The zero pages for GPTBot and ClaudeBot on JS-only nav is the part that sticks. A lot of teams assume Google can render it so we’re fine, and miss that most AI crawlers never get past the initial HTML. Converting those links to real anchors and seeing GPTBot pick up roughly 250 pages in about 48 hours is a clean before/after. For the important paths, server-delivered hrefs still feel like the baseline, with JS as enhancement rather than the only discovery route.