What is Prompting?
A prompt is the instruction you give a generative AI model. The quality of the AI's output depends almost entirely on the quality of your prompt. Prompting is a learnable skill — the more precise, structured, and contextual your instructions, the better the results.
The quality of your very first prompt sets the tone for the entire conversation. A well-structured opening gives AI the role, the goal, and the constraints it needs from the start — saving you time, avoiding confusion, and producing a far more useful response. A poor first prompt leads to a poor first answer, which requires multiple corrections to fix. Get it right first time.
What is Generative AI?
Generative AI creates text, images, code, and more from patterns learned during training on vast datasets. It can answer questions, write content, summarise, translate, and analyse — at speed.
The Biggest Limitation
The biggest limitation of generative AI is not the technology itself, but the creativity and prompt engineering skills of the people who use it. A powerful model with a poor prompt gives a poor result.
Why Prompting Matters
AI cannot read your mind. Without context, role, format, and goal, it guesses what you need — and often guesses wrong. A well-structured prompt unlocks dramatically better outputs every time.
Advantages of Generative AI
Highly proficient in many languages · Can perform complex computations in seconds · Highly customisable · Fast, efficient, cost-effective, and consistent · Can help with writing emails, summarising data, sorting information, generating content, and brainstorming ideas · Constantly updating and learning.
Limitations to keep in mind
Prone to hallucinations — may produce incorrect or nonsensical answers · May produce unethical or biased information · Can be repetitive · Has a knowledge cutoff date (no real-time data unless it has browsing access) · Does not know your personal context unless you provide it.
ChatGPT (OpenAI)
Good for content creation, creative tasks, and versatile everyday prompting. Casual, friendly, and conversational. Adapts easily to different tones.
Gemini (Google)
More formal and factual. Enhanced reasoning and coding abilities. Best for researchers and professionals needing precise, informative responses.
Claude (Anthropic)
Recommended for its ethical training and safety focus. Direct, structured responses suitable for complex tasks. Strong on analysis and nuanced writing.
Prompt Frameworks
Frameworks give your prompts reliable structure. Each acronym is a checklist ensuring you have included the key ingredients. Choose the one that best fits your task — they can also be combined.
Assign a Role so the AI adopts the right perspective. Define the Task precisely. Specify the Format — table, bullet list, email, storyboard, report, etc.
Define the Task objectively. State the Action you want AI to take. Clarify the Goal — what success looks like in concrete terms.
Describe the current problem (Before). State the ideal outcome (After). Ask AI to provide the Bridge — the plan, steps, or solution that gets you there.
Give Context about your situation. Describe the Action you need. Clarify the desired Result. Provide an Example to model from.
Assign a Role. Provide the relevant Input data. Ask for specific Steps. State the Expectation — the success metric or deliverable.
INTERVIEW
Instead of front-loading all information yourself, ask AI to extract it through structured questions. It then summarises your responses and creates a plan. Use SHIFT+ENTER to go to a new line without sending.
Context · Goal · Instruction
The C·G·I framework is the foundation of effective prompting. Use it for every prompt you write, then layer in a specific framework (R-T-F, T-A-G, etc.) when you need more structure. It covers what the AI needs to know, what you want it to produce, and how to get it there.
Full example — Financial context
Full example — Library / Education context
Full example — EFL lesson planning context
Follow-up instructions — refining the response
Instruction Words for Prompting
The instruction word you choose at the start of your prompt determines the type of response you get. Each word signals a different cognitive task to the AI. Below you will find the full list, followed by the most powerful combinations — two instruction words used together for more precise, professional results.
Two instruction words used together in the same prompt are far more precise than one alone. They tell AI both what to do and how to frame the result.
Prompting by Level
Each example below shows a weak prompt on the left and a strong, structured prompt on the right. The improvement always comes from adding Context, Goal, and Action — not simply making the prompt longer.
Goal: Request Thursday off for a medical appointment. Keep it brief and professional.
Instruction: Draft a short email, under 80 words, with a respectful tone and an offer to arrange cover.
