How to Prompt Correctly
To prompt correctly, turn a vague intention into a structured instruction: state your goal, give the model a role and the context it needs, specify the output format, and add the constraints that matter. A correct prompt removes the ambiguity that forces a model to guess — so it spends its capacity answering your actual question instead of inventing the parts you left out. Everything below is the structure that makes that repeatable.
Prompting correctly is specification, not phrasing.
Most people treat a prompt as a question and hope the model fills in the rest. Prompting correctly is the opposite move: you do the thinking first and hand the model a clear specification. The skill is not memorising magic words — it is making your intent explicit enough that there is only one reasonable way to answer.
There is a cognitive reason this works. John Sweller's cognitive-load theory shows that working memory is narrow and easily overloaded — and a language model behaves the same way under a sprawling, ambiguous request. When a prompt is vague, the model burns capacity resolving contradictions and guessing at missing pieces, and the answer gets shallow. A structured prompt offloads that work: by naming the role, the context, and the format up front, you leave the model's full capacity free for the reasoning you actually care about. Structure is not bureaucracy — it is how you keep the hard part hard and everything else easy.
The bottleneck is almost never the model. It is the gap between what you meant and what you typed — and a correct prompt closes it before a single token is generated.
The seven building blocks of a correct prompt.
Almost every strong prompt is assembled from the same parts. You will not need all seven every time, but knowing them turns prompting from guesswork into a checklist: Goal · Role · Context · Instruction · Output format · Constraints · Chain-of-thought.
Before — a vague prompt
This is how most prompts start. It names a topic but no goal, no audience, no format — so the model invents all three, and the result is generic.
After — the same request, structured
Same intent, now specified. Each line maps to a building block, and the model has nothing left to guess.
The second prompt is longer, but every extra word changes the output. That is the test for whether detail belongs: if it would change the answer, keep it; if not, cut it.
Seven rules that hold across every model.
Internalise these and you will write better prompts without thinking about it. Each is one rule and one concrete example — the kind of move you can apply in your next message.
Be specific — replace every adjective with a number or a noun.
Not "write a short intro" but "write a 60-word intro for first-time visitors who have never heard of us."
Prime a role — tell the model who it is before you tell it what to do.
Open with "You are a senior tax accountant reviewing a freelancer's return" and the answer changes register, depth, and caution.
Show, don't tell — give an example of the output you want (few-shot).
Paste one product description in your house style, then write "Now write one for this product in the same style," and consistency follows.
Decompose the task — break a big ask into ordered steps.
Instead of "write the report," ask it to "first outline the sections, then draft section one," so each step is checkable.
Constrain the output — define the format, length, and what to avoid.
Add "Return exactly five bullet points, max 12 words each, no marketing adjectives" and you get a usable answer, not an essay.
Ask for reasoning — make the model think before it commits.
End with "Think through the trade-offs step by step before giving your recommendation" and accuracy on judgement tasks rises sharply.
Iterate — read the gap, then fix the single weakest block.
If the tone is off, change only the role line; if it rambled, tighten only the output format. One variable at a time.
Five reasons prompts fail — and the fix for each.
When an output disappoints, it is rarely the model. It is almost always one of these five fixable mistakes in how the prompt was framed. Learn to spot them and you can repair a bad prompt in one edit.
Mistake · Vague goal — you ask for help instead of an outcome.
Fix — State the deliverable: "Draft a two-paragraph rejection email that keeps the relationship warm," not "help me with this email."
Mistake · Missing context — the model has to guess your audience, facts, or intent.
Fix — Hand it the inputs: who the reader is, the source material, the decision already made. It cannot read your situation.
Mistake · No output format — you get a wall of prose when you needed a table.
Fix — Specify the shape up front: "Return a markdown table with columns Risk, Likelihood, Mitigation."
Mistake · Stacking many tasks into one sentence — the model does some and drops the rest.
Fix — Split it. One instruction per prompt, or an explicit numbered list of steps to complete in order.
Mistake · Politeness as instruction — "make it better" or "be creative" tells the model nothing.
Fix — Define "better" operationally: shorter, more concrete, fewer claims, a specific tone. Ambiguous praise is not a spec.
The goal is to prompt well without thinking about it.
Reading the rules is the easy part. The skill only becomes useful once structuring a prompt is automatic — when you reach for a role and a format without a checklist, the way a fluent writer reaches for a topic sentence. That shift comes from deliberate practice with feedback, not from collecting more tips.
That is exactly what PromptCorrectly is built to train. The Brain Trainer's 36 courses walk you from the mindset shift to advanced techniques, each grounded in cognitive science and gated so you cannot skip the part that sticks. When you want to see a prompt's parts laid out in front of you, the Studio visual canvas turns Goal, Role, Context, and Constraints into nodes you can wire together and run. In a hurry, the Composer lets you spec a prompt in 60 seconds from five fields, and the Library of 2,750+ prompts gives you correctly structured starting points to fork and adapt. Use them together and the structure stops being something you apply and starts being how you think.
Questions about prompting correctly.
What does it mean to prompt correctly?
Prompting correctly means translating a fuzzy intention into a structured instruction the model can act on without guessing. In practice that means stating the goal, giving the model a role and the context it needs, specifying the output format, and adding the constraints that matter. A correct prompt removes ambiguity so the model spends its capacity on your task instead of inventing the missing pieces.
How do I write a good AI prompt?
Start with the outcome you want, then build the prompt around it: assign a relevant role, supply the background facts, give one clear instruction, define the output format, and list any hard constraints. For harder tasks, ask the model to reason step by step before it answers. Then run it, read the gap between what you got and what you wanted, and tighten the weakest part. Specific beats clever every time.
What is prompt engineering?
Prompt engineering is the practice of designing the input to a language model so it reliably produces the output you want. It covers structuring instructions, giving examples (few-shot), assigning roles, controlling format and length, and decomposing complex tasks into steps. It is less about secret words and more about clear specification — the same discipline a good brief shows a human expert.
Does prompting work the same in ChatGPT, Claude, and Gemini?
The fundamentals transfer: clear goals, roles, context, format, constraints, and step-by-step reasoning improve output on every major model. The details differ — Claude responds well to XML-style tags, OpenAI models follow system messages and developer instructions, and Gemini handles long multimodal context — but a correctly structured prompt works across all of them. Learn the structure once and adapt the surface.
How long should a prompt be?
As long as it needs to be to remove ambiguity, and no longer. A one-line prompt is fine for a trivial task; a complex deliverable may need a paragraph of role and context plus a worked example. The right length is set by how much the model would otherwise have to guess — add detail that changes the output, and cut words that do not.
How do I get better at prompting?
Practice deliberately with feedback. Write a prompt, judge the output against your intent, identify which building block was weak, and fix only that. Repeat across varied tasks until structuring a prompt becomes automatic rather than effortful. Structured courses, before/after comparisons, and a visual canvas that shows each part of a prompt all accelerate the move from conscious effort to fluent habit.
Learn it once. Keep it forever.
Prompting correctly is a transferable skill — it works the same whether you use Claude, ChatGPT, or Gemini, and it does not expire when the next model ships. Start with the Brain Trainer, then build your first structured prompt in the Studio.