Creating Request Content (Prompts)

The Generative AI action executes processes based on the instructions entered in the 'Request content'.
Therefore, it is necessary to give clear instructions to the Generative AI for desired results.
Here, we will explain the key points for creating these instructions (hereinafter referred to as prompts).

Basic Principles

Prompts should be written concisely with the minimum content in a natural, easy-to-understand sentence.
This is because describing in sentences rather than fragmented words makes it easier for the AI to understand your intentention accurately.
Here are some basic principles you should be aware of.

1. Write clearly and be specific

Prompts must include clear and specific instructions.
Vague expressions can lead to misinterpretation by the AI, and cause unexpected results to be generated.

Hint

  • Bad example: 'Write a short summary'
  • Good example: 'Please explain this product in a paragraph of 3-5 sentences'
Describe the expected result, length, style, etc., specifically as needed.
Also, writing the instructions at the beginning of the prompt leads to higher quality output.

2. Provide background information (context)

The AI does not comprehend any information other than what is provided in the prompt.
Therefore, it is important to provide the necessary background information (context) to improve the accuracy of the response.
By providing context, You can get more relevant answers.

Hint

  • Bad example: 'Write about a sales position'
  • Good example: 'Create a job description for the [Job Title] position at [Company Name], including required skills and experience, and an overview of the duties'
It is also effective to specify the target audience (e.g., experts, beginners) and specific constraints.

See also

In actions, you can include files and other items into the context.


3. Specify the output format

By clearly specifying the desired response format, it will be generated accordingly.
  • Bulleted list
  • Table format

Hint

  • Example: 'Please present the ideas in a bulleted list'
In AI actions, multiple values will be returned automatically in a table format, however,
clearly specifying it in the prompt ensures consistency.

Advanced Techniques: Creating More Advanced Prompts

In addition to the basic principles, you can use the following advanced techniques.
These techniques may improve the execution accuracy of AI action requests.

1. Define a role for the AI

By defining a role for the AI, such as a specific expert, you can effectively specify the tone, style, and perspective of the response.

Hint

  • Example: 'You are an experienced marketing analyst. Based on the following information about a new product, please list three sales points that you think will resonate most with the target customers, along with the reasons.'

2. Provide specific examples

When you want to execute complex processes or output in a specific format, it is effective to provide 'model' input and output examples.
This allows the AI to generate responses that are more aligned with the user’s purpose.

Hint

Example: Rewriting to a specific tone
Please rewrite the following sentences in a more detailed expression.

# Example 1
Input: I think this feature is pretty groundbreaking.
Output: I believe this innovative feature can provide great value to users.

# Example 2
Input: I think this approach should be fine.
Output: In my analysis, I predict that this approach will effectively solve the problem.

# Text to rewrite
Input: I’ve put the data together, so please check.
Output:

3. Break down complex tasks into smaller steps before executing them

The accuracy of response may improve by breaking down and executing complex tasks with the following methods.
・ Break down complex requests into small steps and execute them as individual Generative AI actions
・ Instructing 'Think step-by-step' or describing the execution procedure in detail to have the AI organize its thought process and execute

Hint

Example: Creating a marketing plan

Step 1 Prompt:
Please brainstorm ideas for a marketing strategy for the new product 'Smart Coffee Maker' in a bulleted list.

Step 2 Prompt (following the response from Step 1):
Thank you. From the ideas presented, please propose a concrete action plan for 'Utilizing influencers on SNS' step-by-step.

4. Structure prompts with Markdown or XML

When including multiple pieces of information (instructions, text to be processed, output format specification, etc.) in a single prompt, you can improve the AI’s understanding by clearly separating each part using Markdown headings (titling with #) or XML tags.
This method helps to accurately convey the components of the prompt to the AI and prevent misinterpretation of instructions.

Samples of Request Content for Different Patterns

1. Text Summarization/Proofreading

Here is an example prompt for asking an AI to summarize or explain a piece of text.

Example 1: Summarization

Read the following passage and briefly summarize the most important points in 3-5 sentences.
Include major themes and conclusions; specific figures and examples may be omitted.
The summary should be short and capture the core of the information.

Example 2: Proofreading

You are an expert in proofreading.
Please proofread the given text.
Pay attention to grammar, spelling, punctuation, and style of expression to ensure that the entire text is readable, clear, and consistent.
In addition, please remove redundant or ambiguous expressions and supplement explanations as necessary.
The final text should maintain a professional tone and not change the main idea from the original text.

2. OCR

Here is an example of a prompt for having the AI read the text in an image.

