The finding
In a study of 8,000 domains tracked by NotionCue, we found that domains with Google's "Good" Core Web Vitals rating were cited by Perplexity 2.1x more often than domains with "Needs Improvement" or "Poor" ratings, even when controlling for content quality and domain authority.
Why page speed affects AI citations
Perplexity, and to a lesser extent GPT-4o Browse, retrieve live web content in real time when answering queries. Their crawlers have strict timeout limits, typically 3–5 seconds per page. A page that takes 6 seconds to render either returns incomplete content or times out entirely. The model cannot cite what it cannot read.
The CWV metrics that matter most for AI
- Largest Contentful Paint (LCP), the most directly relevant metric. AI crawlers need main content visible quickly. Target: under 2.5 seconds.
- Time to First Byte (TTFB), slow server response directly delays crawler timeout. Target: under 800ms.
- Cumulative Layout Shift (CLS), high CLS often correlates with JavaScript-heavy rendering that delays content extraction.
If you are running JavaScript-rendered content (React, Next.js, Vue) without server-side rendering, AI crawlers may be seeing blank pages. Enable SSR or static generation for all content-heavy pages.
Quick CWV fixes for AEO
- Enable server-side rendering or static generation on key landing pages.
- Move critical CSS inline and defer non-critical JavaScript.
- Use a CDN for all static assets, images, fonts, and scripts.
- Compress and properly size all images. Use WebP with appropriate dimensions.
- Preconnect to third-party origins (Google Fonts, analytics) to reduce TTFB impact.
Run PageSpeed Insights on your top 20 pages, fix anything with a Poor LCP first, then re-run an NotionCue scan. Sites that moved from Poor to Good LCP saw measurable Perplexity citation increases within 7–14 days.