Skip to main content
Every page ChatGPT Deep Research opens returns only a capped window of text — roughly 5,700 characters median on first read, ~8,000 max — never the whole page. Peec measured this from recorded session logs (full study, Jun 2026). Every link on the page renders inline as a marker that spends characters from that same window, so navigation competes directly with your content:
  • Light navigation (under 20 links): ~78% of the first read is real content.
  • Medium (20–59 links): ~55% — nearly half the read goes to menus and markup; most pages sit here.
  • Heavy navigation (60+ links): ~33% — two-thirds of the read is chrome before the agent reaches your answer.
“Skip to main content” links do not help: activating one requires a click, and the agent never clicks — it reads the link as text and keeps going through the menu. Source order is what fills the budget, not on-screen position: a nav rendered at top by CSS but sitting late in the HTML costs nothing. The fixes (Jun 2026 findings): put the main answer in the first screen of HTML as plain text, keep the header slim, move mega-menus below the content in source order, and unblock OAI-SearchBot. A page the agent dismisses on first read almost never gets a second look — re-reads are earned by keyword find hits. Full pipeline mechanics: How Deep Research reads your site. Source: How ChatGPT Deep Research reads your site: What the logs reveal, Peec (David Konitzny), Jun 2026.