Gemini and Copilot read JavaScript pages, ChatGPT, Claude and Perplexity do not

Glenn Gabe tested whether the major chat models can read JavaScript-rendered content when grounding via web search, and Chris Long shared the results on LinkedIn on 19 October 2026. Gabe used three sites: one fully client-rendered, one where main content loaded via JS, and one that rendered on scroll or event triggers.

Gemini read all three. Copilot read all three. ChatGPT failed on every test, usually acknowledging the URL and then flagging dynamic content it could not parse. Claude also failed on every test, and on one it referenced attempting a second fetcher. Perplexity failed on every test too, often opening with "Yes I can read the page" before admitting parts were unrendered.

Why the split tracks the crawler

Google runs Googlebot and Microsoft runs Bingbot. Both already render JavaScript at scale for search. Their chat products appear to inherit that capability. The other three models do not operate a rendering crawler of their own, and in Gabe's tests they read what amounts to the static HTML.

What to check on your own templates

Before acting, run the test on pages you care about. Ask each model to quote or summarise content that only appears after JS executes, and compare with what the raw HTML contains. If ChatGPT, Claude or Perplexity can only see your nav and footer, server-side rendering or prerendering of the main content is the fix, and the same change helps every retriever including the two that already cope.

This is a snapshot of five products in October 2026. Capabilities here have shifted fast, so re-run the check on your own URLs rather than assuming today's result holds.