Structure of a prompt
The best structured prompts typically start with the relevant data or context at the top and include the specific instructions at the bottom. This allows the AI to first understand the information it has to work with, and then what to do with that information.
Example: If you want to AI to write a short story about a pirate named Captain Hook who loves tropical fruit, you might structure your prompt like this:
Data: Captain Hook is a pirate who loves tropical fruit.Instruction: Write a short story centered around Captain Hook's quest for a legendary tropical fruit.
Use of Positive Instructions
Positive instructions tend to work better with AI compared to negatives. When you specify what you do want, it gives the AI a clear direction. Negative instructions can sometimes leave a bit too much room for interpretation.
Example: Instead of saying: Don't use complicated words in the story
It's better to instruct: Use simple, easy-to-understand language in the story.
Specific Action Verbs
Using precise and descriptive action verbs can help get more intended results from the AI. A broad action like 'write' can be interpreted in many ways, depending on the context.
Example: Instead of saying: Give an ending to the story
You might get more dramatic and satisfying endings with a command like: Elaborate an adventure-packed ending to the story.
The 'Sheet of Paper Test'
Just like humans, AI also needs clear and context-rich instructions. The 'Sheet of Paper Test' means that if another person can understand your instructions as written on a sheet of paper, then likely the AI will too.
Example: Imagine you left a sheet of paper with this written on it: Write about Captain Hook's adventure.
A human might wonder what kind of adventure? Where is it set? What's the tone? If you provide more context: Write a humorous short story about Captain Hook's adventure in search of a legendary tropical fruit in the Caribbean.
Both a human and AI have much more context to work with, leading to a more satisfying result.
Use of Explicit Requirements
Being clear from the start about your specific requirements, like formatting or style, helps in returning the desired output.
Example: If you need a report summary in bullet points, make sure to mention this in your prompt.
Instead of saying Summarize the report, say Summarize the report in bullet points.
Avoid Duplicate Labels
When providing data to the AI, ensure each piece of data is labeled uniquely to avoid any confusion or incorrect analysis.
Example: If you have data on sales figures from different years, rather than labeling them as 'Sales 1', 'Sales 2,' etc., label them as 'Sales 2019', 'Sales 2020,' etc.
Experiment with Synonyms
If your initial prompt doesn't get the desired results, try the same instruction but with different words or phrasing.
Example: If Summarize the report is not providing the depth of information you need, try Provide a detailed overview of the report.
Using "Integrative" or "Transformative" Phrasing
Such language can help guide the AI's approach to your prompt.
Example: If the regular instruction, Write a story about a haunted house brings a cliche response, you could apply transformative phrasing as, Reimagine the story of a haunted house from the perspective of the ghost.
Prompts Should Be Rich and Clear
Providing the AI with detailed, coherent prompts will yield the most comprehensive results.
Example: Instead of requesting, Write an article, a prompt like Write an informative article about the effects of climate change on biodiversity over the last decade gives the AI much more substance, tone, and direction to work with.