DEV Community

Cover image for time on page lies. engaged time is the number i actually read
Om Yaduvanshi
Om Yaduvanshi

Posted on

time on page lies. engaged time is the number i actually read

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.

cloudline dashboard

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.

cloudline visitors over time chart

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.

https://trycloudline.space

— om

Top comments (10)

Collapse
 
beusebiu profile image
Eusebiu Balan •

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.

Collapse
 
omyvnss profile image
Om Yaduvanshi •

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?

Collapse
 
beusebiu profile image
Eusebiu Balan •

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.

Thread Thread
 
omyvnss profile image
Om Yaduvanshi •

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?

Thread Thread
 
beusebiu profile image
Eusebiu Balan •

Straightforward once I saw it. Each platform already writes one row per person per day, and "All" counts distinct people across those rows, so the app plus Claude on the same day stays one.

The longer list was what isn't use at all: background syncs, widget refreshes, analytics flushed as the app goes to the background.

Thread Thread
 
omyvnss profile image
Om Yaduvanshi •

background syncs counting as use is the sneakiest one on that list. the app doing work on its own looks identical to a person doing something if all you count is the hit. same problem as engaged time, just on your side of the stack.

Collapse
 
ascendra_ventures_92c21f2 profile image
Ascendra Ventures •

Sorting by engaged time instead of pageviews is a good habit, and pairing it with scroll depth makes it more useful: high scroll with low engaged time means skimming, low scroll with high engaged time means the top of the page did the work. Eusebiu's thread about only counting a real tool call, and not counting refused or failed calls, is the same idea applied to daily actives.

It reminds me of a dashboard we built at Ascendra Ventures for an egg distribution startup that had asked for a full ERP. The numbers that ran the business sat in spreadsheets and end-of-day reports that were stale by the time they were compiled. What they needed was one live view of sales against target, stock, demand and logistics, with shortfall flagged per region before a route ran short. The lesson we wrote up was to ask which decision you cannot make today before buying more software. Engaged time passes that test, because it tells you what to write next.

Do you plan to show engaged time per section, so you can see where people stop reading?

Collapse
 
omyvnss profile image
Om Yaduvanshi •

your egg distribution example makes the same point, one live number that decides something beats a report full of stale ones. per-section is the obvious next step, yeah. scroll depth tells you how far someone got and engaged time tells you how long they stayed, but neither one says which part they actually sat with. no timeline on it though, still working out what the first version looks like.

Collapse
 
dpietro profile image
Denise Pietro • • Edited

This is a great example of why better analytics isn't necessarily about collecting more data, but about measuring the right behavior. The distinction between time on page and engaged time seems simple, yet it fundamentally changes how we interpret content performance. A page can generate impressive session-duration metrics without delivering meaningful value, and optimizing around those misleading signals can send content creators in entirely the wrong direction. I particularly like the combination of engaged time and scroll depth because neither metric tells the complete story independently. Someone might spend several minutes deeply reading the introduction without scrolling far, while another visitor might rapidly skim the entire article. Those behaviors require very different interpretations and content decisions. One additional consideration worth exploring is how engaged time accounts for accessibility tools, keyboard navigation, and readers who pause to think or take notes without actively interacting with the page. Getting those details right could make the metric even more representative of genuine attention while preserving the privacy advantages of cookieless analytics. Ultimately, the strongest insight here is that traffic measures reach, but meaningful engagement provides a much better indication of whether content is actually resonating. I'd much rather understand what readers genuinely consume than celebrate inflated numbers that merely look impressive on a dashboard. Great work building toward a more honest approach to content analytics!

Collapse
 
omyvnss profile image
Om Yaduvanshi •

the accessibility point is a real gap and i don't have a clean answer for it yet. someone reading slowly with a keyboard, or just sitting and thinking, looks idle to anything activity-based. engaged time beats raw time on page, but it's still a proxy. it can't tell thinking from idle, and that's the part i need to figure out.