Introduction
ChatGPT, Claude, and Deepseek respond to the request in front of them. A useful prompt says what you need and what a good answer looks like.
How LLMs work
Language models generate a response one token at a time. They use the conversation as context, but that context has a limit. They can also state false details with confidence, so check important claims.
Basic prompting techniques
State the task, add the facts that matter, and name the output format. A role can set the point of view when it helps.
specific requestGive me a brief overview of the five major oceans of the world and mention one unique characteristic of each.
contextI’m traveling to Italy in summer and love history. Can you suggest a city to visit and why?
roleYou are a knowledgeable chef. Explain the process of making sourdough bread step by step.
formatList 5 key points about climate change in bullet form.
When an answer misses, rewrite the request with a tighter target. Ask follow-up questions that identify what to expand, simplify, or change.
rephraseExplain the top 3 health benefits of regular exercise for an adult in simple terms.
stepsPlease explain the solution step by step.
follow-upOkay, I like point 2 you gave — can you expand on that more?
Advanced prompting strategies
Use examples when the model needs to match a pattern. Separate multi-part work into clear steps. Give precise instructions for voice and format.
few-shotEnglish: Good morning -> French: Bonjour. English: Thank you -> French: Merci. English: How are you? -> French:
reasoning taskWhat is the sum of all even numbers from 1 to 50? Let’s solve this step by step. First, identify the even numbers… (and so on)
multi-part taskFirst, list the top 5 causes of climate change. Then, in a second paragraph, suggest one solution for each cause.
Creative approaches
Roles and narrative frames can change how an answer is presented. Combine them only when each instruction helps the task.
role-playLet’s role-play: you’re an IT support technician and I’m a confused customer. I’ll ask you about a computer issue and you help me solve it.
storyTell me about the water cycle, but explain it as a short story about a water droplet’s adventure.
Troubleshooting and iteration
Ambiguous requests lead to ambiguous results. Break long requests into parts. If a conversation loses the thread, start a fresh chat with the full task and the details you learned matter.
clarifyTell me about jaguars, the animals, and their habitat.
check a claimAre you sure about that? Double-check the facts above.
Real-world applications
Specific prompts are useful for writing, code, research, and planning. Give the model the relevant audience, inputs, constraints, and result.
writingYou are a travel blogger. Write a friendly, conversational blog post about visiting Paris for the first time, including 3 insider tips for budget travelers. End with an encouraging note to inspire new travelers.
researchSummarize 3 major discoveries about Mars from NASA’s missions in the last 5 years. For each discovery, briefly explain its significance. If possible, mention the mission or year it was made. Keep it concise and factual.
planningI run a small online bakery business that sells cookies and cakes. Lately, sales have been dropping. What are some strategies I can use to attract new customers and boost repeat sales? Provide a list of ideas, and for each idea, explain briefly how to implement it in the context of an online bakery.
Future trends in prompting
Models can hold more context and use more tools. Clear intent, relevant inputs, and direct constraints still give the model a stronger task.
Conclusion
Prompting improves through use. Start with a direct request, inspect the answer, and revise the next turn with the missing context or constraint.
Published here first.