Survey of recent patent cases related to AI prompts in Japan

📌 For those reading this article: Please also take a look at the IT Intellectual Property Portal, which provides one-stop support for permits, trademarks, and copyrights for IT companies.

List of patent cases related to AI prompts in the last 5 years

The following is a summary of patent cases related to "AI prompt" technology that were filed and registered in Japan within the last five years (mainly from 2019 to 2024). Targeting inventions related to "prompts" in the broad sense of the word, such as natural language prompt design (prompt engineering) and image generation prompts, the applicant, patent number (publication number/registration number), title of the invention, and summary (prompt-related features and effects) are organized in a table format.

*Among the above, "Unexamined Patent Publication" indicates an open publication (application publication), and "Patent No. ○○" indicates a patent publication (registered patent). [Number†Ln-Lm] in the citation indicates source information (J-PlatPat equivalent data, patent analysis article, etc.).

Detailed explanation of notable representative cases (5 cases)

From the list above, we will explain in detail the key points, technical features, and specific effects of the inventions for 5 representative cases that are particularly noteworthy.

SoftBank Group Corp. – Service name generation system according to user emotions (Patent No. 7597971)

Patent No. 7597971 obtained by the SoftBank Group is an invention that creates an optimal service name from the user's emotions and business parameters. For example, it analyzes the user's current emotional state (e.g. happiness, anger, sadness, etc.) from the user's facial expressions and tone of voice, and generates a prompt sentence (instruction sentence) by combining that emotion with information provided by the user, such as the industry, target group, and service content. By inputting this prompt into a large-scale language model (LLM), candidate service names will be output. Furthermore, this system automatically checks whether the generated name can be registered as a trade name in each country and whether it infringes on existing trademark rights, and presents the results in a list.

The feature of this invention is that it integrates creative name suggestions by AI and legal checks. It is possible to automatically generate unique names that reflect the user's emotions, and at the same time, it is possible to instantly check the availability of candidate names, which has the effect of streamlining branding operations. For example, if a user is expressing positive emotions, we can suggest a service name with a positive impression based on that, and we can also verify that the name does not conflict with existing trademarks, making it possible to quickly obtain a name that satisfies both sensibilities and legality.

Rakuten Group, Inc. – Product review summary generation system (Patent No. 7606028)

Patent No. 7606028 held by the Rakuten Group is an invention that balances information when automatically summarizing a large number of user reviews. Specifically, we collect a large amount of customer reviews and evaluation comments for a product, and input them as prompts into a generative AI (large-scale language model) to generate a summary statement. The key here is to instruct the prompt to ``Create a summary that reflects the ratio of positive to negative content in the post''. In other words, if the original review set contains half positive and half negative reviews, the generation AI will be controlled so that the ratio of positive and negative expressions in the generated summary is approximately 1:1. This is clearly stated in the patent claims, ``Obtaining a summary of submitted text adjusted to have a positive/negative ratio.''

This technical feature has the effect of preventing bias in the evaluation of summaries and obtaining fair and reliable surveys. Normally, products with many positive reviews tend to have positive summaries, but in this invention we intentionally maintain the ratio, so users can understand the balance of pros and cons just by looking at the summaries. As a result, consumers don't have to read all the lengthy reviews, but instead get an accurate overview that includes both positive and negative opinions about the product.

I'mbesideyou Co., Ltd. – Chatbot response system using dialogue history summary (Patent No. 7620369)

Patent No. 7620369 of startup company I'mbesideyou is Chatbot (dialogue avatar) technology that efficiently handles long conversation history. The new feature is that it retains a large amount of past message history between the user and the avatar AI, and automatically summarizes and includes old messages when used as prompts to LLM. For example, the full text of the most recent exchanges is entered in the prompt, but past logs that are a certain amount old are summarized briefly, and the algorithm compresses the summaries into shorter lengths as the past messages become older. In this way, the overall length of the prompt (number of tokens) is kept low while keeping important historical information intact. LLM uses the summarized history prompt as input to generate a new reply message.

This method makes it less likely that the relevance and quality of responses will deteriorate even if the chat AI maintains a long conversation. Large-scale language models have an upper limit on input length, but this invention overcomes this restriction through prompt engineering that selects and compresses past logs. As a result, even during a 100-turn interaction with the user, old topics are concisely summarized and only the key points are maintained, allowing the AI ​​to continually provide contextually appropriate responses. This is a very useful improvement for long chats, customer support bots, etc.

