OpenAI Boosts ChatGPT Search Precision With Site Filters

OpenAI has significantly increased the use of domain-specific filtering in its ChatGPT search queries, according to data from tracking firm Promptwatch. The company, which monitors chatbot responses for Generative Engine Optimization (GEO), observed a dramatic shift in how ChatGPT handles search queries following the rollout of the GPT-5.6 model. The percentage of tracked ChatGPT Search fanout queries utilizing the site: operator skyrocketed from its historical baseline of 0.3% to 0.5% up to 16% to 17% on August 8, 2026. This surge followed a brief dip to 0.15% between August 3 and August 5, which likely indicated a pre-launch experiment.

The timing of this change aligns with an official announcement from OpenAI on August 6, 2026. The company stated that it was updating GPT-5.6 Sol in Chat for Plus and Pro users to make the model more reliable with facts and to provide more focused answers. While OpenAI's system prompts remain obscured, analysis suggests the underlying search mechanism now functions similarly to a structured query like search(query, recency, domains) rather than injecting a literal site: operator directly into user prompts.

Additionally, Promptwatch reported on August 18, 2026, that ChatGPT has greatly reduced its reliance on Reddit during these searches. For web publishers and search engine optimization practitioners, this development marks a critical evolution in the emerging GEO space. Instead of scanning the open web indiscriminately, ChatGPT is increasingly narrowing its focus to specific, trusted domains to verify facts. Creators and businesses must now adapt to a landscape where LLM-driven search engines actively filter their sources, making direct domain authority and targeted indexing more critical than ever for appearing in AI-generated answers.