The Prompt Engineering Cheat Sheet

Open with who the model should be. 'You are a senior B2B copywriter with 10 years writing SaaS landing pages.' The more specific the role, the more consistent the tone and judgment. Generic roles ('you are a helpful assistant') produce generic output.

Tell the model the situation it is operating in. Who is the audience? What is the goal? What has already happened? One short paragraph here eliminates the most common failure: output that is technically correct but wrong for your actual situation.

State one clear, specific action. 'Write a 200-word subject line test for a cold email sequence targeting e-commerce founders' is a task. 'Help me with email' is a request for the model to guess. Use verbs: write, list, rewrite, compare, summarise, draft, extract.

Specify exactly what the output should look like before you see it. Bullet list? Numbered steps? Table? Three options side by side? Word count? Heading structure? If you do not specify format, you get whatever the model defaults to, which is rarely what you need.

Tell the model what NOT to do. 'Do not use bullet points. Do not use the word leverage. Do not include an introduction paragraph.' Negative constraints cut the most common AI-tells before they appear. Build a short personal banned-word list and paste it into every prompt for written content.

Give a word count or a count of items. 'Three options, each under 50 words' is more useful than 'keep it short.' AI will always write more than you need unless you cap it. Cap it.

State the specific person reading the output. 'Write this for a founder who has never run paid ads, not for a marketer.' The audience shapes vocabulary, assumed knowledge, and tone. Without it, the model aims at a generic reader who does not exist.

Paste one example of what good looks like. 'Here is an email I wrote that worked well. Match this tone.' One real example is worth ten adjectives. If you want output that sounds like you, show it your actual writing, not a description of your writing.

For complex outputs, break the task into sequential prompts. First ask for an outline. Review it. Then ask for the full draft. Bundling 'outline + draft + social posts' into one prompt produces rushed output on all three. Separate prompts give each task full attention.

For decisions, analysis, or strategy, add: 'Before you give me the answer, walk me through your reasoning.' This catches wrong assumptions early and produces more accurate conclusions. Useful for any prompt where the logic matters, not just the output.

Tell the model whose point of view to write from. 'Write from the perspective of the buyer, not the seller.' Or: 'Write this from the perspective of someone who is sceptical about AI tools.' Perspective shifts the framing of every sentence without you having to rewrite manually.

If output is too generic, the problem is almost always under-specified context. Do not ask it to 'be more specific.' Tell it exactly what specific means: 'include the actual tactic name, the platform, and the expected result.' Adjectives ('more vivid,' 'more engaging') produce inconsistent results. Concrete instructions produce consistent ones.

If the tone is off, do not describe the tone you want. Paste a sentence or two of what the right tone looks like and say 'rewrite to match this.' Showing is always faster than describing for tone corrections.

If most of the output is good and one section is wrong, do not regenerate the whole thing. Paste only the broken section and ask for a targeted fix. 'Rewrite only the third paragraph. The rest is fine. The problem with the third paragraph is [specific issue].' This saves the good parts and avoids new errors appearing elsewhere.

Models pay most attention to the beginning and end of a prompt. If you have one critical rule (e.g. 'write in first person' or 'do not mention pricing'), repeat it at the end of the prompt, after all the context. Instructions buried in the middle of a long prompt are the ones most likely to be missed.

Pasting everything you know about a topic does not improve output. It often makes it worse because the model spreads attention across irrelevant material. Include only the context that directly affects the task. If you are writing a subject line, the model does not need the full brand history.

Long conversation threads carry context from earlier turns. If you are starting a new task, start a new conversation. Old instructions, tones, and assumptions from earlier in the thread will bleed into new outputs if you stay in the same session.

Ask yourself: (1) Did I specify the role? (2) Did I give enough context for a stranger to understand the situation? (3) Is the task one specific action or several bundled together? (4) Did I specify the output format and length? If the answer to any of these is no, add it before sending. Two minutes of prompt work saves twenty minutes of editing.