Input: My startup offers AI-powered English learning for corporate clients in Southern Europe. Budget: 50k euros. Team: 2 people.
Steps: Outline an executive summary, market analysis, value proposition, revenue model, and 12-month milestone roadmap.
Expectation: A structured plan I can present to angel investors.
Goal: A 60-minute lesson plan on mixed conditionals with warmer, presentation, controlled practice, freer practice, and an error correction slot.
Instruction: Provide the plan with timing, instructions per stage, and 2 ready-to-use activities requiring no printing.
After: We want a clear, defensible policy on algorithmic fairness to present to our board.
Bridge: Analyse the main types of AI bias relevant to hiring, identify 3 audit methodologies, and recommend a practical 4-step review process to implement in 90 days.
Goal: Write a 250-word story about an English teacher who receives an unexpected message from a former student years later.
Result: The story should be emotionally engaging, use B2-level vocabulary, and contain at least 3 examples of reported speech.
Example: The tone and length should resemble a short story from a graded reader series.
Most Common Prompting Mistakes
These errors consistently produce poor, irrelevant, or unreliable AI outputs. Each card shows what goes wrong and how to fix it.
Ethics, Privacy & Responsible Use
Using AI effectively also means using it responsibly. Each issue below shows the problem and the responsible solution. These principles apply to students, professionals, and teachers alike.
AI can amplify existing biases
AI algorithms perpetuate biases from training data. Outputs about groups of people can reflect historical prejudice and lead to discriminatory outcomes in hiring, lending, or allocation.
Always critically review AI outputs involving groups or demographics. Apply your own professional judgement. Use diverse, representative prompts and data sources.
Sensitive data entered into AI is exposed
Real names, addresses, medical records, passwords, and personal data entered into AI tools may be stored, processed, or used for training. This violates privacy and data protection laws.
Always anonymise data before submitting. Replace real names with codes (e.g. Patient A, Client X). Never paste passwords, ID numbers, or contact details. Collect only what is strictly necessary.
AI invents facts confidently
AI models can produce completely incorrect statistics, citations, names, and dates — and present them with full confidence. This is called hallucination and it is one of the most dangerous limitations.
Never trust AI-generated facts without verification. Cross-check statistics, citations, and names through primary sources before using them professionally or academically.
Undisclosed AI use undermines trust
Submitting AI-generated content as entirely your own — in professional reports, academic work, or client deliverables — without disclosure can be misleading and may violate institutional policies.
Be transparent about AI use where appropriate. Always review, edit, and take ownership of AI-generated content. You remain responsible for the accuracy and appropriateness of the output.
Dependency erodes critical thinking
Over-reliance on AI for decision-making, writing, and analysis can lead to a gradual loss of professional expertise, critical thinking, and independent judgement — skills that define professional value.
Use AI as an assistant, not a replacement. Actively practise your own thinking skills. Question AI outputs, seek clarity, and maintain the professional expertise that only experience builds.
Students submit AI work without critical engagement
When students use AI without understanding how it works, they miss the learning process entirely — and teachers receive work that does not reflect the student's own knowledge or development.
Teach prompting as a skill. Set clear expectations about AI use. Focus on the critical reading and evaluation of AI output as part of the learning process, not just the final product.
AI-generated content may infringe copyright
AI models are trained on vast amounts of copyrighted text and data. Content they generate may closely resemble or reproduce protected material without attribution, creating legal and ethical risks.
Treat AI output as a draft, not a final product. Always edit, add your own perspective, and verify that any specific phrasing, data, or examples are not directly lifted from a protected source.
AI is displacing certain job roles and skills
Repetitive, formulaic, and data-processing tasks are increasingly automated by AI. Workers in writing, translation, data entry, and customer service roles are already seeing significant disruption.
Focus on developing skills AI cannot replicate: critical judgement, creativity, emotional intelligence, and professional expertise. Use AI to handle the repetitive so you can focus on the valuable.
AI makes it easy to create convincing false content
Generative AI can produce realistic fake images, audio, video, and text. This technology is increasingly used to spread misinformation, impersonate real people, and manipulate public opinion.
