From Generative Engine Optimization to Conversational Commerce
Intent orchestration, consumer world models, and agentic recommendation for multi-turn LLM experiences.
GUIDE — GEO-driven Understanding, Intent & Decision Experiences
GEO is IR's next chapter, not a departure from it
SIGIR has always been about connecting people to relevant information — through ranking, relevance modeling, and rigorous evaluation. Generative Engine Optimization (GEO) is the natural extension of that lineage into the LLM era: instead of a ranked list, users now receive a single synthesized, cited, conversational answer; instead of a one-shot query, they engage in an extended, multi-turn conversation in which intent forms and evolves turn by turn.
GUIDE 2026 takes GEO as its entry point into a broader, IR-rooted research agenda.
How do we retrieve, rank, and synthesize information faithfully inside a conversation? How do we track and satisfy user intent as it evolves across turns? How do we orchestrate agents, tools, and recommendation signals to carry a conversation from information-seeking through to a completed action — online or offline? And how do we measure the resulting consumer experience, from relevance and citation quality to satisfaction and NPS?
This agenda is increasingly urgent. LLM-based systems are no longer confined to information service — they now guide product discovery, assist offline purchase decisions in physical retail, and in many cases close the transaction directly, collapsing the boundary between search, recommendation, and commerce. At the same time, rapidly improving LLM memory capabilities open the door to genuinely individual-level personalization, which recommendation techniques — including lightweight adaptation methods such as LoRA — are needed to support at scale. And as LLM systems increasingly compose multiple specialized agent modules to serve a single user journey, recommendation becomes the connective tissue deciding which agent, tool, or piece of content to invoke next: an orchestration problem that is, at its core, an IR and recommendation problem.
A further, cross-cutting need is consumer simulation — building the shared datasets and simulation environments the community needs to study conversation generation and orchestration systematically. Our goal: better consumer experience and agent experience in LLM-native commerce, from information-seeking, to intent fulfillment, to transaction.
Five themes, one throughline: from retrieval to transaction
We welcome original research, position papers, and industry case studies across the following areas — reviewed double-blind.
GEO as the New Retrieval & Ranking Frontier
- Generative Engine Optimization (GEO): methods, metrics, and analytics
- LLM citation and attribution as a relevance / evaluation signal
- Ranking versus synthesis: rethinking relevance for generative answers
- Retrieval-augmented generation for multi-turn recommendation
Intent-Driven Multi-Turn Conversational Recommendation
- Intent modeling and tracking across conversation turns
- Feedback-based conversation orchestration for recommendation
- Relevance versus chain-of-thought reasoning in conversational search
- Conversational recommendation for e-commerce and offline / in-store journeys
Memory, Personalization & Consumer World Models
- Consumer World Models (CWMs): digital-twin representations of users
- Long-term memory architectures for individualized LLM experience
- Parameter-efficient personalization (e.g., LoRA) per user or segment
- Persona simulation for LLM-driven conversation
Recommendation-Guided Agent Orchestration
- Recommendation-guided agent and tool selection
- DAG-based workflow orchestration for multi-agent customer journeys
- Standardized lifecycle simulation for orchestrated agent experiences
Evaluation, Benchmarks & Consumer Simulation
- Consumer / user simulation for generating conversation datasets
- LLM conversation benchmark datasets — open resources for the community
- Customer experience and NPS-style metrics for LLM-driven commerce
- Measurement and statistical evaluation of multi-turn LLM conversations
Two 90-minute sessions
Session 1
From GEO to Conversational Recommendation- 00:00Opening remarks & keynote: GEO as the new frontier for IR and recommendation
- 00:15Four peer-reviewed papers on intent-driven multi-turn recommendation and consumer world models
- 01:00Panel: measuring relevance, satisfaction, and NPS in generative, multi-turn experiences
Session 2
Orchestration, Simulation & Breakouts- 00:00Lightning talks on agent orchestration and personalization engineering
- 00:30Breakouts: Evaluation & Metrics track and Orchestration & Simulation track
- 01:15Group synthesis: toward shared consumer-simulation datasets and benchmarks
Key dates for 2026
All dates are Anywhere-on-Earth (AoE) unless otherwise noted.
Who should submit — and attend
We invite original research papers, position papers, and industry case studies. All submissions undergo double-blind peer review, evaluated on technical rigor, fit with the five themes above, and potential to catalyze discussion, shared datasets, and collaboration across the SIGIR-AP community.
We expect 40–60 participants from academia — IR, recommender systems, NLP, and behavioral / consumer simulation researchers — and industry: search, e-commerce, and conversational recommendation platform builders working on GEO, agent orchestration, or personalization.
- Submission types
- Research paper · Position paper · Industry case study
- Review process
- Double-blind peer review
- Format & system
- To be announced
- Enquiries
- qkzhao@guyuai.com
Organizer & Program Committee
Dr. Qiankun Zhao
Founder & CEO, Guyu AI · Beijing
PhD, Nanyang Technological University; postdoctoral research, Pennsylvania State University. Prior experience at Microsoft Research, AOL, and Telefónica.
researchers across IR, recommender systems, HCI, and agentic systems, from institutions including Stanford, CMU, University of Amsterdam, and NIST.