Back to Explore
GuideVibeFix editorialUpdated Nov 20, 2026

Messages Vanishing Into Thin Air? Read Receipts and Online Status in Practice: First Separate Delivered, Read, and Replied

Of 1,000 messages, 974 arrive, 861 get seen, 602 get seen within an hour, and 437 get answered. A receipt is not a dot but a funnel: fix late seeing before fixing seen-but-unanswered. The author, product, and numbers in this walk-through are fictional.

A real photo: a person holding a phone in both hands and typing, with chat bubbles on the screen.

Messages vanish into thin air — but at which step?

Xiaobei built a small collaboration tool for small teams called Lighthouse Squad, and chat is its most-used feature. One day a user asked her support team a very practical question: “When I send a colleague a message, did they not receive it, not see it, or see it and choose not to reply?” She could not answer. Her chat had a single “Sent” state, and after that messages vanished into thin air: a colleague says it never arrived, and you cannot tell whether that is a network problem or someone who never opened it; a colleague says they saw it, and you cannot tell whether they really saw it or are just being polite; hardest of all is the read-but-unanswered message — you do not know whether they are busy, forgot, or waiting for you to change your mind first. A read receipt looks like a small dot, but it is the trust layer the whole chat stands on.

Spread the numbers out, and receipts stop looking like a yes-or-no choice and start looking like a funnel. Before Lighthouse Squad shipped receipts, the team sampled 1,000 messages users had sent: 974 reached the other person's device, 861 were opened and seen, 602 were seen within one hour, and 437 eventually got a reply. In other words, out of every 100 messages, 97 arrive, 86 get seen, 60 get seen within an hour, and 44 come back with a reply. The 26 that never arrived are a network and push problem; the 113 that arrived but were never opened belong to people who never opened them; the 259 that were seen but not within an hour belong to people who saw them late, busy with something else; and the 165 that were seen within the hour but never answered are the hardest part — people who saw it and did not reply.

To be clear: the author, product, and full set of numbers in this guide are fictional examples; the methods themselves are real, usable practice. Treat the numbers as worked examples for doing the math, not as industry data to quote.

A real photo: a person holding a phone in both hands and typing, with chat bubbles on the screen

First, separate three things: delivered, read, and replied

Before building anything, separate three states. They are not the same thing, so do not count them as if they were. First is delivered: the message reached the other person's device and a push went out, but they have not opened it. Second is read: they opened the message and saw its content. Third is replied: they wrote something back, even if it is just “got it.” The three form a funnel, not a set of equals: delivered does not mean read, and read does not mean replied. Many teams treat “read” as “they agreed,” and that is where the misunderstanding starts — someone glancing at a message only proves they know about it, and knowing is still a long way from agreeing.

Put the funnel on the table before deciding which step to fix first. Lighthouse Squad's numbers run: 1,000 sent; 974 delivered, losing 26 on the way, a delivery rate of about 97% (974 / 1,000 = 97.4%), lost to pushes that never arrived, devices offline, and apps killed in the background; 861 read, about 88% (861 / 974 = 88.4%), losing 113 that arrived but were never opened; 602 read within one hour, about 70% (602 / 861 = 69.9%), losing 259 that were opened more than an hour later; 437 answered, about 73% (437 / 602 = 72.6%), losing 165 that were seen but never answered. The losses add up: 26 + 113 + 259 + 165 = 563, exactly 1,000 − 437. The books balance.

Message status funnel: 1,000 messages pass through delivered, read, and read within one hour, and 437 end up answered

The biggest hole in the funnel is not the last step but the third: 259 messages were seen more than an hour late. The problem is not “they won't reply” but “they see it too late” — when the message arrived they were in a meeting, driving, or putting a child to bed, and by the time they opened it two hours had passed. Fix in this order: first the lateness (write the reason in the push text, flag important messages, do not send at midnight), then the seen-but-unanswered (ask a specific question, offer an answer they can tap), and only then the undelivered 26 (a network and push problem, and the slowest to fix). Many teams start by agonizing over what shape the read-receipt dot should be, which is effort spent at the narrowest part of the funnel — it recovers none of the 259 late-seen messages.

Four states, no more: sending, delivered, read, failed

The state machine needs exactly four states, no more and no fewer. First is sending: the message has just left the sender's device and the server has not confirmed it yet — a small spinner is enough, no words needed. Second is delivered: the server has it and the push went out — show one check mark. Third is read: the other person opened the message — the check mark changes color or becomes their avatar. Fourth is failed: the network dropped or the server refused it — a red exclamation mark next to it, tap to resend. Four states need only two visual elements: a check mark and an avatar. Do not invent a new icon per state; nobody will remember them. And do not draw “delivered” and “read” so alike that users cannot tell them apart.

Group chats need their own treatment — do not copy the one-to-one design. In a direct message, read means the other person saw it; in a group, “read” has to carry a count: in a group of 10, if 3 people have seen it, write “Read by 3” and let a tap show which 3. Never show the bare word “Read” in a group before everyone has read it — if one person has not seen it, “Read” is wrong. One more case to decide in advance: if someone blocks you or leaves the group, the message stays at “Delivered” forever. Do not leave it spinning — a spinner tells the sender “almost there,” which is a lie.

