Avg performance score: most- vs least-cited. 76.5 vs 74.9.
Faster, better-performing sites are cited somewhat more: the most-cited domains average a 76.5 performance score against 74.9 for the least-cited. Site speed tracks citation frequency in the expected direction, but the effect is small.
Site performance score vs citation frequency
Grouping every cited domain by how many of AmICited’s tracked-prompt responses cited it, and joining Google’s real-user Core Web Vitals data, a consistent pattern appears across the tiers. Cited 10+ times (659 domains with data) sit at one end and cited 1–2 times (4,313 domains) at the other. The direction is the same for every health metric we can measure — pass rate, server response time, and performance score — which is why the (modest) signal is credible rather than noise.
The underlying numbers
| Citation frequency | Domains w/ CrUX data | Pass Core Web Vitals | Median TTFB | Avg perf score |
|---|---|---|---|---|
| Cited 10+ times | 659 | 60% | 804 ms | 76.5 |
| Cited 3–9 times | 1,584 | 58% | 893 ms | 74.9 |
| Cited 1–2 times | 4,313 | 57% | 910 ms | 74.9 |
Ready to Monitor Your AI Visibility?
Track how AI chatbots mention your brand across ChatGPT, Perplexity, and other platforms.
What this means for AI search visibility
Technically healthy pages are cited somewhat more often, but the effect here is small — site health looks like a supporting factor, not a primary driver of whether AI engines cite you. Content relevance almost certainly matters more (see the source- and topic-level reports). The practical read: fixing Core Web Vitals and server response time is worth doing — it removes a mild headwind and helps users regardless — but it will not, on its own, move you into an engine’s cited sources. Treat it as table stakes, then compete on relevance.
What PageSpeed score actually measures (and what AI engines care about)
Google’s PageSpeed performance score is a synthetic metric — it simulates a page load on a throttled connection and measures how quickly the page becomes visually complete and interactive. It is a useful proxy for real-world performance, but it is not the same thing as what an AI crawler experiences.
AI crawlers like GPTBot (OpenAI), PerplexityBot, and Google-Extended do not render pages visually. They fetch the HTML , extract the text content, and process it through natural language pipelines. The aspects of PageSpeed that matter most to them are:
Server response time (TTFB ): How quickly the server delivers the first byte of HTML. A slow server means the crawler waits longer for content, which can trigger timeouts or incomplete fetches.
HTML size and complexity: Pages with bloated HTML (excessive inline styles, large DOMs) take longer to parse, even for non-rendering crawlers.
Availability and reliability: A page that returns 5xx errors or times out under load will not be crawled successfully, regardless of its content quality.
The aspects of PageSpeed that matter least to AI crawlers are: image optimization, JavaScript execution time, layout shift metrics, and interactivity measures. These are important for human visitors but irrelevant to text-extracting crawlers.
This means that if you are optimizing for AI visibility specifically, you should prioritize server-side performance (TTFB, HTML delivery, error rates) over front-end optimizations (image compression, lazy loading, CSS minification). The two are correlated — a well-optimized site tends to be good at both — but if you have limited engineering time, focus on the server side.
Stay Updated on AI Visibility Trends
Get the latest insights on AI mentions, brand monitoring, and optimization strategies.
The 1.6-point gap in context
A 1.6-point difference in PageSpeed score (76.5 vs. 74.9) is remarkably small. To put it in perspective, switching from a shared hosting plan to a dedicated VPS can swing a PageSpeed score by 10–15 points. Enabling a CDN can add 5–8 points. The fact that the most- and least-cited domains are separated by only 1.6 points means that performance is not a meaningful differentiator within the cited set.
This is consistent with the broader finding across all of our Core Web Vitals analyses: AI engines are selecting sources primarily on content, not performance. A page that is slow but uniquely authoritative on a topic will be cited over a fast but generic page. The practical implication is that you should not obsess over PageSpeed scores once you are in the 70+ range. The ROI of performance work for AI visibility drops sharply at that point.
Practical recommendations
Get your PageSpeed score above 50 first. If you are in the “poor” range (0–49 on mobile), you almost certainly have server-side issues that are affecting both human visitors and AI crawlers. This is the highest-ROI performance work.
Target 70+ as your “good enough” threshold. Once your score is in the 70s, you are in the range where AI-cited pages live. Further improvements will have diminishing returns for AI citations.
Focus on the server-side metrics within PageSpeed. Pay attention to TTFB, server response time, and the “Reduce initial server response time” audit. These are the aspects of PageSpeed that most directly affect AI crawler experience.
Don’t chase 90+ for AI visibility. A score of 90+ is impressive and will benefit your human visitors, but our data shows no evidence that it provides additional citation benefits beyond what you get at 70+.
Methodology
This is an association among the pages AI already cites, not proof of cause: it is computed by joining Google CrUX / PageSpeed field data to how often each domain was cited across AmICited’s 1,905 tracked prompts. CrUX data was available for 6,556 of 8,845 cited domains (74%). Domains are grouped into how many responses cited them; within each group we average the site-health metric. A domain “passes Core Web Vitals” when the majority of its audited URLs pass Google’s LCP /INP /CLS thresholds in real-user (CrUX) data; TTFB and performance score are averaged likewise. Pages without sufficient CrUX data are excluded. Because AmICited’s prompts skew toward SaaS, e-commerce and support topics, these figures describe the sites cited for that kind of query. The relationship is real but modest, and correlational — we are not claiming faster pages cause more citations.
