.svg){kind=logo}
To utilize generative AI effectively, the way you write "prompts" is extremely important. A prompt is an instruction given to the generative AI, and by creating appropriate prompts, you can output high-quality text, images, program code, and more.
In this article, we will provide an easy-to-understand explanation ranging from the basic overview of generative AI prompts to tips for writing them well and examples of their use in business scenes. Additionally, based on cautions when entering prompts, we will introduce hints for prompt creation that can be safely utilized in business.
This content is useful for those who want to maximize the use of generative AI and streamline their operations and content production.
|
【Table of Contents】 |
1. What is a "Prompt" Important for Generative AI?
A prompt refers to the input instructions or information given to generative AI. Simple examples include prompts such as "Summarize the following report" or "Create an image of a business person in a meeting."
In recent years, generative AI, starting with OpenAI's ChatGPT, has significantly improved its natural language understanding capabilities each time the built-in AI model is updated. Consequently, modern generative AI can generate high-quality text and images that flexibly reflect the user's intent and nuances, even with prompts similar to conversations between humans.
At the same time, the range of what can be generated has expanded dramatically, and the accuracy, originality, and creativity of the generated content change significantly depending on the precision of the prompt. In other words, setting what kind of prompt determines the quality of the generated output.
Against this background, along with the development of generative AI, the importance of technology and know-how for improving prompt quality is increasing.
The role of prompts is "extraction," not "command"
A prompt functions as a "blueprint" for instructing generative AI on specific tasks and output content. Through prompts, generative AI accurately reads the user's intent and outputs text, images, or videos by flexibly combining the content of the vast data it holds.
To use an analogy for the relationship between generative AI and prompt design, an LLM (Large Language Model) is like a pantry containing all sorts of ingredients from all times and places. Depending on the combination, quality, order, thawing method, cooking method, and presentation of the ingredients taken from there, dishes of completely different levels are completed.
And the prompt is like a recipe that instructs everything from the combination of ingredients to the presentation.
In other words, a prompt is considered an important element not just for conveying commands, but for "extracting" necessary information from the rich knowledge and pre-trained data possessed by generative AI.
What you can do by writing prompts
By designing and utilizing prompts well, generative AI can output various things.
Below are examples of specific output content:
|
In other words, the accuracy, originality, and clarity of the output in generative AI can vary greatly depending on the prompt design (prompt engineering).
Steps for creating generative AI prompts
Prompts become of higher quality by creating them according to the following steps:
|
By following the above steps, you can create prompts that can accurately elicit the expected results from generative AI.
2. Tips for Writing Good Generative AI Prompts
For effective prompt creation, several techniques are needed rather than just giving instructions. Below are tips for writing good generative AI prompts.
Give context
Generative AI reads the overall background and context from the content of the prompt. Therefore, for generative AI to provide accurate answers through prompts, it is important to include background information and specific data (= context) within the prompt.
In particular, in generative AI systems equipped with RAG (Retrieval-Augmented Generation), combining information retrieval from external knowledge sources enables answers based on more accurate and up-to-date information.
Examples of context that can enhance the generative capabilities of generative AI are below.
| Type of Context | Example |
| Role of Generative AI | You are a professional web marketer. You can refer to the latest industry data and success stories using the RAG system. |
| Background Information | Our company is a small-to-medium enterprise with 50 employees that develops B2B software. We face challenges in acquiring customers online. |
| Expected Output | Please include implementation steps and expected effects in each proposal. |
By providing detailed context like the above, the AI can understand the content of the prompt more deeply and generate higher-precision answers.
Give specific instructions
No matter how sophisticated the natural language understanding power of a generative AI model is, abstract instructions similar to the feeling of giving instructions to humans will limit information, resulting in only ambiguous answers.
Therefore, when creating prompts, the point is to state specific and clear instructions instead of abstract ones.
|
【Non-specific Example】 |
|
【Specific Example ①】 |
|
【Specific Example ②】 |
Give examples of ideal answers
By including specific answer examples or formats in the prompt, the AI will generate output using them as a model.
For example, if you want to create a column of corporate cases, it is good to provide about 1 to 3 exemplary corporate cases as follows:
| 【Prompt Example】 Based on the writing style, volume, and information elements of the example below, investigate security risk cases of generative AI and create a report: In July 2024, a situation occurred where an employee of XXXX Electronics mistakenly entered internal confidential source code into ChatGPT, resulting in information leaking externally. |
By presenting examples of ideal answers, the AI can read the patterns and habits of the given examples, which is useful when you want to generate text with a consistent format or taste without giving detailed conditions.
Specify necessary conditions
In prompts, it is also important to clearly indicate specific requirements under what conditions the output content should be generated.
