If you’ve used Claude, ChatGPT, or any AI assistant, a “prompt” is simply the request you type in — a question, an instruction, a task. The quality of that request is the single biggest factor in the quality of the response you get back.
Every prompt that works well is built from up to three ingredients: instructions, context, and examples. Most people who use AI tools regularly have absorbed these ideas without naming them. This article names them, so you can use them on purpose.
Knowing the names of these three ingredients is not the same as knowing which one is missing when a response falls short. This article is a diagnostic guide. It explains what each ingredient actually does, what a response looks like when one is missing, and how to identify the gap quickly instead of guessing. If you’re new to prompting in general, our complete prompting guide is the best starting point before this one.
Why This Distinction Matters
A vague prompt and a flat response can have several different causes. Sometimes the task itself wasn’t clear. Sometimes the task was clear, but the model had no relevant background to work with. Sometimes both were fine, but the desired style was never demonstrated.
These are three different problems. Each has a different fix. Treating them as one undifferentiated issue — “the prompt needs more detail” — leads to padding a prompt with extra words that don’t address the actual gap.
The faster path is to diagnose first. Look at what the response got wrong, then identify which ingredient was missing.
Instructions: What They Actually Do
An instruction defines the task, its format, its length, and its scope. It answers the question: what, specifically, should the output be?
What a response looks like when instructions are unclear: The output addresses the general topic but not the actual task. Ask for “help with our pricing page” and you’ll get generic thoughts on pricing pages. Ask to “rewrite this pricing page section to lead with value instead of features, in under 100 words” and the response has an actual job to do.
How to fix it: State the task as an action, not a subject. Add format, length, and scope. “Help with X” is a topic. “Do Y to X, in this format, at this length” is an instruction.
Context: What It Actually Does
Context is the situational information that makes an answer specific to your circumstances rather than generic advice that could apply to anyone.
What a response looks like when context is missing: The output is technically correct but not useful, because it wasn’t built around your actual situation. Ask “how should I price this product?” and you’ll get general pricing theory. Add that you’re a two-person business selling to enterprise clients with a 6-month sales cycle, and the answer changes entirely.
How to fix it: State the specific facts that would change the answer: your business type, your audience, your constraints, what you’ve already tried. A rule of thumb: if removing a sentence from your prompt wouldn’t change what a good answer looks like, that sentence isn’t context — it’s filler. If removing it would change the answer, it belongs in the prompt.
Examples: What They Actually Do
An example shows the model a target — a tone, a format, a level of detail — instead of describing it in the abstract.
What a response looks like when an example is missing: The instructions and context are both clear, but the style still doesn’t match what you had in mind. This usually shows up on subjective qualities: tone, voice, structure, level of formality. Words like “professional but approachable” can be interpreted many different ways. A single sample sentence removes that ambiguity.
How to fix it: Include one representative sample of the output you want — a sentence written in the right tone, a past email that worked well, a competitor’s structure you want to follow (with your own content, not theirs). One example is usually enough. More than two or three tends to add noise rather than clarity.
The Diagnostic Framework
Use this table when a response isn’t working. Match the symptom to the likely missing ingredient before rewriting the whole prompt.
| Symptom | Likely Missing Ingredient | Fix |
|---|---|---|
| Response addresses the topic but not the actual task | Instructions | Add a specific action, format, and length |
| Response is correct but generic, could apply to any business | Context | Add specific facts about your situation |
| Task and background are clear, but tone or style is off | Examples | Add one sample of the target style |
| Response covers too much, or the wrong things | Instructions | Narrow the scope explicitly |
| Response ignores something you assumed was obvious | Context | State it directly — nothing is assumed |
This table is a starting point, not a rulebook. Some weak responses involve more than one missing ingredient. When that happens, fix the most obvious gap first, then reassess.
A Common Misdiagnosis
The most common mistake is treating a context problem as an instruction problem. When a response feels generic, it’s tempting to add more formatting requirements or restate the task differently. This rarely helps, because the task was never the issue. The response was missing something about the actual situation it was meant to address.
Before adding more instructions to a prompt that already has clear ones, check whether the real gap is context instead.
Practical Takeaway
When a response underperforms, ask three questions in order:
- Did it do the wrong thing? Missing or unclear instructions.
- Did it do the right thing, but generically? Missing context.
- Did it do the right thing, but in the wrong style? Missing an example.
This order matters. Instructions problems are usually the most visible, context problems are the most common, and example problems are the easiest to fix once identified. Diagnosing in this order prevents the most frequent mistake: adding more instructions to solve what was actually a context gap.
FAQ
Do I need all three ingredients in every prompt? No. Simple, factual requests often need only clear instructions. Complex or subjective tasks are more likely to need context, an example, or both.
Which ingredient matters most? Instructions are necessary in almost every prompt. Context is the most common gap in prompts that already have clear instructions. Examples matter most for tasks with a subjective style component.
Can too much context hurt a prompt? Irrelevant detail adds length without adding clarity. Relevant detail — anything that would change what a good answer looks like — is worth including regardless of length.
What if I’m not sure which ingredient is missing? Look at what the response got wrong first, using the diagnostic table above. Diagnosing before rewriting is faster than guessing and revising blindly.