Develop media literacy skills: question the source of content, look for verification marks, and cross-reference information. Never share AI-generated content presenting fiction as fact.
AI has a significant and growing carbon footprint
Training and running large AI models requires enormous computing power and energy. The environmental cost of AI is rarely visible to users but is a real and growing contributor to carbon emissions globally.
Use AI purposefully, not excessively. A well-crafted single prompt is more efficient than 10 poorly-written ones. Be aware of the resource cost and factor it into decisions about when AI use is justified.
Tone & Register
Tone is one of the most powerful — and most overlooked — elements of a prompt. The same information delivered in the wrong tone can confuse, offend, or simply fail to connect. Always specify the tone you need.
Register: how to specify it in a prompt
Register is the level of formality relative to your audience and context. Use these phrases directly in your prompts:
| Register | Prompt phrase to use | Avoid these words | Best for |
|---|---|---|---|
| Very formal | "Use very formal language. No contractions. Third person where possible." | I think, maybe, kind of, lots of | Legal, diplomatic, official |
| Professional | "Professional and clear. Contractions are acceptable. First person is fine." | Slang, filler words, emojis | Business, workplace |
| Semi-formal | "Semi-formal — friendly but structured. Suitable for a professional acquaintance." | Very technical jargon, very casual slang | Networking, introductions |
| Informal | "Informal and conversational — as if writing to a friend or colleague you know well." | Overly stiff vocabulary, passive constructions | Team chats, social posts |
| Plain English | "Use plain English. Short sentences. Simple words. Anyone should understand this." | Jargon, complex clauses, abstract nouns | Public information, A2-B1 learners |
Specifying Output Format
Without a format instruction, AI chooses the structure for you — often producing a wall of text when you needed a table, or a long essay when you needed three bullet points. Always tell AI how to present the output, not just what to produce.
Length control phrases
Always pair a format instruction with a length constraint. These phrases work in any prompt:
Advanced Prompting Techniques
Once you are comfortable with the CGI framework and instruction words, these techniques will take your prompting to the next level. Each one solves a specific limitation of basic prompting.
AI performs better on focused, single-step tasks than on large compound requests. Chaining lets you review and refine each stage before moving forward — like building with blocks rather than pouring everything at once.
The role you assign shapes the vocabulary, depth, assumptions, and perspective of the response. "Explain investment risk" gets a different answer from a financial advisor, a teacher, or a journalist — and rightly so.
Examples communicate what instructions cannot. Instead of describing the format in words, you show it — and AI matches the pattern. This is especially powerful for consistent style, tone, or structure.
For analytical or problem-solving tasks, asking AI to "think out loud" produces more accurate and verifiable answers. You can follow the logic and catch errors before they reach the conclusion.
Your first prompt is never the final product — it is the starting point. Each follow-up instruction narrows, sharpens, or redirects the output. Three focused refinements produce a far better result than one long initial prompt.
Constraints eliminate the most common problems before they appear. Instead of correcting output after the fact, you prevent bad output from being generated in the first place.
Prompt Starter Templates
These ready-to-use templates follow the CGI framework. Copy, fill in the [brackets], and send. Each template includes the Context, Goal, and Instruction already structured for you.
Quick Reference & Cheat Sheet
Everything you need on one page. Use this as a classroom handout, a desk reference, or a checklist before you send any prompt.
AI Prompting — Quick Reference
abcenglishonline.com · Kyle Atkins · Generative AI Course
| Framework | Best for | Key question to ask |
|---|---|---|
| R·T·F | Single clear deliverable | What role, what task, what format? |
| T·A·G | Performance or evaluation | What task, what action, what measurable goal? |
| B·A·B | Problem solving | What is wrong, what should be true, what is the path? |
| C·A·R·E | Marketing & content | What context, what to do, what result, what example? |
| R·I·S·E | Multi-step planning | What role, what input, what steps, what expectation? |
| AI Interview | Complex, unclear needs | Ask AI to ask you the right questions first |