A Multi-Agent Query Fan-Out Analysis of Popular Google Searches in Brazil

• Vagner Figueredo de Santana NIC.br
• Diogo Cortiz NIC.br / PUC-SP
• Beatriz Rocha NIC.br
• Henrique S. Xavier NIC.br

DOI: https://doi.org/10.5753/webmedia.2026.30219

Resumo

Large Language Model (LLM) solutions are being more and more embedded into Web search engines, being query fanout one of them. While decomposing a query and using multiple sub-queries to one or more search engines can increase coverage and diversity, it is not clear how this impact biases in terms of language and location of these sources. In this context, this work employs a multi-agent system to perform a query fan-out study contrasting search engines vs. query fan-out using 10 different search engines via a SearXNG tool provided to agents. Results point out that larger models (e.g., deepseek-r1) tend to answer based on internal knowledge while smaller models invoke tools from 2 up to 4.4 times on average per task. In terms of URLs used by agents to answer queries, even tasks considering Brazilian context and terms in Portuguese, URLs from .𝑐𝑜𝑚 top level domains are used 6.1 times more than .𝑐𝑜𝑚.𝑏𝑟 and sources in English are selected over 3.7 times more than non-English sources. With the presented results, we propose the following bias mitigation strategies in agentic Web search: (1) controlling for dominant languages and sources, (2) employing context-aware search engines to better balance priority given to English content, (3) applying prompt engineering techniques to increase transparency and responsibility, and (4) testing different models+frameworks for revealing differences in terms of token usage, query augmentation, and tool usage for different solutions.

Palavras-chave: Artificial Intelligence, Agentic Web, Large Language Models, LLM-based Agents, Multi-Agent Systems, Query Analytics

Referências

Anthropic. 2025. How we built our multi-agent research system. Technical specification. [link]

Ricardo Baeza-Yates. 2020. Bias in Search and Recommender Systems. In Proceedings of the 14th ACM Conference on Recommender Systems (Virtual Event, Brazil) (RecSys ’20). Association for Computing Machinery, New York, NY, USA, 2. DOI: 10.1145/3383313.3418435

Asia J. Biega, Krishna P. Gummadi, and Gerhard Weikum. 2018. Equity of Attention: Amortizing Individual Fairness in Rankings. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval (Ann Arbor, MI, USA) (SIGIR ’18). Association for Computing Machinery, New York, NY, USA, 405–414. DOI: 10.1145/3209978.3210063

Conductor. 2025. What Is Query Fan-Out in AI Search and Why Does It Matter? Technical Report. Conductor. [link] Accessed: 2026-05-29.

CrewAI. [n. d.]. CrewAI. [link] Accessed: 2025-09-19.

Guilherme Fonseca, Gabriel Prenassi, Washington Cunha, Marcos Gonçalves, and Leonardo Rocha. 2024. Estratégias de Undersampling para Redução de Viés em Classificação de Texto Baseada em Transformers. In Proceedings of the 30th Brazilian Symposium on Multimedia and the Web (Juiz de Fora/MG). SBC, Porto Alegre, RS, Brasil, 144–152. DOI: 10.5753/webmedia.2024.241229

Tiago Gomes, André Regino, Rodrigo Caus, Victor Sotelo, and Julio Reis. 2025. Multi-Agent LLM Approach for Moderating E-Commerce Customer Service Responses. In Proceedings of the 31st Brazilian Symposium on Multimedia and the Web (Rio de Janeiro/RJ). SBC, Porto Alegre, RS, Brasil, 349–357. DOI: 10.5753/webmedia.2025.16162

Google. 2025. Year in Search 2025: Brazil. Google Trends. [link] Google Year in Search 2025 for Brazil; accessed: 2026-05-29.

