Most Core Web Vitals monitoring ends at one number per page: the 75th percentile of LCP for /checkout, the median INP for your dashboard. That figure is useful for comparing releases. It tells you almost nothing when a customer writes in saying the page hangs and your dashboard is green.
Percentiles compress thousands of visits into a single value. The slow sessions — the older phone on a train, the visit where a third-party script held the main thread, the layout that jumped under the cursor — fall into the tail and stop being addressable. You cannot watch a percentile. You can watch a session.
This guide is about running Web Vitals per session: record LCP, INP and CLS against the visit that experienced them, then use the recording, the errors and the clicks around the metric to fix the cause instead of guessing.
What a page-level average cannot tell you
Google’s published thresholds give you a vocabulary: LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1. What a page-level p75 gives you is a description of the middle of your traffic. Two things stay hidden: which sessions sit in the tail, and what those sessions were doing at the moment.
Slow sessions cluster
A slow session is rarely random. It repeats: one route, one device class, one browser, one release, one network condition. A p75 per page mixes all of those together, so a route that is slow only on mid-range Android, and only after a marketing script loads, still looks acceptable in aggregate. Per-session data lets you filter to that group and see it as a group.
Support tickets come from the tail
The average visitor does not complain about a 2.1-second LCP. The visitor who hit the nine-second session does, and by the time the ticket arrives the context is gone. Keeping vitals attached to the session preserves that context: device, route, network, errors, clicks and the recording.
How per-session vitals are recorded
Prynt Replay records Web Vitals with each session and shows them beside the recording, so the metric and the experience are the same object. You open the session, scrub to the moment, and watch what the user watched: the spinner, the late hero image, the layout jump. Bot sessions are filtered out and do not count against your quota, so the list you triage is human traffic.

Field data per session, not a lab score
A lab audit tells you what the page can do on a fast machine with a clean cache. Per-session field data tells you what it did, on this device, on this connection, on this release. Only one of the two can be attached to a support ticket.
Py reads the session with you
Py, the Prynt AI built into Replay, summarises a session, explains what went wrong and includes the highlights in a daily brief. Use the summary to decide which of twenty sessions deserves your afternoon, then watch the one you picked.
A triage loop from red metric to shipped fix
Run the same six steps every time, and save the filter you used.
- Filter by metric and threshold: LCP above 4 seconds, INP above 500 milliseconds, CLS above 0.25 on /checkout.
- Group the result by route, device type, browser and release.
- Watch three sessions from the worst group, not one. A single recording is an anecdote; three is a pattern.
- Check JS errors, rage clicks and dead clicks before you blame the network.
- Find the slow step in the timeline: first paint, hero image, font swap, ad slot, a hydration pass.
- Ship the smallest fix, then re-run the same filter after the release.
INP needs the session more than LCP does
An aggregate INP says that interactions are slow somewhere. A session says which interaction: the click on Apply coupon, the keypress in the OTP field, the tap on the third step of a checkout. In the recording you also see whether the interface answered at all, whether the user clicked twice, and whether the second click created a duplicate request.
CLS is a story about what loaded late
0.3 CLS is not a number you can act on by itself. In the session you see what moved: the banner that pushed the primary button down, the font that swapped, the notice that inserted itself above the fold. Watch the two seconds before the shift; the culprit is usually the element that finished loading then.
Put the session in front of the person who owns the page
A metric that only the performance team can see gets fixed slowly. Replay lets you embed sessions inside your own admin, per company or per user, with identify(userId, { accountId }), list them with the secret key and open them in the embedded player. Support sees the session the ticket is about; engineering sees the same session with the errors attached. Share links and comments keep the discussion on the recording.

Every form field is masked in the browser and again on the server, with no unmask option. That is what makes it reasonable to hand a session to support at all. Setup details are in the docs, and the platform overview shows how Replay shares one agent and one account with the rest of the suite.
FAQ
Is per-session monitoring the same as field data?
It is field data at a finer grain. Standard field data aggregates real visits into a percentile per page. Per-session monitoring keeps the number attached to the visit, so you can filter by device or route and watch what happened.
Which sessions should I look at first?
Start with the metric that matches the complaint: LCP for slow to load, INP for a button that does not respond, CLS for a misclick. Then filter to the worst sessions of that metric and group them by route.
Does recording vitals per session slow the page down?
Any monitoring adds work to the page. Measure the agent on your own pages before rollout and load it the way you load other third-party scripts, against a budget you actually check.
Start with the sessions that hurt
Prynt Replay records Web Vitals with every session, masks every form field, keeps bot traffic out of your quota and includes Py to explain what went wrong. The free plan covers 1,000 sessions a month with 14-day retention on one site; Pro is $279/month for 10,000 sessions and 30-day retention, with a 14-day trial.

Start on the free plan at pryntid.com, or read the glossary first if you want the metric definitions.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.