Page speed isn't a direct AI-ranking dial, but the chain from "fast, light page" to "cited in AI answers" is real. Crawlers have to fetch, render, and parse your pages before anything you wrote can be retrieved, and slow or bloated pages leak at every step.
The mechanism: fetch → render → parse
- Fetch (crawl efficiency). Crawlers work within budgets. Slow responses mean fewer of your pages get crawled, and less often, so new or updated content takes longer to become eligible.
- Render (the JS trap). Many AI crawlers don't reliably execute JavaScript. If your key content and structured data only appear after JS runs, they may simply not exist for the crawler. Server-rendered HTML is the safe default.
- Parse (the byte budget). Crawlers process a finite amount of a page. Google's crawler, for instance, works within a roughly 2MB window, if bloated inline CSS/JS pushes your real content and JSON-LD past that, the important facts can be missed even though they're "on the page."
Why this maps to citations
AI citation happens at the passage level: a model lifts a clean chunk from content it could fetch and read. A page that's slow to fetch, renders late, or buries its facts below a wall of inline code is harder to extract from, so it loses to a leaner competitor that says the same thing more accessibly. Performance isn't a vanity metric here; it's an extractability prerequisite.
What to fix
- Server-render key content and structured data so they're in the initial HTML, not JS-injected.
- Keep the important stuff early and light: get your core answer, headings, and JSON-LD well within the byte budget; don't bury them under heavy inline code.
- Hit Core Web Vitals: fast loads help crawlers and users alike.
- Trim render-blocking bloat: minimise inline CSS/JS that pushes content down.
"We have great content" doesn't help if the crawler times out, can't render it, or never reaches the facts before its budget runs out. Speed and clean delivery are what turn good content into retrievable content.
This pairs with making sure you're reachable at all (is your CDN blocking AI crawlers?) and structured for extraction. Catching when a slow or JS-heavy template is quietly costing you AI presence is part of what a readiness audit, and ongoing monitoring with Buffy Intel: surfaces.