---
name: core-web-vitals-audit
description: Find the highest-traffic pages on a website that fail Google's Core Web Vitals for real visitors, explain why, and produce a prioritised fix list. Use when someone asks whether their site is fast enough, which pages are slow, why their PageSpeed score is low, or what to fix first for SEO page experience. Needs the Caybl.ai connector (pagespeed_ tools; Google Analytics or Search Console optional).
---

# Core Web Vitals audit

Uses the Caybl.ai connector. If its `pagespeed_` tools aren't available, tell the user to add Caybl.ai (https://caybl.ai) as a connector and stop.

## 1. Pick the pages that matter

- If Google Analytics is connected: call `run_report` for the last 28 days with dimension `landingPage` and metric `sessions`, ordered by sessions, limit 10. Build full URLs from the site's domain.
- Otherwise, if Search Console is connected: use its top pages by clicks for the last 28 days.
- Otherwise: ask the user for their domain and up to 10 important URLs (home, top product or service pages, pricing, a blog post).

Keep the share of traffic for each page; you'll use it to prioritise.

## 2. Check real-user data first

Call `pagespeed_core_web_vitals` with all the URLs at once, `formFactor: "PHONE"`. Then once more with just the domain (whole site) for `PHONE` and `DESKTOP`.

- A page **fails** when LCP > 2.5 s, INP > 200 ms or CLS > 0.1 at the 75th percentile.
- Note pages that fell back to "whole site" data. Don't present those numbers as that page's own.

## 3. Find the causes

Run `pagespeed_audit` (mobile) on the **failing pages only**, worst and highest-traffic first, at most 4 audits. Pages built from the same template usually share causes; audit one per template.

## 4. Report

Lead with the verdict in one sentence: does the site pass on mobile, and how much traffic lands on failing pages.

Then a table: page · sessions share · LCP · INP · CLS · pass/fail.

Then **the fix list**, merged across pages and ranked by (estimated time saved × traffic share of the pages it affects). For each fix: what to do in plain words, which pages, estimated saving, and who usually does it (developer, content editor, or a plugin/theme setting).

Rules:
- Real-user (field) numbers decide pass/fail. The lab score only explains causes. Never say a page "fails" because its Lighthouse score is low.
- Say which date range the field data covers (the tool returns it).
- Don't recommend generic advice the audit didn't flag.