Loading…
Loading your learning journey…
Loading your learning journey…
Loading…
Prompt Engineering
Most people who say “AI doesn't work for me” are writing weak prompts. This guide gives you the frameworks, failure modes, and real examples you need to get consistently excellent results from ChatGPT, Gemini, Claude, and every other AI tool — all free to apply right now.
The RCTF framework covers the four ingredients that, combined, make almost any prompt dramatically more effective. You do not need all four every time — but adding even one improves your results.
Tell the AI who it should be. This sets the expertise level, vocabulary, and perspective of the response.
"Act as an experienced financial analyst..."
The same question asked of a 'financial analyst' vs no role gets different depth and framing.
Give the background information the AI needs to give a relevant answer.
"I'm preparing a pitch for a Series A startup in the edtech space targeting India..."
Without context, the AI makes assumptions that may not match your situation.
Be precise about what you want done. Use action verbs: write, summarise, compare, critique, list, explain.
"Summarise the three strongest arguments and the three weakest ones..."
Vague tasks produce vague outputs. Specific tasks produce specific outputs.
Specify how the output should be structured — length, style, structure, and audience.
"Write this as a 5-bullet executive summary, under 150 words, for a non-technical audience."
Format instructions are the single quickest win for most users — AI will match whatever structure you specify.
These are the patterns that reliably produce disappointing AI output. Recognising them is the fastest way to improve.
Example: "Write a business plan."
Fix: Add what kind of business, who it is for, what stage the business is at, and what format you need. The AI cannot guess context you have not shared.
Example: "Explain machine learning."
Fix: Add 'to a 12-year-old' or 'to a senior executive with no technical background' or 'to a first-year computer science student'. Audience is the most powerful single word you can add.
Example: "Tell me everything about marketing."
Fix: Narrow the scope radically: a specific aspect, a specific industry, a specific goal, a specific time constraint. Broad questions produce broad, shallow answers.
Example: "Make it better." (after pasting text)
Fix: Define what 'better' means: shorter, clearer, more persuasive, more formal, more engaging for X audience. Without a criterion, the AI edits in a direction you may not want.
Example: Giving up after one mediocre response.
Fix: Treat every AI interaction as a conversation. The first response is a draft. Follow up: 'That's too formal — try again more conversationally' or 'Give me three alternatives to the third paragraph.'
Example: Accepting statistics, quotes, or citations without checking.
Fix: For any factual claim that matters, verify it independently. Ask the AI: 'How confident are you in this statistic, and where should I verify it?' Use Perplexity for research that needs real citations.
For complex reasoning tasks — analysis, strategy, problem-solving — add “Think step by step” or “First, identify the key issues. Then...” to your prompt. This forces the AI to reason through the problem rather than pattern-match to a surface answer. The improvement on analytical tasks is substantial.
In ChatGPT's Custom Instructions (or Claude Projects), set a standing system prompt that defines the AI's role and rules for all conversations. You never have to repeat your context.
The most productive AI users do not try to write the perfect prompt upfront. They start rough, get a first draft, then refine in three to four turns. Typical refinement sequence:
Summarise a long text
Summarise the following in 5 bullet points for a busy executive. Focus on decisions and action items. [PASTE TEXT]
Improve your writing
Improve this text. Make it clearer and more direct. Cut any repetition. Keep my original meaning. [PASTE TEXT]
Generate ideas
Give me 10 ideas for [TOPIC]. Include a mix of conventional and unexpected approaches. One sentence per idea.
Explain a concept
Explain [CONCEPT] to a 14-year-old with no prior knowledge. Use an everyday analogy and avoid jargon.
Create a to-do list
I need to accomplish [GOAL] in the next two weeks. Create a realistic daily to-do list with 3–5 tasks per day.
Critique a plan
Here is my plan: [PLAN]. Give me 3 reasons this could fail and how I should adjust for each one.
Write a professional email
Write a professional email to [RECIPIENT TYPE] asking for [REQUEST]. Tone: [formal/friendly]. Under 120 words.
Learn something new
Teach me [SKILL/TOPIC] from scratch. Give me a 5-step learning path with one specific action per step.
Translate tone
Rewrite the following text in a [formal/casual/urgent/empathetic] tone while keeping the same meaning. [PASTE TEXT]
Debate check
I believe [POSITION]. Give me the 3 strongest counter-arguments I should be prepared to answer.
Prompt engineering is the skill of writing instructions to AI tools that reliably produce the output you need. It is not coding — it is clear communication. Just as asking a clear question to a colleague gets a better answer than a vague one, a well-structured prompt consistently outperforms a casual one.
You can get basic results without any specific technique. But prompt engineering dramatically improves consistency and quality — especially for complex tasks. Most experienced AI users develop prompt habits that become second nature within a few weeks of practice.
The core principles (specificity, role, format, context) apply to ChatGPT, Gemini, Claude, and Perplexity equally. Some advanced techniques — like system prompts — are specific to certain tools or API access, but everyday prompting skills transfer directly.
Add a format instruction. Instead of just asking a question, add: 'Give your answer as a numbered list of 5 items' or 'Respond in under 100 words' or 'Format this as a comparison table'. Format instructions alone dramatically improve usability of AI outputs.
A system prompt is a set of instructions you give the AI before the main conversation to set its persona, rules, and style. In ChatGPT, you can set this in the custom instructions section. In Claude Projects, it is the project instructions. System prompts let you create reusable AI assistants tuned to specific tasks.
CertifAI's Level 1 Module 2 covers prompt engineering in depth — the RCTF framework, 6 failure modes, 3 advanced patterns, and hands-on practice with ChatGPT and Gemini. Part of the free Foundation sessions.