The Visual Prompting Playbook
A practical, colour-coded, interactive handbook for using generative AI more effectively at work. Learn the CRAFT framework, core prompting techniques, text prompting, image prompting, risks, hygiene, and reusable templates.
Tip: use this file as index.html for GitHub Pages, Netlify, Vercel, or any static website host.
CRAFT: your universal prompting checklist
Use CRAFT whenever you want clearer, more reliable outputs. Each colour becomes a visual anchor throughout this handbook.
C — Context
Give the AI background, role, situation, constraints, and purpose.
R — Request
State the exact action: create, summarise, classify, compare, rewrite, analyse.
A — Assets
Add the raw material: notes, documents, examples, emails, tables, screenshots.
F — Format
Specify the structure: table, bullets, JSON, email, slide outline, checklist.
T — Target Audience
Define who will read or use the answer and their level of expertise.
Foundations of prompting
Prompting is the process of communicating with a generative AI system using natural language. The prompt is the interface between your intent and the model’s output.
Prompt design
User-facing practice of writing clear, structured instructions that minimise ambiguity.
Prompt engineering
More technical optimisation using tools, APIs, parameters, retrieval, or system design.
Prompt hygiene
The habit of checking assumptions, verifying outputs, protecting data, and iterating.
Anatomy of a great prompt
A strong prompt usually contains instruction, context, input data, and output format. CRAFT expands this into a practical business checklist.
Weak prompt
Write me a social media post.
This is vague: no audience, purpose, tone, source material, or output requirements.
CRAFT prompt
You are a B2B marketing manager. Create a LinkedIn post announcing our AI Prompting Course. Use the notes below as input. Audience: senior business professionals with limited technical knowledge. Format: under 120 words, confident but accessible tone, clear CTA.
| Prompt element | Question to ask | Example |
|---|---|---|
| Instruction | What should the AI do? | Create, summarise, classify, analyse, rewrite. |
| Context | What does the AI need to know? | Role, situation, goal, background. |
| Input data | What should the AI use? | Email thread, transcript, notes, table, uploaded file. |
| Output format | How should the answer be presented? | Table, checklist, executive summary, markdown, JSON. |
Core prompting techniques
Use these patterns when basic prompting is not enough. Each technique solves a different quality problem.
1. Zero-shot prompting
No examples are provided. Best for simple requests, brainstorming, or quick classification.
Classify the following customer email as one of: - Order Inquiry - Technical Problem - Complaint Return only the category and one sentence explaining why.
2. One-shot prompting
Provide one example so the model understands the desired style or structure.
Rewrite this update in the same style as the example. Example: “Project Atlas is on track, with two risks requiring leadership attention.” Now rewrite: [paste update]
3. Few-shot prompting
Provide several examples. Best for repeatable formatting, tone consistency, and specialist outputs.
4. Chain-of-Thought style prompting
Ask for structured reasoning or a step-by-step explanation. For sensitive or complex tasks, ask for a concise rationale rather than hidden reasoning.
Analyse the decision using a structured approach. 1. Identify the decision criteria. 2. Compare the options. 3. Provide a concise recommendation with rationale.
5. Self-consistency
Generate multiple candidate answers and compare them. Useful when accuracy and robustness matter.
Generate three possible answers independently. Compare them for strengths and weaknesses. Then provide the most reliable final answer.
6. Tree-of-Thought
Ask multiple expert personas to evaluate the problem from different angles, then synthesize.
Imagine three experts are discussing this decision: - CFO - Marketing Strategist - Operations Lead Let each identify risks and opportunities. Then provide a joint recommendation.
7. Prompt chaining
Break a large task into smaller sequential steps. Each output becomes the input for the next prompt.
Text prompting mastery
Text prompting works best when you define the role, audience, purpose, constraints, and output structure.
Assign a useful role
Instead of “be a lawyer,” give a detailed role with task, domain, and audience.
Define the tone
Professional, concise, persuasive, supportive, executive-ready, plain English.
Name the audience
Adjust complexity for CEO, customer, beginner, technical team, or student.
Use Markdown
Headings, bullets, dividers, and labelled sections reduce ambiguity in long prompts.
Ask for a preview
Generate the first 10 rows, first paragraph, or first section before scaling output.
