A Practical Manual for Crafting High-Impact Prompts
The AI Prompting Handbook
This handbook covers practical prompting principles across six major modalities. A good prompt defines the task, the context, the role of the AI, the intended audience, the constraints, the output format, and the quality standard.
Weak vs. Strong Prompts
Collapse
Treat AI Like a Highly Capable but Context-Limited Specialist
AI models can process complex instructions, synthesize large amounts of information, generate creative options, analyze documents, debug code, interpret images, and simulate expert reasoning. However, they do not automatically know your intent, your business context, your risk tolerance, your preferred style, or your quality standard unless you provide those inputs.
Weak Prompt:
"Write a marketing plan."
This gives the AI no role, no audience, no context, no constraints, and no format. The output will be generic and unusable for real business decisions.
Strong Prompt:
"You are a senior B2B marketing strategist. Create a 90-day marketing plan for a mid-sized medical device company launching a new sleep therapy product in Germany. The target audience is pulmonologists and sleep clinics. Include objectives, target segments, messaging, channel strategy, budget assumptions, KPIs, risks, and a weekly action plan. Use a concise executive style."
The difference is specificity.
The first answer is often a draft. The expert user improves it.
Anatomy of a Strong Prompt
Collapse
A strong prompt contains: Role, Task, Context, Input Material, Output Format, Constraints, and Quality Criteria.
Role & Task
Role: Define the expertise you want the AI to apply.
Examples: "Act as a senior commercial strategist." "Act as a regulatory affairs expert." "Act as a Python code reviewer."
Use roles when expertise, judgment, or style matters.
Task: State exactly what you want done.
Examples: "Summarize the document." "Extract the risks." "Compare the options." "Generate five campaign concepts."
Avoid vague verbs such as "help" or "improve" unless you define what improvement means.
Context & Input
Context: Provide the background the model needs. Examples: "Our company sells SaaS solutions to SMEs." "The audience is non-technical." "This will be used in a board presentation." Context reduces generic output.
Input Material: Provide the source text, image, transcript, data, or reference material. Use delimiters to separate your instructions from the source material.
Example: "Use only the information between the triple quotes."
Output Format & Constraints
Output Format: Tell the model how to structure the answer.
Examples: "Return as a table." "Use bullet points." "Return valid JSON only."
Constraints: Define boundaries.
Examples: "Maximum 300 words." "Do not invent facts." "Use only the provided source material." "Flag assumptions separately."
Quality Criteria: Tell the model what good looks like. Examples: "Prioritize commercial practicality." "Use MECE structure." "Include the strongest counterargument."
The Universal Prompting Formula
Collapse
Be Specific: "Rewrite this proposal to make it more concise, commercially persuasive, and suitable for a CFO. Keep the meaning unchanged. Reduce the length by 30%."
Give Examples: Show the model the pattern you want. Provide 2-3 examples of desired output before asking it to process new input.
Use this formula as a default structure:
Act as [role]. Your task is to [task]. The context is [context]. Use the following input: [input]. Follow these constraints: [constraints]. Return the output in this format: [format]. Before finalizing, check for [quality criteria].
Act as a senior AI automation consultant. Your task is to identify automation opportunities for a 50-person accounting firm. The firm handles invoices, payroll, client reporting, tax preparation, and customer service. Follow these constraints: focus on practical tools available today, avoid highly technical implementation details, and flag any data privacy risks. Return the output as a table with columns for Process, Pain Point, AI Opportunity, Business Value, Complexity, Risk, and First Step. Prioritize ideas that could be piloted within 30 days.
BE SPECIFIC, NOT VERBOSE. GIVE EXAMPLES WHEN QUALITY MATTERS. SEPARATE INSTRUCTIONS FROM SOURCE MATERIAL. ASK FOR ASSUMPTIONS AND RISKS.
Advanced Prompting Patterns
Collapse
Role Prompting
Use when you need domain-specific judgment.
Act as a [specific expert]. Evaluate [topic] for [audience/context]. Prioritize [criteria].
Few-Shot Prompting
Use when the output must follow a specific pattern. Provide 2-3 examples of desired output before asking it to process new input.
Chain-of-Work
Instead of asking the AI to jump to the final answer, ask it to work through stages.