Example:

You are a highly accurate OCR system.
All text from the provided image is read completely and accurately for the value of the target item.
# Procedure
1. Recognize and extract the text in the image.
2. Find the specified item in the text and extract its value.
3. The image is a purchase order. The ordering company lists both a supervisor and a subordinate, with the supervisor typically listed above. Do not include field names in the output.
# Output format
1. The extracted values should be displayed side by side.
2. If the target is not found, output an empty string.
3. Do not output the item name, but only display the contents of the image.
4. If multiple files can be selected, display multiple files.
# Rules
1. The values contain uncommon words such as company or product names, be sure to output them exactly as they appear in the original.
2. Do not make any modifications to the extracted text, including corrections, adjustments, deletions, or omissions.
# Target item
1. 'Purchase Order Number': Please extract the issue date, payment date, receipt date, etc., included in the receipt, using the Gregorian calendar format.
2. 'Issue Date': Please extract the issue date included in the purchase order in Gregorian calendar format.
3. 'Purchasing Company Name': Please extract the name of the company or organization to which the order is addressed. It will include the honorific term 'Onchu' (御中).
4. 'Ordering Company Name': From the numeric postal code following the 〒 symbol, look up the address and extract the name of the company or organization placing the order within that building.

3. Category Classification

Here is an example of a prompt where the AI analyzes data in CSV format and classifies it into the given categories accordingly.

Example:

Analyze the provided file of survey response results in csv format and sort them into the specified categories.
# Procedure
1. Analyze the results of your answers.
2. Based on the analysis, assign them to the specified categories.
# Output format
1. Output should be category name only.
# Rules
1. Output only the categories you have sorted.
2. If more than one category applies, select the category that best represents the most salient element.
3. If it is difficult to classify, output “unclassified”.
4. In any case, do not include the content of the review in the output.
# Category
- Highly rated
- Claim
- Request
# Example of input output
## Example 1:
Input: I have lost weight by taking this product!
Output: Highly rated
## Example 2:
Input:This is a scam!
Output: Claim
## Example 3:
Input: I need it to be lighter.
Output: Request

4. Email Analysis

Here is an example of a prompt where the AI analyzes a received email, classifies the next action, and extracts necessary information.

Example:

The specified cell is the body of the received email and the path of the attached file.
Analyze the content of the body and the file, and output the appropriate action for the next step as a word from the choices.
The specified cell is the body of the received email and the path of the attached file. Analyze the content of the body and the file, and output the appropriate action for the next step as a word from the choices. Also, retrieve the necessary information to execute this action from the email and be sure to return it from the next column. Consolidate similar information into one.
# Procedure
1. Analyze the email body and attached files.
2. Select the appropriate action.
3. Extract necessary information.
# Output format
1. Output the action and necessary information in CSV format.
2. Information is displayed left-aligned in no particular order.
# Choices
- Create Quote
- Resend Invoice
- Handle Inquiry
- Schedule Adjustment
- Acknowledge Receipt
# Example of input output
# Rules
I cannot find the invoice for the following order that you sent the other day, and I am unable to confirm it.
Order Number: ORDER2025001
Products: 5 Laptops, 3 Monitors, 10 Wireless Keyboards
Delivery Date: June 15, 2025
Could you please resend the relevant invoice?
Thank you for your cooperation.
## Action for the above: Resend Invoice
Resend Invoice
## Necessary Information
- Order Number: ORDER2025001
- Products: 5 Laptops, 3 Monitors, 10 Wireless Keyboards
- Delivery Date: June 15, 2025
# Output format
In this case, it will be a CSV with the following 4 columns. Although there are multiple products, they are similar information so they will be in one column. Please output in no particular order and left-aligned.
"Resend Invoice", "Order Number: ORDER2025001", "Products: 5 Laptops, 3 Monitors, 10 Wireless Keyboards",
"Delivery Date: June 15, 2025"

5. Data Analysis

Here is an example of a prompt where the AI analyzes given data and provides the analysis results and suggestions based on them.

Example:

You are a excellent consultant.
Based on the last three years of sales data contained in the specified cells, analyze sales trends and related factors.
Then, from the perspective of providing excellent consulting, propose a specific action plan for the future.
# Procedure
1. Examine the entire sales data in detail.
2. Derive clear trends and correlations from the data.
3. Estimate the factors behind the identified trends and explain why.
4. Consider and propose specific actions and strategies to improve future sales.
# Output format
1. Rcognized sales trends: patterns and trends in sales that can be understood from the past three years of data
2. Factors contributing to the trend: External or internal factors that contributed to the identified trend and the reasons for the trend.
3. Future actions: concrete steps and strategies to increase sales and improve operations
# Rules
1. Be sure to base your analysis solely on the data provided.
2. Please look over all sales data and perform the analysis.
3. Be sure to clearly state the basis for any inferences or deductions you make.