Save proven prompts
Create a reusable prompt database for tasks that recur in your workflow.
# Role You are a [role]. # Objective Your task is to [request]. # Context [background] # Assets [paste data, notes, examples] # Output Format [table / bullets / email / report] # Target Audience [who this is for] # Quality Check Before finalising, challenge your answer and identify anything important that may be missing.
Image prompting mastery
Image prompts need visual specificity. Describe what should appear, where it appears, how it should look, and how it should be composed.
Subject
The main person, product, animal, object, scene, or concept.
Environment
Office, forest, beach, city, studio, classroom, futuristic lab.
Style
Photorealistic, doodle, art deco, editorial, cinematic, infographic.
Lighting
Soft natural, studio, golden hour, neon, dramatic, high contrast.
Camera
Wide angle, close-up, 85mm portrait, overhead, macro, drone view.
Composition
Centred, rule of thirds, symmetrical, negative space, poster layout.
Create a [subject] in [environment]. Style: [style] Lighting: [lighting] Camera: [camera angle / lens] Composition: [framing] Mood: [mood] Colour palette: [palette] Aspect ratio: [ratio] Avoid: [things to exclude]
Prompt risks and responsible AI
Prompting is powerful, but not risk-free. Treat AI outputs as drafts, not unquestionable truth.
Prompt injection
Malicious instructions attempt to manipulate the model or connected tools into unintended behaviour.
Prompt leaking
Attempts to reveal hidden system prompts, internal policies, private instructions, or confidential context.
Jailbreaking
Attempts to bypass safeguards and make a model produce content it should refuse or handle safely.
Hallucinations
AI can produce incorrect facts, fake citations, invented details, or confident but unreliable claims.
| Risk | Practical control |
|---|---|
| Sensitive data exposure | Do not paste confidential, personal, customer, or regulated information unless the tool is approved for that use. |
| Outdated information | Verify facts that may have changed, especially laws, prices, policies, product specs, and news. |
| Overreliance | Use AI for drafts and options; keep human review for judgement, compliance, and accountability. |
| Poor prompt hygiene | Ask the model to check assumptions, missing context, and possible inaccuracies. |
Prompt hygiene checklist
Before accepting an output, run a quality check. This reduces rework, hallucinations, and incomplete answers.
Before you prompt
- What outcome do I need?
- What source material should the AI use?
- Who is the audience?
- What format do I need?
- What should be excluded?
Before you accept
- What might be inaccurate?
- What assumptions were made?
- What context is missing?
- Are sources or facts verifiable?
- Does this meet the real objective?
Quality phrase
Add this to important prompts: “Challenge your answer before finalising. Identify anything important that may be missing, inaccurate, or based on assumptions.”
Prompt template library
Copy, adapt, and save these templates in Notion, your knowledge base, or a shared team prompt library.
You are a management consultant. Summarise the attached document for an executive leadership audience. Include: - Key findings - Risks - Recommendations - Decisions required Format: One-page executive summary with clear headings.
You are a strategic account director. Review the notes below and create a meeting preparation brief. Include: - Customer context - Key issues - Opportunities - Risks - Suggested questions - Next actions Format: Table.
I want to achieve the following outcome: [goal] Ask me up to five clarifying questions. Then rewrite my prompt using the CRAFT framework.
Create an image of: [subject] Environment: [environment] Style: [style] Lighting: [lighting] Camera: [camera] Composition: [composition] Mood: [mood] Aspect ratio: [ratio] Avoid: [exclusions]
Prompting in one page
CRAFT
🔵 Context
🟢 Request
🟠 Assets
🟣 Format
🔴 Target Audience
Techniques
Zero-shot, one-shot, few-shot, structured reasoning, self-consistency, Tree-of-Thought, prompt chaining.
Text tips
Assign role, define audience, use Markdown, request previews, save reusable prompts.
Image formula
Subject + Environment + Style + Lighting + Camera + Composition.
Risks
Prompt injection, leaking, jailbreaking, hallucinations, outdated information.
Final check
What is missing? What is assumed? What could be wrong? Is the output fit for purpose?
Final takeaway
The best AI users are not the people who know the most about AI. They are the people who communicate most clearly with it.