Prompting isn't magic words. It's clear communication with a very literal colleague. These ten habits cover most of the gains.
1. Say who it's for
“Explain DNS” vs “Explain DNS to a non-technical small-business owner in 3 sentences”. The second gets a far better answer.
2. Give the context it can't guess
Your product, your constraints, what you've already tried. Models don't know your situation unless you tell them.
3. State the goal, not just the task
“Write an email declining the meeting so that we keep a good relationship and propose next month instead.”
4. Specify the format
Bullet points, a table, JSON, “under 100 words”, “with headings”. Format instructions are followed very reliably.
5. Show an example
One good example beats a paragraph of description. See few-shot prompting.
6. Use clear sections
Separate instructions from data with headings or tags:
## Instructions
Summarise the customer's issue in one sentence.
## Customer message
<message>...</message>7. Tell it what to do when unsure
“If the answer isn't in the text, say ‘not found’.” This prevents confident guesses.
8. Ask for reasoning on hard problems
“Think through the edge cases first, then give the final answer.” Or use a reasoning model.
9. Break big tasks into steps
Outline → draft → review works better than one giant request.
10. Iterate with real examples
Keep a handful of test inputs. Change one thing at a time and compare. See testing prompts.
Claude, GPT, Gemini, DeepSeek, image and video models through one OpenAI- and Anthropic-compatible endpoint. Prepaid, pay only for what you use.