I love building things that move fast, but I hate when the thing that moves fast is invisible to search. Launched a Three.js/WebGL game on a fresh domain. The visible page is one <canvas> plus a JS bundle - basically nothing crawlable. My default assumption: never trust the render pass for anything that needs to rank. Put it all in the static shell.

What shipped at launch:

Static edge-served HTML with real descriptive copy + one semantic H1 (game mounts into it, not an empty root div)
JSON-LD VideoGame schema, canonical, OG/Twitter tags
sitemap + robots, submitted, URL manually queued
Sub-second LCP from Cloudflare edge, no CMP, nothing render-blocking

Result: indexed fast, #1 for its primary term in about two days, page one for a couple of variants.

Two things I want the nerds on:

Render-trust in 2026. Is "put everything rankable in the static HTML and treat the WebGL app as content Google ignores" still the correct default? Anyone actually get canvas-derived or post-hydration content indexed without a pre-rendered equivalent?
Indexation latency. Manual submit obviously drove the speed. But new-URL indexation feels compressed over the last few months vs the 2023-24 crawl-budget era. Matching your logs or noise?

Confounds up front: exact-match term, low competition, n=1. Not a method, just an observation.

A colleague pointed out that Google still doesn't render and index canvas. Another split it further: for classic Google, a thin exact-match page can rank on links with almost no body text. But for the AI answer layer, it needs a self-contained span of text that asserts what the game is in plain declarative sentences. So the static-shell instinct is not just still correct - it's more correct for that surface, as long as the copy is fact-shaped rather than keyword-shaped. Descriptive SEO filler ranks fine but rarely gets cited.

On indexation latency, someone noted that indexed and retrievable-by-an-answer-engine have quietly decoupled. A page can be indexed within a day and still never show up in a generated answer because there's no passage clean enough to lift. If retrieval visibility is what you actually care about, test that separately from the crawl log.

Happy to paste the shell markup + schema if useful. Show it pls.