ChatGPT search now uses the site:operator at scale

Background: Why a seemingly minor syntax shift matters

You might think, search engines have supported the site: operator for years. What is the big deal? The twist here is that it is not human users driving this change. It is the underlying system of ChatGPT Search itself. Third-party tracking from Promptwatch shows that after the GPT-5.6 update, the usage rate of site: in ChatGPT search queries jumped from a stagnant 0.3% to 0.5% straight up to 16% to 17%. This is not a user behavior trend. It is a deliberate, systemic strategy shift.

Breaking it down: From casting a wide net to precision targeting

The site: operator does one simple thing: it restricts search results to a specific domain. When an AI search system starts deploying it at scale, it means the model is no longer blindly scraping the entire web. It is actively focusing on high-trust or high-relevance sources. Combined with OpenAI's vague release notes for GPT-5.6, which mention improved factual accuracy and more focused answers, we can reasonably infer a structural change under the hood. The search tool likely evolved from a free-text query function to a structured call like search(query, recency, domains).

This reveals a deeper trend: AI search is moving from a find-everything approach to a find-precisely strategy. Early AI search acted like an eager but careless intern, grabbing whatever it could find. The current iteration behaves more like a seasoned researcher. It knows which sources are reliable, uses limiting parameters to narrow the scope, and actively filters out noisy platforms. Promptwatch data confirms a significant drop in Reddit citations, showing the system is now curating, not just collecting.

Trend insight: GEO is replacing SEO as the new traffic battleground

You might assume AI search is just a technical upgrade, but it is actually restructuring the content ecosystem. Traditional SEO taught us how to please Google's crawlers. Generative Engine Optimization, or GEO, teaches us how to get our content cited by AI when it answers user questions. As ChatGPT uses site: to actively whitelist sources, the websites included in that circle will see unprecedented visibility. Those left out could face a traffic cliff.

Promptwatch itself is a GEO tool, tracking AI responses to provide optimization advice. Its rise proves GEO is moving from theory to practice. In the near future, creators and developers will need to optimize for AI citation logic just as rigorously as they once did for search engine rankings.

Practical value: How to adapt and what to watch

For developers building apps that rely on AI search, source controllability is now critical. You can simulate the site: effect through prompt design or explicit tool calls to ensure the AI pulls from trusted domains. For content creators and webmasters, the playbook is changing. Chasing keyword rankings is less effective than positioning your content as a default source for AI. This means prioritizing authority, structured data, and ensuring your site is visible in the AI's indexing and retrieval layer.

The counterintuitive angle: Restraint is the real progress

We are used to AI getting smarter by knowing more. But this shift reminds us that the real maturity of AI search lies in knowing when to say less. By using site: operators and source filters, the AI is learning restraint. It is doing subtraction in an era of information overload. That discipline, not just raw capability, is what will define the next generation of generative search.