Prompt engineering sounds intimidating. It isn’t. It’s just the practice of asking AI well — and these tips cover 90% of what matters, jargon-free.
1. One job per prompt
Prompts fail when they ask for five things at once. “Write a blog post, make a social caption, and suggest images” gives you three mediocre outputs. Split it: three prompts, three good outputs.
2. State the audience
“Explain photosynthesis” vs. “Explain photosynthesis to a curious 8-year-old” vs. “…to a biology undergrad.” Same topic, three correct-but-different answers. The audience line is the highest-ROI sentence you can add.
3. Use delimiters for your content
Wrap pasted material in triple quotes or XML tags: Summarize the following: “[text]”. This stops the AI confusing your instructions with your content — the #1 cause of weird outputs.
4. Ask for a format, not just an answer
“List,” “table,” “under 100 words,” “as JSON” — format instructions are obeyed far more reliably than vague quality words like “good” or “detailed.”
5. The “think step by step” trick
For reasoning tasks (math, logic, planning), add: “Think through this step by step before answering.” This single sentence measurably improves accuracy on complex problems.
6. Give a quality bar with an example
Instead of “write a great headline,” paste a headline you admire: “Write 5 headlines in the style of this one: [example].” Examples calibrate quality instantly.
7. Iterate in the same chat
Follow-ups (“shorter,” “funnier,” “now for email”) build on context. Starting a new chat for each tweak throws away the AI’s understanding of what you liked.
8. Save what works
When a prompt nails it, save the exact wording. Build a personal library of proven prompts — or start from our prompt library of 32 ready-made examples.
9. Know your model’s quirks
Chat models love conversation and follow-up; image models need visual nouns and style words; video models need camera language. The same idea needs different phrasing per medium — match the prompt to the model’s strengths instead of fighting them.
10. Test the same prompt twice
AI output varies run to run. If a prompt works once, run it twice more before declaring it reliable. The prompts worth keeping are the ones that work three times in a row — those are your templates.
The 80/20 of prompt engineering
Specific task + audience + format + one example beats 95% of prompts people write. Everything else is refinement. For the full deep-dive, read our guide on how to write AI prompts.