When exactly a message counts as read deserves a decision too. Read should be recorded at the moment the message appears on the other person's screen — not the moment they open the chat window. If they open the window but the message is still below the fold, it is not read. What about reading the whole message from the notification shade — does that count? Both answers are used in the wild; pick one, write it down, and stay consistent. The worst option is mixing both in the same product: counted as read from the notification shade but not from the list, and users will conclude the receipts are lying.

“Typing…” and online status: do not turn them into surveillance

“Typing…” is the easiest feature to overdo. The right behavior: show it one second after the other person starts typing, hide it three seconds after they stop, and write only “Typing…” — never stream the actual keystrokes. Streaming keystrokes is surveillance, not a feature. Two more details: if they delete everything and start over, the indicator should not flicker; if they type on two devices at once, show it once. In a group, once more than two people are typing, stop listing names and write “3 people are typing.” The indicator exists to give the sender a sense that a reply is coming, not to let them count keystrokes.

Online status has more traps than typing. Draw a line first: what does “online” mean? Lighthouse Squad's rule is a heartbeat from the client every 30 seconds, and the server marks someone offline after 90 seconds without one; open the app but idle for 5 minutes, and they show as “away,” not “offline.” Do not turn “online” into second-by-second monitoring: nobody cares whether someone was last online 3 minutes ago or 5 minutes ago — they care whether that person can reply now. So three levels are enough: online, away, offline. Do not add a fourth, “last seen at,” timestamp — that timestamp only breeds “they were online 2 minutes ago, why aren't they replying?”

Invisible mode is mandatory, and it belongs in the most visible place. Some people simply do not want others to know they are online: they are working late, in a meeting, or just want to read a message in peace. With invisible mode on, others cannot see their online status — and they cannot see others' either. An equal exchange, and fair. Do not ship a half version where someone is invisible but read receipts still go out: if the other side sees “read,” they know you are online, and the invisibility was pointless. One more rule: admins and group owners get no special privilege. Do not open a back door that lets anyone see everyone's true status — open that door once, and trust is gone.

Three receipt policies compared: default-on is not always best

Receipt policy comes in three levels, and Lighthouse Squad ran the math on fictional comparison data. Plan A is on by default: every conversation sends read receipts; out of every 100 messages, 86 get read and 44 get a reply within an hour — but 9% of users go into settings and turn receipts off. Plan B is on in groups, optional in direct messages: 81 get read and 41 get a reply within an hour per 100 messages, and only 4% turn receipts off. Plan C is off by default: only 52 get read and 28 get a reply within an hour per 100 messages, and 0% turn receipts off — because there was nothing on to turn off. Laid side by side, Plan A has the prettiest numbers, but 9% of users voted against it with their settings; Plan C respects privacy best, but half the messages vanish unanswered.

Comparison of three receipt policies: on by default, on in groups and optional in direct messages, and off by default, with read rates and one-hour reply rates

Lighthouse Squad picked Plan B, because it splits the difference between “visible” and “not pushy.” In groups it is on by default: group messages are public anyway, and receipts help everyone confirm “how many people know about this” without hurting anyone. In direct messages it is off by default and users turn it on themselves: direct messages are where “read but unanswered” pressure bites hardest, so the choice goes back to the user. One companion rule: anyone who turns receipts off cannot see other people's receipts either — an equal exchange, same as invisible mode. When choosing a plan, do not look only at the read rate; the share of users turning receipts off is the more honest number. Users vote with their settings more accurately than with their words.

Do not roll a plan out to everyone at once, either — pilot it in one small group for two weeks first: on by default in groups, off by default in direct messages, then watch three numbers — whether the read rate rose, whether the one-hour reply rate moved, and whether the share of users turning receipts off climbed. If all three move in the right direction, roll it out everywhere; if the opt-out share climbs, the receipts are too pushy — loosen them first, by switching direct messages to off by default, or turning “Typing…” off along with them. A pilot is cheap to fix: a problem in one small group takes a day to reverse, while a problem found after a full rollout takes a week, plus explaining the change to every user.

How to build it: three tables, two heartbeats, one dedup

The implementation is simpler than it looks — three tables are enough. The first is a messages table: sender, recipient, conversation, content, and sent time per message. The second is a receipts table: delivered time, read time, and read device per message, plus a per-person read detail for groups recording when each person read it. The third is a conversations table holding the last message and the last read message id (last_read_message_id) per conversation — a receipt is computed by walking backward from that id: everything at or below it, sent by the other side, counts as read, with no need to record messages one by one. Typing uses a temporary key that expires after 3 seconds; presence uses a heartbeat every 30 seconds, with offline after 90 seconds of silence.

Dedup and ordering are the two easiest places to ship a bug. Dedup: the same read receipt can arrive twice when the client retries over a flaky network, so the server dedups on “message id + reader” and never lets one message's read time get stamped twice. Ordering: a receipt can arrive before the message itself — when the other side reads fast and the network is slow — so the server stores the receipt first and reconciles the state once the message lands, instead of dropping it. The offline queue matters too: messages sent while the other side is offline should be stored on the server and pushed when they come back — never lost in the offline gap, and never letting receipts jump around while offline.

