Workflow: weekly Core Web Vitals check for a list of URLs, alert Slack when a page is slow

Sharing a simple pattern for keeping an eye on site speed without opening PageSpeed Insights by hand.

Nodes: Schedule Trigger (weekly) → Apify (“Run an Actor and get dataset items”) → IF → Slack (or Google Sheets “Append row” if you want a history).

Apify node settings: Actor birchwood-data/pagespeed-cwv-auditor, input:

{"urls": ["https://example.com", "https://example.com/pricing"], "strategy": "mobile"}

Each returned item is one URL with fields like performanceScore, labLcpMs, labCls, labTbtMs, plus real-user data when Google has it (fieldUrlLcpRating, fieldUrlInpRating, fieldUrlClsRating). In the IF node I use labLcpMs greater than 2500 OR performanceScore less than 50 (adjust to taste), and send the URL and the numbers to Slack.

Disclosure: I built that Actor, and it is a paid one (a fraction of a cent per URL, billed by Apify). You can do the same for free with an HTTP Request node calling Google’s PageSpeed Insights API (https://www.googleapis.com/pagespeedonline/v5/runPagespeed?url=...&strategy=mobile&key=YOUR_KEY) plus a Split In Batches node to stay under the quota. The Actor just handles batching, retries, mobile and desktop in one run, and a flat output. Happy to share the workflow JSON if anyone wants it.

Store page: Bulk PageSpeed Insights & Core Web Vitals Checker · Apify

Thank you for sharing the pattern, and for being open about the paid Actor alongside the free API route. Two small additions that may help anyone building this.

First, lab values such as LCP can vary from one run to the next, so alerting on a single weekly result may create noise. One option is to alert only when a page fails the threshold in two consecutive runs, or to store each run in Google Sheets and alert on the trend. Second, the IF condition could also cover INP (200 ms is the usual “good” limit) and CLS (0.1), since LCP alone does not describe the whole experience. Where field data is available, it is generally the better pass or fail signal, with lab data used to investigate the cause.

Thanks, Jessica, both points land. I’ll update the pattern to require two consecutive fails before Slack fires (or Sheets plus a trend), and gate on the field LCP, INP and CLS ratings when CrUX data exists, with lab data as the fallback. Happy to drop a revised IF example if useful.