Below are examples of conditions that are good to specify:
- Writing style in output (e.g., polite/desu-masu tone)
- Output format (e.g., table, bullet points)
- Target audience (e.g., CHRO of a domestic listed company)
| 【Prompt Example】 Please create a market research report for XXXX under the following conditions. # Conditions Polite tone (desu-masu) Use tables and bullet points frequently For CHROs of domestic listed companies |
By setting conditions as above, you can effectively narrow down the AI's answer range and obtain an output closer to the desired results. This is recommended for business emails or presentation materials for companies where a fixed structure or wording is required.
Write constraint conditions you don't want in detail
In addition to "absolute conditions" that summarize what you want done, you can obtain an output closer to desired results by clearly stating "constraint conditions" that you do not want output as follows:
- "NG words" such as names of competitor products
- "Content to avoid" such as discriminatory elements or topics currently causing controversy on SNS
- "Diction and tone" inappropriate for your own content
| 【Prompt Example】 Please create a marketing article for a new product. # Constraint Conditions Do not use specific product names, including our own or others' Diction using "da" or "deshou" is prohibited |
Constraint conditions lead to ensuring consistency of output and improving accuracy. They are particularly effective when producing content based on brand guidelines or internal rules.
3. [By Business Scene] Examples of Generative AI Prompts
In business scenes, text generation utilizing generative AI demonstrates significant effects in terms of efficient communication, information organization, and data management. Here, we introduce examples of prompts for text generation that can be used universally in business scenes.
Creation of business emails
Business emails are often in a fixed format and are a field where it is easy to create with generative AI.
For example, when generating business emails tailored to the purpose, such as arranging a meeting, sharing proposal content, or contacting a customer, prompts like the following are effective:
| 【Prompt Example】 Please write an email to Mr. △△ of 〇〇 Company requesting a re-adjustment of next week's meeting schedule. In formal style, please present 3 candidate dates. |
The point of prompts for business emails is to clearly indicate the "recipient, purpose of the email, and style." If there are business emails you create frequently, by formatting the prompts, you can generate them more smoothly.
Summarization of text
Summarizing existing text is a specialty of generative AI. Even vast documents or business reports spanning several pages can be summarized quickly and accurately.
For example, when summarizing the contents of a long report or meeting minutes, a prompt that extracts key information and summarizes it concisely as follows is effective:
| 【Prompt Example】 Please summarize the following text, including the main points. |
If you want a detailed summary, the point is to provide keywords you want to narrow down, such as "Summarize centering on 〇〇."
Data ingestion of business card images
The organization of business card data that exists in large quantities within a company can be easily managed by leaving it to generative AI.
For example, a prompt to extract information such as company name and phone number from a business card image and summarize it in a table is as follows:
| 【Prompt Example】 Please recognize multiple business card images within the image individually, extract data, and output in table format according to the following steps: 1. Extract the following information from each business card: - Name - Company Name - Title - Phone Number - Email Address - Address - Website (if present) 2. Output the extracted data in a table format like the following: | Name | Company Name | Title | Phone Number | Email Address | Address | Website | |--------|----------|--------|--------------|----------------------|--------|------------------| | | | | | | | | 3. If data is unclear or unreadable, enter "Unknown" in the corresponding cell. 4. Once data extraction for each business card is complete, write "Data Extraction Complete" under the table. 5. If there are cautions regarding image quality or business card design, set a "Remarks" column at the right end of each row and comment concisely. Now, please begin processing the uploaded business card image. |
The point of the prompt for ingesting business card image data is to have information extraction and data formatting executed step-by-step.
Also, clearly stating the error response method to prevent stopping due to errors during reading and specifically specifying the output format are important tips.
4. Cautions When Entering Prompts into Generative AI
When utilizing generative AI in business, sufficient consideration for security and legal risks is required. Since the input content of prompts affects the outputs, observing cautions at the input stage leads to preventing copyright troubles and reducing information leakage risks.
Do not include confidential information
When entering prompts into generative AI, from security and legal perspectives, it is basic not to include information that would be problematic if leaked externally, such as corporate confidential information, personal information, or customer information.
This is because some generative AI services utilize data input through prompts for AI learning and feedback. Therefore, when utilizing generative AI, it is basic to limit content to that intended for use in public spaces.
Consider copyright
Content output by generative AI may also have latent issues related to copyright. For example, instructing it to refer to existing works during prompt entry constitutes copyright infringement.
To avoid risks of copyright infringement, you must always confirm the source for information to be quoted/referenced or images to be utilized as motifs, and strive for use that complies with copyright law.
| Related Article: Does generative AI constitute copyright infringement? Explaining the problematic conditions, examples, and countermeasures! |
5. Summary
In generative AI, prompts are not mere input sentences; they play a major role as a blueprint for accurately conveying the user's intent and eliciting expected output.
The basics of prompt creation are three: "giving context," "giving specific instructions," and "specifying conditions." By firmly grasping these points, generative AI can provide results close to the ideal, no matter what kind of output is required.
Author
Latest Articles
-1.webp)
-2.webp)