Push and receipt ordering deserves thought too. The message pushes first and the receipt comes later — never the other way around: a receipt cannot arrive before the other person has the message. Push text should state the reason — “Xiaobei @mentioned you in the group: tomorrow's event moved to 3 pm” works far better than “you have a new message,” because the reader can tell at a glance whether to open it now. During do-not-disturb hours, pushes stay silent but receipts still go out: if the other person opens and reads it, the read state updates as usual — just do not wake the sender at midnight. The easiest ordering mistake is the receipt arriving before the message: store the receipt first and reconcile once the message lands, never drop it, and never stamp a receipt time earlier than the message's arrival — a timeline where the receipt predates the message tells the user the system is lying.

Advanced group uses: @mentions, announcement reads, and multiple devices

Once direct-message receipts are done, groups have three advanced uses left. The first is pairing @mentions with reads: when someone is @mentioned in a group and has not read it, that message deserves its own note — the push can be escalated once, say half an hour later, with a “someone @mentioned you in the group” reminder. Only once, though: more than that is spam. The second is announcement reads: when a group announcement goes out, the owner should see “7 of 10 have read it” instead of asking around one by one. Announcement reads should be counted separately from ordinary message reads: an announcement is something everyone needs to know, and a low read rate tells the owner to send it again or at another time. The third is the boundary case: if the other person read a message on their computer and their phone has not synced yet, does it count as read? Yes — read is recorded against the person, not the device. Whichever device read it first is the one that counts, and users should not have to refresh state back and forth between two devices.

One more case to decide in advance: forwarding. A sends to B, B forwards to C, C reads it — does A's read state change? No. C read the message B forwarded, not the one A sent; the two messages are two different things, so do not link them. Read states on a forward chain are counted separately. And a last rule: never use receipt data for performance reviews. Some teams take “read rate” and grade employees on how fast they answer messages — and the moment receipts ship, employees start fighting back with “read but unanswered” or just turn receipts off. Receipts exist so two people can agree on an answer, not so a third person can score them.

Before launch, check off this list

First, the state machine has exactly four states — sending, delivered, read, failed — no more, no fewer; group reads must carry a count. Second, read is recorded the moment the message appears on the other person's screen, and whether reading from the notification shade counts is one consistent rule across the product. Third, “Typing…” appears after one second and disappears after three, and never transmits what the other person is typing. Fourth, online status has three levels — online, away, offline — and no second-accurate “last seen” timestamp. Fifth, invisible mode and the receipt switch must be reciprocal: turn yours off, and others cannot see yours; do not keep a back door for yourself. Sixth, never write “Read” in a group before everyone has read it. Seventh, store offline messages first and push later, and dedup receipts by id. Eighth, watch four numbers daily in the first week after launch: delivery rate, read rate, one-hour read rate, and the share of users turning receipts off — if that last number climbs, your receipts are too pushy; loosen them before optimizing anything else.

One last thought: read receipts do not exist to reassure the sender; they exist so both sides agree on “does the other person actually know?” Done well, users barely notice receipts exist. Done badly, users argue about them every day. Ship Plan B, run it for a month, watch the opt-out rate, and only then decide whether to tighten or loosen.

Views 9Comments 0

Comments (0)

ME
0/1000
Loading comments...
Browse projectsPublish your project

Related articles

Real photo: a laptop, tablet, and phone on a wooden table showing the same green-leaf website, the phone view with navigation in one column, a potted plant nearby
Guide
Squashed on phones? Responsive layout and media queries, explained plainly

One page should stand up on a phone, a tablet, and a computer. This guide follows MDN's Responsive web design, Using media queries, and the viewport meta reference: fluid HTML, the viewport tag, fluid grids and max-width, images that never burst their container, media query syntax, breakpoints set by content, mobile first, responsive type, testing capability instead of width, and a pre-launch checklist.

Mobile AppsDesign ExperienceFrontend Engineering
Real photo: a bank of light gray metal filing cabinets with one drawer pulled open, revealing rows of hanging file folders inside
Guide
A real database inside the browser: IndexedDB local storage, explained plainly

localStorage only stores strings and has no indexes. IndexedDB is a real database in the browser: structured data, far larger capacity, indexes, and transactions. Following MDN's Using IndexedDB, IDBDatabase, and IDBObjectStore: the five-step routine, opening, upgrades, keys, indexes, transactions, add versus put, get, cursors, ranges, and multi-tab upgrades.

Tool TipsData WorkflowsFrontend Engineering
Real photo: hands typing on a MacBook Pro showing JavaScript code that changes HTML content, with a white mouse, papers, and binders on a white desk
Guide
Functions without the confusion: JavaScript functions, parameters, and arrow functions, explained plainly

You already call functions every day; this guide names what you have been doing. Following MDN's Functions, Arrow function expressions, and Rest parameters: declarations versus expressions, parameters and arguments, default values, rest parameters, spread syntax, return values, arrow function shorthand, the this trap, callbacks, and scope.

Tool TipsFrontend EngineeringAI Coding