site: operator as a core retrieval tool. David Konitzny’s analysis of GPT-5.6 (Luna) fan-outs (full article, Jul 2026) found it in roughly 43% of query fan-outs, versus 0.004% in ChatGPT 5.5.
How it is used (Jul 2026 data):
- Mostly one domain per fan-out — only 9% of cases carry two or three site: commands. The operator narrows the information space to a trusted source.
- 84% of targeted domains are branded — the model goes straight to the domain offering the relevant product or service. Only 16% are non-branded secondary sources.
- Reddit is 71% of all non-branded site-operator domains (93% of the Social & Forums category); Trustpilot is 54% of the Review & Comparison category. arXiv appears among public institutions at ~9% of that category.
- Root domains beat URL paths. Root-domain targeting pairs with broad informational terms (review, official, service, comparison); specific URL paths pair with product-specific and pricing queries. Product attributes (size, model names, colors) are the most common accompanying terms at 12%.