First identify the key issues. Then group them into themes. Then recommend actions. Then list risks.
Critique-and-Improve
Review the following draft. Identify weaknesses in clarity, logic, evidence, structure, and persuasiveness. Then rewrite it.
Red-Team
Act as a skeptical executive. Challenge this proposal. Identify weak assumptions, hidden risks, missing data, implementation barriers, and reasons the initiative might fail.
Constraint-Based
Design a solution under these constraints: budget below €20,000, implementation within 60 days, no new full-time hires, GDPR compliance required.
Structured extraction prompting
Extract the following fields from the text: customer name, company, pain point, urgency, budget signal, decision-maker, next step, and risk. If a field is missing, write ‘Not stated.’
Comparative prompting
Compare Options A, B, and C using the following criteria: cost, speed, complexity, scalability, risk, and strategic fit. Recommend one option and explain why.
Text-to-Text and Vision-to-Text Prompting
Collapse
Text-to-text prompting
Text-to-text prompting is the most common form of AI interaction.
Typical uses: Summarization, Email drafting, Market research synthesis, Strategy development, Data interpretation, Customer feedback analysis, Meeting preparation, Proposal writing, Training content creation, Decision support.
Best Practices: Define the role. State the task clearly. Give context. Provide source material. Specify the output format. Define the audience. Ask for assumptions and risks.
Act as a customer insights analyst. Analyze the following customer feedback. Tasks: 1. Identify recurring themes. 2. Rank themes by frequency and business impact. 3. Extract representative quotes. 4. Identify urgent issues. 5. Recommend three actions. Return the output as a table with columns: Theme, Evidence, Frequency, Business Impact, Recommended Action, Owner, Urgency.
Act as a senior B2B sales copywriter. Rewrite the email below for a managing director of a 75-person manufacturing company. Goals: Make it concise, make the value proposition clearer, reduce hype, add a stronger call to action, keep the tone professional and direct. Return: 1. Revised email 2. Explanation of changes 3. Subject line options.
Vision-to-text prompting
Vision-to-text prompting uses images, screenshots, charts, invoice or document extraction, diagrams, scanned documents, whiteboards, or product photos as input. The AI describes, analyzes, extracts, compares, or interprets the visual content.
Best Practices: Tell the AI what to focus on. Specify the level of detail. Ask it to separate observation from interpretation. For charts, ask for axes, units, trends, outliers, and possible caveats. Do not assume the AI can read every small detail perfectly.
Review this screenshot as a senior UX consultant for a B2B SaaS product. Tasks: 1. Describe the screen objectively. 2. Identify usability issues. 3. Identify confusing labels, layout problems, or missing calls to action. 4. Recommend improvements. 5. Prioritize recommendations by expected user impact. Return a table with columns: Issue, Evidence from Screenshot, User Impact, Recommendation, Priority.
Extract structured information from the attached invoice image. Fields: Vendor name, Invoice number, Invoice date, Due date, Total amount, Currency, Tax amount, Line items, Payment terms, Bank details. Rules: If a field is not visible, write 'Not visible.' If a field is unclear, write 'Unclear' and explain why. Do not guess. Return the result as a table.
Text-to-Image / Text-to-Video Prompting
Collapse
Text-to-Image/ Text-to-Video uses written descriptions to generate visual assets.
Best Practices for Text-to-Image / Video
Describe the subject clearly. Define the environment. Specify style, mood, lighting, camera angle, composition, and format. Mention what should not appear. For commercial work, define brand tone and audience. For video, describe sequence, motion, pacing, camera movement, duration, and transitions. Avoid overloaded prompts with too many competing visual ideas. Use iterations: concept first, then refine.
Create an image of [subject] in [setting], doing [action], in the style of [style]. Use [composition], [lighting], and [mood]. Format: [aspect ratio/resolution]. Avoid [negative elements].
Create a [duration] video showing [subject] in [scene]. The subject [action]. Camera [movement]. Lighting is [lighting]. Mood is [mood]. Style is [style]. Avoid [negative elements].
