the right format,
• ask for verification when facts matter,
• protect sensitive information,
• improve through iteration.

When you write prompts this way, AI becomes less like a search box and more like a capable work partner: faster, clearer, and more aligned with the result you actually need.

Sources

1. OpenAI API Docs — Prompt guidance
2. Atlassian — The ultimate guide to writing effective AI prompts
3. Journal of Accountancy — 9 tips to write more effective AI prompts
4. MIT Sloan Teaching & Learning Technologies — Effective Prompts for AI: The Essentials
5. OpenAI Help Center — Best practices for prompt engineering with the OpenAI API
6. Parloa — The complete guide to prompt engineering frameworks
7. Denys Dinkevych — CRISPE: ChatGPT Prompt Engineering Framework