--- title: "Google research makes query fan-out an order of magnitude faster" url: https://seomadman.com/news/google-r4t-query-fan-out-research section: news language: en published: 2026-09-27T00:00:00.000Z modified: 2026-09-27T00:00:00.000Z author: Adam Hafez topics: ["AI search"] --- # Google research makes query fan-out an order of magnitude faster ## The short answer Google Research published R4T, a method that trains a small diffusion model to produce the varied sub-queries behind query fan-out in one fast pass. The paper reports fan-out latency cut by an order of magnitude on fashion and music benchmarks. Google has not said R4T runs in Search, where AI Overviews and AI Mode may use fan-out. ## Key takeaways - Google Research described R4T on 15 September 2026, based on a paper accepted at ICML 2026 and first posted in March. - R4T uses reinforcement learning once, offline, then trains a 53.9 million parameter diffusion model to generate fan-outs in a single pass. - Its training reward balances three goals, grounded in real content, diverse, and aligned with the original query. - The paper reports an order of magnitude less fan-out latency on large fashion and music benchmarks, not on web search. - Google's documentation already says AI Overviews and AI Mode may use query fan-out, but not that they use R4T. Google Research has published R4T, short for Retrieve-for-Train, a method that makes query fan-out much cheaper to run. The blog post, dated 15 September 2026, describes a paper accepted at ICML 2026. It is research, tested on fashion and music datasets, and Google has not said it runs in Search. ## What query fan-out is Query fan-out turns one broad request into several related sub-queries. Google's own example is "camping gear": a searcher wants a tent, a sleeping bag, a stove and a headlamp, not "ten slight variations of four-person tents." Google's [AI features documentation](https://developers.google.com/search/docs/appearance/ai-features) already says AI Overviews and AI Mode may use fan-out, issuing "multiple related searches across subtopics and data sources" to build an answer. The problem R4T tackles is cost. Asking a large language model to plan sub-queries takes many reasoning tokens, which Google says is "fundamentally at odds with the sub-second response times required by a production search bar." ## How R4T works The method moves the expensive part offline, in three steps: - **Train a fan-out model with reinforcement learning.** Open models (Gemma3-4B and Qwen3-4B) learn to write sub-queries scored on three balanced goals: grounded in real items, diverse, and aligned with the original query. - **Generate training data.** The trained model produces query and result-set pairs, with no human labels. - **Train a small, fast model.** A 53.9 million parameter diffusion model learns to produce the whole set of results in one pass, with no step-by-step reasoning. The balance matters: optimizing for grounding alone produced nonsense, and adding alignment alone produced near-duplicate paraphrases. Diversity, measured with the Vendi Score, closed both shortcuts. ## What the results cover The paper reports better result sets than strong baselines and fan-out latency cut "by an order of magnitude", on large fashion and music benchmarks. Those are curated outfits and music playlists, not the web. Neither the blog post nor the paper mentions Google Search, AI Mode or AI Overviews. Coverage describing R4T as production-ready search goes beyond what Google has published. ## Why we care Google says its AI features may already use fan-out, and research like this aims to make it cheaper to run. The practical lesson is unchanged by R4T: pages get cited for the sub-queries they answer well, so cover the related questions around a topic, not only its head term. The [AI answer readiness checker](/tools/ai-answer-readiness-checker) scores whether a page answers a question directly. Follow [AI search](/topics/ai-search) for whether Google ever confirms R4T. ## Frequently asked questions ### What is query fan-out? Breaking one broad query into several related sub-queries so the results cover different parts of what the searcher wants. Google says AI Overviews and AI Mode may use it, issuing multiple related searches across subtopics and data sources. ### Does Google Search use R4T? Google has not said so. The research blog and paper test R4T on fashion and music benchmarks, and neither mentions Google Search, AI Mode or AI Overviews. ### What should site owners do about query fan-out? Cover the related questions around a topic, not only the head term. A fan-out system looks for pages that answer the sub-queries, so a page that handles one facet well can be cited even when it would not rank for the broad query. ## Sources 1. [Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train](https://research.google/blog/bypassing-inference-bottlenecks-accelerating-complex-ai-search-with-retrieve-for-train/) - Google Research Blog (primary) 2. [Efficient, Property-Aligned Fan-Out Retrieval via RL-Compiled Diffusion](https://arxiv.org/abs/2603.06397) - arXiv (primary) 3. [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) - Google Search Central (primary) 4. [Google Announces New Query Fan-Out Framework: R4T-Diffusion](https://www.searchenginejournal.com/google-announces-new-query-fan-out-framework-r4t-diffusion/590700/) - Search Engine Journal