Prompting 101

You wouldn't tell a builder to "Just build a house". You would give them a blueprint. Prompting works the same way. It is the set of instructions we give to AI systems that turns plain english into code, which tells the AI what output you want.

It requires components such as the context (background), constraints (boundaries) and format to shape the raw data you put in, to generate and narrow down the output for your desired goal and standards. [1–2]

Official sources such as OpenAI (ChatGPT) and Google (Gemini) guide their users to be clear, specific and use relevant context to get the best output [1–4]. By standardizing your inputs, you bypass the AIs randomness and guarantee a better result.

Regular prompting is basically just typing and crossing your fingers for a good response. Prompt engineering is different. It uses a design system to make sure the outputs are consistent, reliable and repeatable. It provides the structure and clarity needed to stop the AI from guessing, ensuring you get the best output for the tasks you do over and over again. [2–4]

This is It's proven too, with research showing that changing only the prompt, significantly improved how well Large Language Models (LLMs) solve complex problems without the need of retraining [7,8]. The quality of your input is the difference between getting the job done in seconds or wasting hours with corrections!

The same prompt you paste into one Large Language Model will behave differently. AIs such as ChatGPT, Gemini, Claude and other systems can widely differ in quality of output due to:

• How they’re trained and tuned.
• What safety layers they apply.
• How literally they follow instructions.
• What the AIs are optimised for (tool use, style, formatting, long context, etc.) [1,4]

So, yes, there is no single universal prompt. However, the principles carry over to each AI: Clarity, context, constraints and format are essential for the best output.

Most "prompt hacks" you see online are based on outdated data or "vibes" rather than how the technology actually works. Instead, rely on official documentations written by the engineers who build and tested these models. These guides define exactly what the systems can and cannot do especially with safety protocols and limitations. Understanding the mechanics is the only way to get consistent results without wasting time with trial and error.

Official guides also say prompting is an iterative process. Write a prompt, review and refine, like drafting an essay. This is very practical and true. [3,6]

But here's the part most people miss: Iterations are easier when you start well.

A vague first prompt confidently sends the AI into the wrong direction which not only produces a weak output but you pay for it with:

• More back and forth
• More tokens processed
• More time lost
• Higher usage cost for paid tools

OpenAI’s own cost optimization guide says reducing tokens and requests reduces cost and improves latency [13]. But the cost is not only money. Every prompt uses computing power and energy. Writing higher quality prompts with fewer iterations reduces waste.

Google's 2025 technical audit of Gemini apps found that a median text prompt consumes an estimated 0.24Wh of energy and 0.26 mL of water [15]. While this number is low, the cost compounds every time you go back and forth. Getting your instructions right the first time will not only help you save time and money, but also the planet.

When making the best general prompt that is backed by official guidance from Google, OpenAI and Anthropic, you must follow these rules:

• Start with the task (e.g., summarize, draft, explain) what you want the AI to actually do.
• Give it context (e.g., what audience, what goal to achieve, what is used for and what inputs to use).
• Set boundaries (e.g., a word count, tone, do/don’t include, assumptions).
• Specify the format (e.g., make a table, give bullet points, or step-by-step instructions).
• Add verification instructions (e.g., separate facts from assumptions, include sources I can verify) [1-4]

The more specific you are, the better the output will be and the less hallucinations it will have.