⚡ 11 AI Prompt Tips from the Google 68-Page Prompt Engineering Guide

Use high-quality examples
• Giving a few examples helps the AI understand your tone, structure and edge cases.
• Just don’t overload it with too many examples—it can confuse the model.

Start simple
• Clear, concise prompts lead to better results.
• Remember: clarity is more important than trying to be clever.

Define the output
• Instead of just saying “Summarize this,” be specific: “Give me a 3-sentence summary in bullet points.”
• The clearer you are, the better the result.

Prefer positive instructions
• Always tell the model what to do, not what to avoid.
• Unless you’re setting safety constraints, keep the instructions affirmative.

Use variables for flexibility
• Add placeholders like [NAME] or [PRODUCT] to make your prompts dynamic and reusable.
• This way you can easily adapt prompts to different situations.

Experiment with input formats
• Tables, lists or even JSON structures can be much more effective than long paragraphs.
• These formats help guide the AI’s attention and make the output clearer.

Keep testing across models
• What works well on GPT-3.5 might not work as well with GPT-4.
• Test your prompts across different models and always iterate.

Ask for structured outputs
• If you’re building workflows, ask for outputs in formats like Markdown, JSON or CSV.
• The cleaner your input, the cleaner your output will be.

Collaborate on prompts
• Prompt engineering works best as a team effort.
• Don’t hesitate to share prompts, compare results and refine them together.

Use Chain-of-Thought sparingly
• “Let’s think step by step” is helpful for guiding reasoning.
• But using it too much can actually slow down more advanced models.

Track prompt versions
• Keep a log of your inputs and outputs.
• This will save you from reinventing the wheel every time you need to use a prompt.