Query Fan-Out and Query Rewriting Appear in 324 Generative AI Patent Families, Including 44.1% of Those Published in 2026

Summary

Imagine someone asking an AI assistant for the best gluten-free bakery near downtown. Before it answers, a system like this may quietly run several searches, such as one for gluten-free bakeries and another for bakeries downtown. That splitting of one question into many is called query fan-out.

The idea has a long patent history. We found 2,404 patent families since 2018 that describe sub-queries, query rewriting or query expansion. What is new is the link to generative AI. In 2022, 1 of 204 such families published that year mentioned generative models. In 2026 so far, 115 of 261 do, or 44.1%.

Across the whole period, 324 families describe fan-out together with generative AI. A patent shows what was filed, not what is live in any product.

What we measured

We pulled every publication in Google Patents Public Data (patents-public-data.patents.publications) published from January 1, 2018 to September 17, 2026 that carries at least one CPC code in G06F16 (information retrieval), G06F40 (natural language processing) or G06N (AI and machine learning) and has an English abstract. That is 2,170,848 publications in 1,214,491 patent families. For each publication we kept the English title, the English abstract and up to the first 20,000 characters of the English claims. English claims exist for 488,406 publications, mostly US, European and international (WO) filings, so text matching is deeper for those offices than for Chinese, Korean or Japanese filings, which we matched on title and abstract.

A family counts as query fan-out when its text describes sub-queries or synthetic queries, query expansion, rewriting, reformulation or decomposition, or generating a plurality of additional or alternative queries from one input. It counts as generative when the text also names generative AI (large language models, LLMs, generative AI or generative models, GPT, foundation models or pre-trained language models). Use cases come from keyword groups in the same text, and a family can match several. Each family counts once, by first publication year.

Fan-out is old. Generative fan-out is new

We counted fan-out families each year and the subset that also describes generative models.

Total fan-out filings dipped from 342 in 2018 to 185 in 2023, then climbed to 422 in 2025. Most of that rebound comes from generative AI families. By first filing year, 158 generative fan-out families were first filed in 2024 and 59 in 2023.

Where generative fan-out is aimed

We grouped the 324 generative fan-out families by the settings their text names.

95 of the 324 (29.3%) also describe a conversation or chat setting, and 50 (15.4%) describe retrieval-augmented generation. Shopping shows up in 41 families, which matters for any business that sells products online.

Who files fan-out patents

Owners with five or more generative fan-out families:

IBM holds the most fan-out families overall (163) but only 5 that describe generative AI. Blank cells mean the owner is outside the top 15 for all fan-out families. Of the 324 generative families, 208 include a US publication and 122 a Chinese one.

What this means

If AI systems expand one buyer question into several searches, your page does not need to match the exact question. It needs to show up for the related searches around it: your service plus location, your service plus price, comparisons and common objections. An AI SEO consultant can map those related searches for your service so you know which pages are missing.

Build pages that cover a topic from several angles, and answer follow-up questions on the same page or on linked pages. A free AI visibility audit shows which related questions already surface your business in AI answers, and the ones that do not become your SEO for AI search plan.

Limits

A patent shows what was filed, not what is live in any product.

Patents usually publish about 18 months after the first filing, so families first filed in 2025 and 2026 are heavily undercounted, and 2026 publication counts cover January 1 to September 17, 2026 only.

Query expansion and rewriting have been studied for decades, so some families describe classic search techniques rather than AI assistants.

Keyword matching is broader in English claims, which are mostly available for US, European and WO filings.

Use-case groups are keyword based and overlap.

Frequently asked questions

Related studies

• Patents behind AI searchSearch Patents 2026: 39.9% Now Mention Generative AI
• Patents behind AI searchFreshness Shows Up in Only 3.6% of AI Ranking Patents
• Patents behind AI searchOnly 3.9% of AI Search Patents Describe Citing Sources
• Patents behind AI searchAI Search Patents: Top 10 Filers Hold Just 17.9%