Hidden elements hurt AI citations when they conceal content the schema markup claims — Google's structured-data guidelines require that marked-up content be visible on the page, and answer engines that grade attribution confidence will down-weight sources whose visible page contradicts their machine-readable claims.
The rule is not about all hidden elements — accordions and tabs are fine as long as the content is present in the DOM and reachable without an interaction gating the render. It is about content that lives only in a JSON-LD block, or in JavaScript-hydrated state, or behind a login while the schema claims it is a public answer.
Two concrete failure modes. First, a FAQPage schema block listing five Q&A pairs when the visible page only shows two — Google's Rich Results Test does not catch this, but manual reviewers and answer engines can. Second, a Person author node claiming a byline that never appears on the visible page, which trips E-E-A-T grading in both classic and generative search.
The fix is straightforward: visible content is the source of truth, schema mirrors it exactly, and any accordion/tab UI ships fully-rendered HTML with CSS-based show/hide, not JavaScript-conditional insertion.
Design accessibility and AEO together — a page that a screen reader can traverse without JavaScript is a page an AI crawler can extract from.