GMO TECH Deciphers the 'Grammar of AI Recommendations' from 40,000 Domestic AI Searches

GMO TECH Inc. (President and CEO: Ryoichi Suzuki), a member of the GMO Internet Group, has published a research report analyzing 41,264 Japanese-language comparative AI responses from ChatGPT and Google AI Mode (*1), summarizing trends in how generative AI proposes products and services to users.

This survey revealed that generative AI does not propose just 'one' answer, but rather offers an average of 4.15 options based on different conditions. This signifies a major shift, allowing users to save the effort of comparing vast amounts of information themselves and making it easier to find a choice that suits them better.

(*1) Google AI Mode: AI summary and generative answer function in Google Search.

[Background of the Survey]

The proliferation of generative AI has dramatically simplified people's 'research' style. Previously, users had to open numerous websites listed in search results one by one to compare and consider. However, now, AI is increasingly summarizing and comparing information, with users directly referencing those answers.

Along with this change, for companies, not only 'search rankings' but also 'under what conditions and context they are introduced or recommended within AI answers' is gaining importance as a new visibility metric.

GMO TECH inductively analyzed and systematized the 'structure' of how AI derives 'just for you' from actual data in this new form of information gathering.

[Survey Summary]

■ AI Does Not Recommend Just 'One', Proposes an Average of 4.15 Options Suited to the User

For a single question, AI presents multiple options, averaging 4.15, such as 'If you want X, choose A; if you prioritize Y, choose B.' This allows users to easily find the option best suited to their situation. For companies, it becomes crucial to find opportunities to be included in one of these conditional branches (a line in the recommendation list).

■ Main Axes of Recommendation are 'Who It's For' x 'What It's Good At'

The primary perspectives for AI proposals were 'use-case specialization' (37.0% occurrence rate), indicating who the information is for, and 'feature specialization' (34.8%), indicating strengths. Furthermore, a combination of three axes – 'For beginners/if unsure x use-case specialization x feature specialization' – was observed in 5.0% of cases, showing a tendency for AI to intelligently select and structure products that best demonstrate their strengths according to the user's level and purpose.

■ 'For [Type of Person]' Patterns, Tailored to User Attributes, Outnumber 'Big Brand = Safe'

Cases where AI recommended based on traditional authority appeals like 'major company,' 'long-established,' or 'brand recognition' accounted for only 7.1% of the total. Instead, proposals tailored to user attributes and usage scenarios (23.8%), such as 'for beginners,' 'standard,' or 'for [type of person],' were approximately 3.4 times more common. This trend was particularly strong in categories where personal preferences vary, such as cosmetics, hair care, and fashion, highlighting the importance of articulating 'who it is for' regardless of company size.

■ Differences in Response Structure between ChatGPT and Google AI Mode

Differences were also observed in response structures across platforms. Google AI Mode tended to use multiple subheadings and comparison tables even for short queries, offering a broad and comprehensive set of options averaging 4.58. In contrast, ChatGPT's response structure varied more depending on the specificity of the question, typically proposing an average of 3.83 options. Therefore, when designing content, it is effective for companies to adopt a structure that can target both platforms, placing a clear conclusion at the beginning while developing the body by category and use case.

■ AI Recommendation Criteria Vary by Search Category

Analysis across 35 genres revealed that the points AI emphasizes (recommendation patterns) differ by industry.

• Financial Services: Economic sphere, point rewards, risk
• Cosmetics: Skin concerns, user attributes, appearance
• Telecommunications: Data volume, economic sphere
• SaaS (*2): Pricing structure, contract terms, number of users

(*2) SaaS (Software as a Service): A software delivery model used over the internet.

[Report Overview]

Report Name: How Brands Are Recommended in AI Searches: The 'Grammar of Recommendations' Revealed by Analyzing Approximately 40,000 Japanese AI Searches

Pages: 52 pages

Release Date: June 5, 2026 (Friday)

Report URL: https://gmotech.jp/semlabo/seo/blog/ai-search-grammar/

*The report page provides an overview of the survey results. A white paper (free) containing analysis by 35 genres and detailed data can be downloaded from the page.

[Survey Overview]

• Survey Target: Japanese AI responses from ChatGPT and Google AI Mode
• Target Queries: Questions containing comparative intent such as 'recommendation,' 'comparison,' 'price,' 'reputation,' 'disadvantages'
• Number of Analyzed Responses: 41,264
• Analyzed Genres: 35 genres
• Analysis Items: 34 types of recommendation perspectives (axes)
• Data Source: Ahrefs Brand Radar API (*3)
• Analysis Period: May 2026

(*3) API (Application Programming Interface): A mechanism for connecting different systems.

[Future Outlook]

In an era where users efficiently find products that suit them using AI, companies must adopt a comprehensive information design strategy (*4) that goes beyond 'just aiming for the top of search results' to consider 'how to have AI explain and propose their products as suitable for certain types of people.'

GMO TECH will continue its research and analysis on AI search, supporting improved consumer convenience and optimal information dissemination for companies.

(*4) LLMO (LLM Optimization): Optimization measures to increase the likelihood that a company's web content will be referenced by large language models like ChatGPT.

[About GMO TECH]

GMO TECH Inc. is a technology company that leverages AI and SaaS to support corporate sales growth and operational efficiency.

It develops customer acquisition support businesses in the SEO, MEO, and AI marketing fields, including the store customer acquisition DX SaaS 'MEO Dash! byGMO,' as well as operational efficiency support through the real estate DX SaaS 'GMO Rent DX,' affiliate advertising, internet media, and payment service businesses.

Leveraging its proprietary AI technology and in-house developed products, the company promotes the creation of systems where AI can continuously demonstrate value in business operations, supporting companies' digital utilization and sustainable growth.