Tavern Co., Ltd. – Image generation prompt creation support system by tag selection (Unexamined Japanese Patent Publication No. 2024-120131)

Tavern Co., Ltd.'s Japanese Patent Application Publication No. 2024-120131 is an invention that helps even beginners easily create prompts for image generation AI. In this system, the AI ​​server that generates illustration background images and user terminals work together over a network. Many tags (keywords) such as "in the forest," "twilight," and "fantasy style" are presented in advance on the user terminal, and the user only has to select the tag that matches their image. The selected tags are sent to the server, and a prompt text incorporating the corresponding elements is automatically composed**. For example, if you select the "twilight" tag, a phrase such as "under the sunset sky" will be entered as a prompt. Then, an image generation AI (for example, an augmented reality background generation model) is executed based on the prompt, and a background image that matches your wishes is output.

The advantage of this invention is that even an amateur painter can obtain high-quality generated images with intuitive operations. There is no need to think of the prompt as a sentence from scratch and input it, and the appropriate sentence is formed behind the scenes using the GUI operation of selecting tags, which greatly reduces the user's burden. Tags are prepared in a variety of ways, such as composition, perspective, atmosphere, and story elements, and by simply selecting keywords that are close to the image in their mind, the user can create a detailed prompt that reflects them. As a result, it is now possible to have the generative AI draw the background illustration you want without any specialized knowledge, and it has the effect of lowering the barrier to prompt design in the field of game development and illustration production.

Cloud Shift Co., Ltd. – Prompt engineering technology for system development (Patent No. 7575148)

Patent No. 7575148, which CloudShift announced in October 2024, relates to a prompt engineering method for incorporating large-scale language models (LLM) into the software development process. In this invention, first, all development documents such as business system requirements definition documents and basic design documents are broken down into detailed items (nodes) and managed hierarchically. It automatically converts the contents of each node (e.g. screen specifications, API specifications, etc.) into a format that is easy for LLM to understand (Markdown or JSON format), and issues prompts for each node to be sequentially interpreted and coded by LLM. Partial source code obtained from LLM is stored linked to the original document hierarchy, and version control is also performed. This1:N automatic prompt generation and code integration management ensures that code throughout the system is generated in a consistent manner.

The effect of this prompting technology is drastically improving productivity in large-scale system development. Previously, programmers coded each part while reading the requirements document, but with the present invention, optimization prompts are automatically created from design information, and code generation and integration proceed all at once, which is said to dramatically reduce development man-hours. In fact, CloudShift says that with this technology, it aims to make effective use of LLM from requirements definition to testing and significantly reduce manpower. In short, by making full use of prompt engineering and leaving most of the system development to AI, human engineers can focus on the overall design and final adjustments, and this is a remarkable case that shows the possibility of innovating the software development process itself.

Reference materials/information sources: The content of the above patent cases is based on summaries and claims of patent publications and publications published in databases provided by the Japan Patent Office (J-PlatPat, etc.), as well as analytical articles on the patent industry. You can refer to the publication text and drawings on J-PlatPat using each patent number/publication number. In addition, the examination case regarding Case 38 ``Prompt sentence generation method'' published by the Japan Patent Office may also be helpful. As mentioned above, in Japan, a large number of inventions that focus on generative AI prompts have been filed and patented over the past few years, and it can be seen that companies are trying to expand the range of AI utilization by devising prompt designs.

🏢 Complete intellectual property strategy support for IT companies

Patents, trademarks, copyrights, trade secrets --- One-stop support for the intellectual property necessary for IT business.
From startups to listed companies, we will propose the optimal intellectual property strategy tailored to your business phase.

View IT intellectual property portal (general information) →

AUTHOR

Takefumi SUGIURA （杉浦 健文）

EVORIX Intellectual Property Law Firm　Managing Patent Attorney

Supports clients across IT, manufacturing, startups, fashion, and medical industries, covering patent, trademark, design, and copyright filings through trials and infringement litigation. Specialized in IP strategy for AI, IoT, Web3, and FinTech. Member of the Japan Patent Attorneys Association (JPAA), Asian Patent Attorneys Association (APAA), and Japan Trademark Association (JTA).

→ Patent Attorney Profile → Contact Us → Consultation Flow