How to Write ChatGPT Prompts: 12 Patterns That Actually Work

Most "how to write ChatGPT prompts" guides give you adjectives ("be specific!", "give context!") and call it a day. This one gives you 12 reusable patterns, each with a working example you can paste into ChatGPT today, plus when to reach for it.

The one rule under every pattern

A prompt is a contract. You are telling the model who it is, what to produce, who it is for, and how to format the answer. Vague contracts return vague work. Every pattern below is a way to make one part of that contract sharper.

1. Role + Goal + Audience + Format (RGAF)

The default skeleton for any prompt longer than a sentence.

You are a senior B2B copywriter. Write a 120-word LinkedIn post announcing our new pricing page. Audience: founders of 10-50 person SaaS companies. Format: 3 short paragraphs, no hashtags, end with a single CTA question.

Use this when you would otherwise type one line and hope.

2. Show, don't tell (few-shot)

Paste 2-3 examples of the output style before asking for a new one. The model copies tone, length, and structure from examples better than from adjectives.

3. Constraints over instructions

"Make it short" is weak. "Maximum 80 words, no adverbs, no semicolons" is enforceable. Hard limits beat soft requests every time.

4. The "expert panel"

Answer as a panel: a skeptical CFO, a growth marketer, and a customer support lead. Each gives a one-paragraph take, then a 3-line consensus.

Great for decisions where you want disagreement surfaced, not smoothed away.

5. Chain of thought, then verdict

Think step by step. List your reasoning as bullet points. Then give the final answer on a new line starting with "Answer:".

Forces the model to commit to a structure you can audit.

6. Negative examples

Here are 3 subject lines I do NOT want, and why. Now write 5 that avoid these failure modes.

Negative examples often teach faster than positive ones.

7. The loop pattern

Instead of one giant prompt, write a small prompt you can run again and again with different inputs. (This is what we call a "loop" and have 1,000+ of in the Loop Library.) Example:

Rewrite this paragraph for a 12-year-old reader. Keep all numbers and proper nouns. Paragraph: {{paste}}

8. The critique pass

After ChatGPT answers, send: "Now critique your own answer as a strict editor. List 5 problems, then rewrite." You usually get the version you actually wanted on the second pass.

9. Anchor to a source

Paste the source material first, then ask the question. "Using ONLY the text above, answer X." Cuts hallucination dramatically.

10. Specify the failure mode

If you are not at least 80% confident, reply with "I don't know" instead of guessing.

The model will actually do this. Most users never ask.

11. Output as JSON

When the answer feeds into anything (spreadsheet, app, another prompt), demand JSON with a schema. "Return only valid JSON matching this shape: {title: string, tags: string[]}." No prose, no markdown fences.

12. The reset

When a conversation drifts, start a new chat and paste the last good output as context. Long threads accumulate noise; fresh threads with good context outperform them.

Putting it together

The best prompt writers do not memorize tricks. They keep a personal library of patterns that worked, and reach for the right one. That is exactly what SimpleAI Prompt's Loop Library is for: 1,000+ pre-built prompts you can copy, adapt, and stop reinventing.

Start with three patterns this week: RGAF, the critique pass, and "if not confident, say I don't know." Your outputs will jump in quality immediately.