What do Web Vitals actually measure?
Core Web Vitals quantify human page experience: LCP (how fast the main content paints), INP (how fast the page responds to interaction), CLS (how stable the layout is). All three require a browser that renders and a user who interacts. AI crawlers do neither: OAI-SearchBot, GPTBot, ClaudeBot and PerplexityBot download HTML and parse it; ChatGPT-User fetches pages live against a user-facing timeout. A page can fail INP catastrophically and be perfectly retrievable — or ace Lighthouse and time out at the origin under bot load. The metrics answer different questions for different audiences.
So do vitals matter for AI visibility at all?
Indirectly, yes, through three paths:
- Shared infrastructure. The same origin speed that produces good LCP produces the low TTFB that batch crawlers reward with deeper crawl and live fetchers reward with successful retrieval — see site speed and AI retrieval.
- Classic rankings still feed retrieval. Page experience remains a (modest) Google ranking factor, and Google’s AI surfaces retrieve heavily from pages that already perform well in classic search.
- Human visits still convert. Roughly 68% of sessions may end zero-click (Ahrefs, February 2026), but the clicks that survive are high-intent; a page that frustrates them wastes the citation you earned.
Which metrics should you watch per audience?
| Audience | Primary metrics | Measured in |
|---|---|---|
| AI crawlers and live fetchers | TTFB, fetch success rate, timeout rate, redirect hops | Server logs — see log file analysis |
| Human visitors | LCP, INP, CLS | CrUX / RUM / Search Console |
| Both | Uptime, CDN cache hit rate, error rate | Infrastructure monitoring |
Where should you invest first?
- Edge caching and CDN — the single intervention that improves both columns of the table at once.
- Static HTML for citation-worthy pages — removes rendering dependence for bots and speeds LCP for humans; see JavaScript rendering and AI.
- Bot-aware WAF rules — verified AI crawlers must never be challenged or throttled; see robots.txt for AI crawlers.
- Fix redirect chains — costly to live fetchers and to crawl efficiency (crawl budget for AI).
- Then polish visual vitals — image optimization, layout stability, script discipline — for the humans who arrive.
Common pitfalls
First, Lighthouse theater: optimizing lab scores while origin TTFB under concurrent bot load stays multi-second — bots experience production, not lab conditions. Second, the reverse dismissal: ‘bots do not render, so vitals are irrelevant’ — visual experience still shapes the human conversions that justify the whole effort. Third, measuring only human RUM and having no per-bot latency data; the two audiences need separate dashboards. Fourth, third-party script bloat that hurts INP and adds render-blocking weight to HTML bots must download. Fifth, ignoring uptime during deploys — a recurring 503 window is invisible in monthly vitals averages but fatal to live fetches that happen to land inside it.
What does a sensible combined monitoring setup look like?
Two dashboards, one review. Dashboard one (machines): per-bot fetch success, TTFB percentiles and error rates from logs, plus uptime — the AI retrieval layer. Dashboard two (humans): CrUX field data for LCP, INP and CLS on key templates — the experience layer. Review both monthly in the same meeting, because they fail differently: the machine dashboard breaks suddenly (a WAF rule, a deploy, a cache misconfiguration) while the human dashboard degrades gradually (script creep, image bloat). Set alerts accordingly — paging thresholds on fetch failure, trend thresholds on vitals. When budgets collide, fund the machine side first if AI visibility is the goal, but treat the human side as deferred revenue: every citation that earns a click lands on the page experience you shipped. Both dashboards ultimately monitor the same infrastructure from two angles; build them on shared data where possible.
Where to go next
Go deep on bot-side speed in site speed and AI retrieval, instrument logs with AI crawler analytics, and watch the human side in Search Console in the AI era.