What Search Queries Does Generative AI Use for Insurance Consultations? Optyino.ai Analyzes 10,528 AI Responses, 20.1% Include Brand Names

Optyino.ai is a GEO/AIO/LLMO platform that visualizes and optimizes brand visibility in the age of AI search. Leveraging accumulated AI response logs, the company analyzed how often generative AI includes brand names in search queries when answering questions about insurance consultations and financial planning (FP).

This study targeted 12 prompts asking for insurance consultation services, FP advice, and investment NISA providers—such as 'Recommend an FP consultation service,' 'Where can I consult about retirement funds?' and 'Which online insurance consultation service is recommended?'—analyzing 10,528 AI responses collected between August 11, 2025, and April 24, 2026.

Survey Summary

The analysis revealed that out of 15,727 extracted search queries, 3,161 (20.1%) were brand-inclusive queries—those containing one or more names from a fixed brand list.

However, search query data was only obtainable from four models: GPT-5 mini, GPT-5.2, Gemini 2.5 Flash, and Claude Sonnet 4.5. Significant differences were observed: GPT-5 mini had 39.8% brand-inclusive queries, GPT-5.2 had 30.9%, while Gemini 2.5 Flash had only 2.3% and Claude Sonnet 4.5 had 0.0%.

At the search execution level (a bundle of searches performed to generate one AI response), 77.9% of GPT-5 mini executions generated at least one brand-inclusive query, with GPT-5.2 reaching 62.5%.

Among individual brands, the Japan FP Association, an industry body, was mentioned most frequently with 970 instances, slightly surpassing Hoken no Madoguchi (867 mentions), an insurance consultation service.

The top three brands—Japan FP Association, Hoken no Madoguchi, and Hoken Minaoshi Honpo—accounted for 2,358 mentions, representing 65.0% of all 3,627 brand-matched instances.

Background of the Survey

Recently, more people are using generative AI to find places for insurance or FP consultations. An increasing number of generative AI models now operate by sending search queries to search engines during response generation, then organizing and presenting information based on the results.

The term 'search query' here refers to the search terms sent by AI to search engines during response creation. Whether specific service or company names are included in these queries from the outset directly affects how easily a brand can gain visibility through AI.

Therefore, Optyino.ai analyzed how often brand names appear in search queries when users ask generative AI about insurance, FP, or investment NISA consultations.

The study also compared selection tendencies across three categories: insurance/FP consultation services, securities firms, and industry associations (group bodies).

Key Findings

### 20.1% of Insurance Consultation Queries Include Brand Names—Large Gap Between GPT and Gemini/Claude Models

Out of 15,727 extracted search queries, 3,161 (20.1%) included brand names.

However, data was limited to four models, with significant variation: GPT-5 mini had 39.8%, GPT-5.2 had 30.9%, while Gemini 2.5 Flash had 2.3% and Claude Sonnet 4.5 had 0.0%.

The gap between GPT-5 mini and Claude Sonnet 4.5 reaches 39.8%, indicating that even with identical questions, different AI models construct search queries in vastly different ways.

This difference may stem from design choices—whether models actively insert proper nouns or stick to generic expressions when generating queries.

### Despite 39.8% at Query Level, 77.9% of GPT-5 mini Executions Include Brand Names at Execution Level

While the brand-inclusive query rate for GPT-5 mini is 39.8% per query, 77.9% of search executions include at least one brand-inclusive query.

- GPT-5 mini: 39.8% (query), 77.9% (execution) - GPT-5.2: 30.9% (query), 62.5% (execution) - Gemini 2.5 Flash: 2.3% (query), 6.6% (execution) - Claude Sonnet 4.5: 0.0% (query), 0.0% (execution)

There is a large gap between query-level and execution-level rates. GPT-5 mini jumps from 39.8% to 77.9%—nearly doubling—when measured per execution.

GPT-5.2 shows a similar pattern: 30.9% at query level vs. 62.5% at execution level—about twice as high. This indicates that while not all queries in a single execution contain brand names, most executions include at least one brand-inclusive query.

In other words, when AI generates responses for insurance or FP consultations, it rarely avoids brand names entirely at the search entry point. In most cases, some brand name appears as a candidate in the search queries.

Gemini 2.5 Flash and Claude Sonnet 4.5 remain low even at the execution level (6.6% and 0.0%, respectively), making model differences even more pronounced at the execution level than at the query level.

### By Category: GPT-5 mini Uses Association Bodies (e.g., Japan FP Association) at Over 70% the Rate of Insurance/FP Services

For GPT-5 mini, the match rate for association bodies (15.2%) reaches over 70% of that for insurance/FP services (21.6%).

- Insurance/FP services: 21.6% - Association bodies: 15.2% - Securities firms: 3.0%

Despite inquiries being about insurance or FP consultations, GPT-5 mini frequently uses industry associations like the Japan FP Association in its search queries.

GPT-5.2 also shows a 7.6% match rate for association bodies—lower than its 23.1% for insurance/FP services, but still significant.

Gemini 2.5 Flash shows nearly zero for association bodies (0.0%) and securities (0.1%), while Claude Sonnet 4.5 shows 0.0% across all categories.

Although the Japan FP Association is not a direct insurance consultation service but an industry body, it provides abundant information from an official standpoint,