Create a clean, professional LinkedIn banner image for an AI automation consultant helping SMEs. Visual concept: A modern office workspace with subtle AI network patterns integrated into the background. A business leader reviews a digital workflow dashboard showing automated tasks, customer service, finance, and marketing processes. Style: Premium B2B consulting aesthetic, realistic but slightly polished. Composition: Wide horizontal banner, central focus on the dashboard, enough negative space on the left for text overlay. Mood: Confident, practical, modern, trustworthy. Lighting: Soft daylight, clean office atmosphere. Avoid: Robots, sci-fi clichés, exaggerated neon effects, clutter, unreadable text.
Create a 20-second B2B explainer video concept for an AI automation consultancy serving SMEs. Scene sequence: 1.A small business team looks overwhelmed by emails, spreadsheets, and manual admin work. 2.The workflow transforms into organized automated processes. 3.A dashboard shows saved time, faster response rates, and fewer manual errors. 4.The business owner appears more focused and confident. Style: Modern, clean, realistic business animation. Camera: Slow, smooth transitions. Start with a slightly chaotic wide shot, then move into clean close-ups of automated workflows. Mood: From overwhelmed to controlled and confident. Text overlays: ‘Manual work slows growth.’ ‘AI automation creates capacity.’ ‘Start with one process. Scale what works.’ Avoid: Hype, futuristic robots, fake-looking dashboards, exaggerated claims.
Audio-to-Text Prompting
Collapse
Audio-to-text prompting uses spoken input or recorded audio for transcription, summarization, analysis, intent detection, sentiment analysis, or extraction.
Typical uses: Meeting transcription, sales call analysis, customer service QA, voice note summarization, interview analysis, training feedback, dictation, tone and intent analysis, action item extraction.
Best Practices: State whether you want verbatim transcription or cleaned-up notes. Specify speaker identification if relevant. Ask for timestamps when needed. Define whether filler words should be preserved or removed. For business calls, ask for decisions, objections, risks, commitments, and next steps.
Transcribe this meeting audio and convert it into executive meeting notes. Requirements: 1. Identify speakers where possible. 2. Remove filler words unless they change meaning. 3. Capture decisions, open questions, risks, and action items. 4. Include timestamps for key decisions. 5. Flag unclear audio sections.
Analyze this sales call recording. Tasks: 1. Summarize the customer's situation. 2. Identify explicit needs. 3. Identify implied pain points. 4. Extract objections. 5. Assess buying intent as High, Medium, or Low. 6. Recommend the next best action. Rules: Separate what the customer actually said from your interpretation. Include short evidence quotes.
Convert this spoken voice note into a structured project brief. The output should include: Project objective, Background, Scope, Out of scope, Stakeholders, Deliverables, Timeline, Risks, Immediate next steps. Clean up unclear phrasing, but do not change the meaning. If something is missing, write 'Not specified.'
Transcribe/analyze the audio. Identify speakers if possible. Preserve [level of detail]. Extract [fields]. Flag unclear sections. Return the output as [format].
Audio-to-Audio Prompting
Collapse
Audio-to-audio prompting involves speaking with an AI system and receiving spoken responses. This is common in voice assistants, real-time coaching, language practice, interview preparation, role-play, and hands-free workflows.
Best Practices: Start by defining the interaction style. Tell the AI whether to interrupt, wait, coach, role-play, or summarize. Specify response length. Use turn-taking rules. For coaching, ask for feedback after each response. For role-play, define the persona and difficulty level.
You are a skeptical SME managing director. I am an AI automation consultant trying to sell you a discovery workshop. Role-play rules: Push back on cost, time, data privacy, and unclear ROI. Keep your answers realistic and concise. Do not make it easy for me. After each of my responses, give me a score from 1 to 10 and one improvement suggestion. Continue until I ask to stop. At the end, summarize my strongest and weakest moments.
You are my executive presentation coach. I will rehearse a five-minute pitch for an AI automation consulting service. Your role: Listen without interrupting unless I pause for more than five seconds. After I finish, assess clarity, structure, credibility, commercial relevance, and executive presence. Identify weak claims or vague language. Suggest a stronger opening and closing. Keep feedback direct and practical.
Code-to-Code Prompting
Collapse
Code-to-code prompting uses programming code as input and asks the AI to generate, debug, explain, refactor, optimize, test, or translate code.