Sayash Kapoor, Benedikt Stroebl, Zachary S. Siegel, Nitya Nadgir, and Arvind Narayanan. 2024. AI Agents That Matter. arXiv:2407.01502 [cs.LG] [link]

Matthew Kay, Cynthia Matuszek, and Sean A. Munson. 2015. Unequal Representation and Gender Stereotypes in Image Search Results for Occupations. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (Seoul, Republic of Korea) (CHI ’15). Association for Computing Machinery, New York, NY, USA, 3819–3828. DOI: 10.1145/2702123.2702520

LangGraph. 2025. LangGraph. [link]

Anthony Lee. 2026. How AI Platforms Search: Fan-Out Query Behavior Across Intent Types, Verticals, and Platforms. DOI: 10.2139/ssrn.6652059

Aldo Lipani, Gizem Gezici, Yücel Saygin, and Emine Yilmaz. 2021. Evaluation Metrics for Measuring Bias in Search Engine Results. Information Retrieval Journal 24, 2 (2021), 85–113. DOI: 10.1007/s10791-020-09386-w

Gabriel Prenassi, Rodrigo Souza, Henrique Sekido, Guilherme Fonseca, Marcelo Manzato, and Leonardo Rocha. 2025. Calibração de Sistemas de Recomendação com LLMs: Otimização de Prompts para Balancear Precisão, Diversidade e Justiça. In Proceedings of the 31st Brazilian Symposium on Multimedia and the Web (Rio de Janeiro/RJ). SBC, Porto Alegre, RS, Brasil, 103–111. DOI: 10.5753/webmedia.2025.15613

Ranqia Research. 2026. Quem Influencia o ChatGPT no Brasil? GEO Research Paper Vol. 01, Ed. 001. Ranqia. [link]

Jonathan J Ross, Ekaterina Khramtsova, Anton van der Vegt, Bevan Koopman, and Guido Zuccon. 2025. RARR Unraveled: Component-Level Insights into Hallucination Detection and Mitigation. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (Padua, Italy) (SIGIR ’25). Association for Computing Machinery, New York, NY, USA, 3286–3295. DOI: 10.1145/3726302.3730337

Stuart Russell and Peter Norvig. 1995. Artificial Intelligence: A Modern Approach. Prentice-Hall.

Vagner Santana, Reinaldo Ferraz, and Jeniffer Vieira de Deus. 2026. On the Limits for Web Accessibility Evaluation Using AI Agents. In Proceedings of the 23rd International Web for All Conference (W4A ’26). Association for Computing Machinery, New York, NY, USA, 187–197. DOI: 10.1145/3800424.3800445

Vagner Figueredo De Santana. 2024. Challenges and Opportunities for Responsible Prompting. In Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems (CHI EA ’24). Association for Computing Machinery, New York, NY, USA, Article 592, 4 pages. DOI: 10.1145/3613905.3636268

Vagner Figueredo de Santana, Sara E Berger, Heloisa Candello, Tiago Machado, Cassia Sampaio Sanctos, Tianyu Su, and Lemara Williams. 2025. Responsible Prompting Recommendation: Fostering Responsible AI Practices in Prompting-Time. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). Association for Computing Machinery, New York, NY, USA, Article 836, 30 pages. DOI: 10.1145/3706598.3713365

Ranjan Sapkota, Konstantinos I. Roumeliotis, and Manoj Karkee. 2026. AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges. Information Fusion 126 (Feb. 2026), 103599. DOI: 10.1016/j.inffus.2025.103599

Sander Schulhoff. 2026. Learn Prompting: Your Guide to Communicating with AI. Learn Prompting. [link] Open-source educational platform on prompt engineering and generative AI.

SearXNG Team. 2026. SearXNG: A Privacy-Respecting Internet Metasearch Engine. [link] Free and opensource metasearch engine; accessed 2026-06-03.

Cole Stryker. [n. d.]. Types of AI agents. [link]

Masaki Suzuki and Yusuke Yamamoto. 2021. Characterizing the influence of confirmation bias on web search behavior. Frontiers in psychology 12 (2021), 771948.

Fons Wijnhoven and Jeanna Van Haren. 2021. Search engine gender bias. Frontiers in big Data 4 (2021), 622106.

Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. ReAct: Synergizing Reasoning and Acting in Language Models. In Advances in Neural Information Processing Systems (NeurIPS). [link]