---
name: page-speed-regression-check
description: Work out whether and when a website got slower for real visitors, what likely changed that week, and whether it affected traffic or conversions. Use when someone says their site feels slower, their Core Web Vitals turned red in Search Console, or conversions dropped and they suspect speed. Needs the Caybl.ai connector (pagespeed_ tools; Google Analytics optional).
---

# Page speed regression check

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. Find when it changed

Call `pagespeed_core_web_vitals_history` for the whole site (bare domain, e.g. `example.com`) with `formFactor: "PHONE"` and `weeks: 25`. If the user named specific pages or templates, also run it for one URL of each.

The tool's `summary` lists the weeks a metric changed rating and the overall change. Each point is a rolling 28-day window, so a sudden change on a given day shows up gradually over about four weekly points; the change started around the **first** week that moved.

If nothing changed rating and nothing moved 10% or more, say so plainly: real visitors haven't seen a slowdown. Then check current status with `pagespeed_core_web_vitals` in case it was always slow.

## 2. Check whether it mattered

If Google Analytics is connected, call `run_report` with dimension `week` (or `date`), metrics `sessions`, `keyEvents` and `sessionKeyEventRate`, covering four weeks either side of the change. Compare the conversion rate before and after. Report correlation as correlation, not proof.

## 3. Find what changed

Run `pagespeed_audit` (mobile) on the most affected page and name the top 3 fixes. Ask the user what was deployed, installed or changed around the start week (theme update, new plugin, tag, chat widget, hero image, A/B test). Third-party code in the audit's diagnostics is a common culprit.

## 4. Report

One sentence: what got slower, by how much, starting when. Then the evidence (trend points), the business impact if GA was available, the likely cause, and the fixes. Keep it short.