New research into ChatGPT Search's internal mechanics has surfaced details that should change how content teams think about discoverability. The system, codenamed web.run internally, triggers a web search on roughly 34.5 per cent of user prompts. When it does search, it fires between 2.3 and 2.8 sub-queries per prompt using a fan-out architecture.
That fan-out pattern is important. When a user asks ChatGPT a complex question, the system does not run a single search query the way a human would in Google. It decomposes the question into multiple sub-queries, fetches results for each, then synthesises the answers. This means the content that gets cited is often answering a narrow, specific sub-question rather than trying to rank for the broad topic.
of ChatGPT prompts trigger a live web search via the [web.run](http://web.run/) system
The citation data is where it gets interesting for SEO teams. Only 12 per cent of URLs cited by ChatGPT Search appear in Google's top 10 results for equivalent queries. That is a fundamental divergence. The content that Google ranks highest is not necessarily the content that ChatGPT chooses to cite.
The research suggests ChatGPT Search favours content that is direct, factual, well-structured and answers specific questions without requiring the reader to parse through introductory paragraphs. Pages with clear data points, defined methodology and structured formats (tables, numbered lists, FAQ sections) appear to be cited at higher rates.
This creates a dual optimisation challenge for content teams. You need pages that rank in Google for traffic and pages that get cited by AI systems for visibility in a post-search world. Those are not always the same page, and they are not always optimised the same way.
For Australian businesses, the practical impact depends on your content volume and type. If you publish research, methodology content, benchmark data or technical guides, your citation potential in AI search is high. If your content is primarily brand storytelling or thought leadership without specific data points, AI systems are less likely to cite it.
Why it matters
AI-assisted search is not replacing Google tomorrow. But it is creating a parallel discovery channel where different content wins. The 12 per cent overlap figure means that 88 per cent of what ChatGPT cites is content that is not in Google's top 10. That is a massive surface area of opportunity for content that answers specific questions well.
What to do about it
Audit your top-performing content for AI citability. Does each page have clear, extractable answers to specific questions? Are your data points formatted in ways that an AI can parse and attribute? Add FAQ sections with direct answers, structure data in tables rather than prose, and ensure your methodology pages are detailed enough to be cited as sources. This is not a replacement for SEO. It is an additional layer that compounds with it.