Best Practices: Provide the full relevant code, not just the error. Include the programming language and version. Include the expected behavior and actual behavior. Paste the exact error message. Describe the environment. Ask for tests. Do not paste secrets, API keys, passwords, or confidential production data.
Act as a senior Python engineer. Debug the following function ( INSERT FUNCTION) Expected behavior: The function should return the total revenue by customer from a list of orders. Actual behavior: It raises a KeyError when some orders do not have a 'customer_id' field. Tasks: 1. Explain the bug. 2. Provide corrected code. 3. Handle missing customer_id safely. 4. Add two simple test cases. 5. Keep the solution readable for a junior developer.
Act as a senior data analyst. Write a SQL query for PostgreSQL. Goal: Calculate monthly recurring revenue by customer segment. Requirements: Include only active subscriptions. Group by month and customer segment. Return month, segment, total_mrr, and number_of_active_customers. Use clear aliases. Explain any assumptions.
Prompting for Business Consulting
Collapse
For an AI and automation consultant, prompting is not just a productivity skill. It becomes part of your consulting methodology.
You can use prompting to: Diagnose client processes, map workflows, identify automation opportunities, estimate business value, analyze customer feedback, create training materials, draft SOPs, generate implementation plans, build prototypes, prepare workshops, create executive reports, design AI policies, and assess risks.
Act as an AI automation consultant for SMEs. Interview me to identify automation opportunities in my business. Ask one question at a time. Cover: 1. Repetitive admin tasks 2. Customer communication 3. Sales and marketing 4. Finance and invoicing 5. Reporting 6. HR and onboarding 7. Operations 8. Compliance. After the interview, produce: Top 10 automation opportunities, estimated business value, complexity, risk, suggested pilot project, 30-day implementation plan.
Act as a business automation strategist. Prioritize the following AI use cases. Use cases: [Insert use cases] Criteria: Business value Implementation complexity Data availability Risk Time to impact Scalability Stakeholder adoption Return: 1.Prioritization matrix 2.Top three recommended pilots 3.Use cases to avoid for now 4.Key implementation risks 5.First 30-day action plan
Act as an AI governance advisor for a 50-person SME. Create a practical AI usage policy covering: Approved use cases Prohibited use cases Confidential data rules Client data handling Human review requirements Tool approval process Employee responsibilities Escalation path Training requirements The policy should be clear, non-technical, and suitable for employees across sales, operations, finance, and customer service.
Prompting Quality Checklist
Collapse
Use this before sending an important prompt. Diagnose and fix weak outputs systematically.
Clarity
Is the task explicit? Is the audience defined? Is the context included? Is the desired output clear?
Completeness
Have you provided the necessary source material? Have you included relevant constraints? Have you stated what should be excluded? Have you defined success criteria?
Reliability
Have you asked the AI not to invent facts? Have you asked it to flag uncertainty? Have you asked for assumptions?
Structure
Have you specified the format? Have you used headings, tables, JSON, or bullet points where useful?
Critical Thinking
Have you asked for risks? Have you asked for counterarguments? Have you asked what could go wrong? Have you asked what evidence would change the recommendation?
Iteration
Have you planned a second prompt to refine the output? Have you asked the AI to critique its own answer? Have you tested the prompt on more than one example?
Diagnose and fix weak outputs systematically
Too generic
Add role, context, audience, constraints, and examples. Follow-up: "Make this more specific to [industry/company/audience]. Remove generic statements."
Too long
Specify length and format. Follow-up: "Condense this to 300 words. Keep only the most commercially important points."
Inaccurate
Provide source material and restrict the model. Follow-up: "Revise using only the provided source. Remove any unsupported claim."
Lacks judgment
Ask for evaluation criteria. Follow-up: "Evaluate options using cost, speed, complexity, risk, and strategic fit."
Too agreeable
Ask for red-team critique. Follow-up: "Challenge this answer. Identify weak assumptions and missing evidence."
Poorly structured
Specify headings or a table. Follow-up: "Reformat into a table with columns for Issue, Evidence, Impact, Recommendation, Priority."
These checks separate amateur prompting from professional prompting. Apply them consistently to important work.