your analytics says someone spent 4 minutes on your page. they opened it, then went to lunch.
i'm om yaduvanshi, 18, an applied ai agentic developer. i built cloudline, a cookieless analytics tool. day 51 of 90.
time on page counts the tab being open. engaged time is the part of the visit where someone was actually there. those are two different numbers, and most dashboards only show you the first one.
what time on page actually measures
a pageview starts a timer when the page loads. the timer stops when the page closes. everything in between counts, whether the person read a word or not. a tab left open while someone makes lunch counts as 4 minutes of reading. it was not reading.
what engaged time is for
cloudline tracks engaged time alongside scroll depth. scroll depth is measured at 25, 50, 75 and 100 percent. put the two together and a page starts to tell the truth. high scroll depth with low engaged time means people skimmed it fast. low scroll depth with high engaged time means the top of the page did the work.
the page that gets opened but not read
every site has one. a page with a big number next to it and nothing behind the number. people open it, then leave it open, then close it later. time on page calls it your best page. engaged time calls it a page people open and barely read.
what i do with the number
i sort by engaged time, not by pageviews. the page at the top of that list is the one people actually read. that is the page i write more of. the page with big pageviews and low engaged time gets rewritten or cut, because the headline did its job and the page did not.
the question i keep coming back to
most analytics tools hand you a big number and let you feel good about it. a smaller, honest number is more useful. it tells you what to write next, and it tells you what to stop writing.
which page on your site gets opened the most and read the least.
cloudline is free while in early access. no card.
ā om


Top comments (4)
Had a version of this in my daily actives. People who use my app only through Claude or ChatGPT never opened it, so they never counted. Adding them had its own open tab: the AI client lists my tools at the start of every chat, so only a real tool call counts now.
yeah, same problem from a different angle. people using your app through claude or chatgpt never open it, so the old count misses them entirely. only counting a real tool call is the honest fix. did you have to throw out anything else when you redrew the line?
A couple of things. Failed calls don't count either: a request refused for the free plan, a missing scope or a revoked token. It reached the server, but nobody got anything done.
And the total is deduped per person, so someone who opens the app and then uses it from Claude on the same day is one active, not two.
failed calls not counting is a good catch, that's the other half of the same problem. a request that reached the server but did nothing still looks like usage if you only count the hit. and the dedupe per person across app and claude is the part most people would get wrong, they'd double count the same person without thinking twice. did the dedupe part need much work or was it straightforward once you saw it?