Prompt Engineering: How to Get Better Results from AI Tools

The quality of what you get from an AI tool is almost entirely determined by the quality of what you put in. Poor prompts produce generic, shallow output. Specific, well-structured prompts produce work that is actually useful. This is not a mystery — it is a skill, and like any skill it can be learned in an afternoon.

Why Vague Prompts Fail

When you type something like “write me a blog post about crypto,” the AI has almost no information to work with. It defaults to the most generic possible interpretation — a surface-level overview aimed at no particular reader, making no particular argument. The result is technically correct and practically useless.

The AI is not being lazy. It is doing exactly what you asked. The problem is that what you asked was almost nothing. The model has been trained on billions of documents and will produce statistically average output when given no constraints. Average is what you get without clear direction.

The Four Elements Every Good Prompt Needs

Four things transform a bad prompt into a useful one: role, task, context, and format. You do not need all four every time, but understanding them gives you the tools to diagnose why any prompt is producing bad output.

Role tells the AI whose perspective to write from. “You are a UK financial journalist with 10 years of experience explaining crypto to mainstream readers” produces dramatically different output than no role at all. The AI adjusts vocabulary, assumed knowledge, tone and emphasis based on the role you assign.

Task is the actual request, stated with precision. “Write a 400-word explainer on Bitcoin halving for someone who has never bought crypto” beats “explain Bitcoin halving.” Specificity of length, audience, and scope all help.

Context is background information the AI needs but would have to guess at otherwise. “This is for the UK audience — include GBP prices and reference HMRC guidance” steers the output without you having to specify every detail.

Format controls how the output is structured. “Give me three options with pros and cons for each” produces a different result than the same content in prose. Be explicit: bullet points, numbered lists, headers, word counts, tables — state what you want.

The Chain of Thought Technique

For complex tasks, telling the AI to reason step by step before producing an answer improves output quality noticeably. Add “Think through this step by step” or “Before answering, consider the key factors” to prompts that require analysis or decision-making. This forces the model to surface its reasoning rather than jumping to a conclusion.

When I tested this on a tax calculation prompt — asking Claude to work out UK capital gains tax on a crypto portfolio — the chain-of-thought version correctly identified that the annual exempt amount had changed in April 2024, while the direct answer version used the old figure. The difference was one sentence added to the prompt.

Few-Shot Examples

Showing the AI what you want is often more effective than describing it. If you want a specific tone or format, include an example of something that matches it. “Write a tweet in this style: [example tweet]” produces much more on-brand output than “write something casual and punchy.”

For business use cases — sales emails, customer responses, content in a specific brand voice — keeping a small library of examples to paste into prompts pays dividends. The model generalises from concrete examples far better than from abstract style descriptions.

Iterating Instead of Starting Over

Treating the first output as a draft to iterate on, rather than a result to accept or reject, changes how you use AI tools. When the first output misses the mark, add a follow-up that specifies exactly what was wrong: “That is too technical — rewrite assuming the reader knows nothing about blockchain.” Or: “Shorten the third paragraph, it is repetitive.” Or: “Add UK-specific regulation information to the compliance section.”

Iterating within a conversation is almost always more efficient than writing a longer initial prompt. You can course-correct as you see what the model produces.

Constraints That Improve Output

Negative constraints — telling the AI what not to do — are underused. “Do not use bullet points,” “do not start with a definition,” “do not include caveats or disclaimers,” and “do not summarise what you just said at the end” all produce meaningfully better output in specific contexts. The model’s default behaviour includes a lot of patterns that are individually annoying but never explicitly forbidden.

Word counts and target audience constraints both sharpen output significantly. An AI asked to write for a specific person — “a 45-year-old UK small business owner with no technical background” — writes with more useful specificity than one asked to write for a generic audience.

What This Means for You

Every hour spent learning to prompt well pays back in every future interaction with any AI tool. The underlying techniques — specific role, precise task, relevant context, explicit format, iteration on drafts — apply equally to ChatGPT, Claude, Gemini and whatever comes next. Prompting is not a ChatGPT skill. It is an AI literacy skill, and it is already separating people who get useful output from tools from people who remain frustrated with them.

This article is for educational purposes only and does not constitute financial advice. Cryptocurrency investments involve significant risk. Always do your own research.