GPT does not read full web pages. It only sees tiny text slices through strict windows, and this guide shows exactly how that works.
People think GPT reads whole pages like a browser. It does not. This guide shows how GPT actually sees only tiny slices of pages, how it “opens” more text, and why it can never load a full article.
GPT does not browse the web. It does not load full pages, HTML, images, or full articles. When it performs a web search, it gets only a tiny object with a title, URL, a short snippet, and an internal ID. That is all it starts with.
If GPT needs more text, it must request it through an action called open, which returns only a slice of the page around a chosen line number. Each slice is small and capped. GPT can request more slices by opening different parts of the page. This creates a sliding window: it moves through the page in steps, never getting the whole thing at once.
Context settings like Low, Medium, and High only change how big each slice is. They never remove the limits. Even if GPT opens many windows, it still cannot rebuild or output the full page. It must summarise what it has seen because both retrieval and output have hard caps.
GPT does not have hidden access. It uses the same Web Search tool developers get. What it receives and what it returns are always limited, windowed, and controlled at every step.