The meeting ends the way good meetings are supposed to end: on time, with nods. Everyone agrees a decision was made. Everyone is slightly wrong about what it was.
A week later it surfaces. The engineer heard “we're cutting the export feature to hit the date.” The product manager heard “we're shipping a smaller version of export.” The designer, who spoke right after the decision landed, barely registered it at all. Three honest people, one meeting, three memories — and now a second meeting on the calendar to settle what the first one supposedly settled.
This is not a discipline problem, and it is not solved by telling people to pay attention. It is a memory problem, and it is one of the better-documented phenomena in cognitive psychology. A conference room full of intelligent adults is, structurally, one of the worst environments human beings have ever built for forming a reliable shared memory.
You weren't listening — you were rehearsing
In 1973, the psychologist Malcolm Brenner ran a deceptively simple experiment. People sat in a group and took turns reading words aloud; afterward, they were tested on how many of the words they could recall. They remembered other people's words reasonably well — with one striking exception. Recall collapsed for the items presented immediately before their own turn.
Brenner called it the next-in-line effect, and the mechanism is straightforward: in the moments before you perform, your attention pivots inward. You are rehearsing what you're about to say, and rehearsal is expensive. The words landing in your ears during that window are heard, in the acoustic sense, but they are never properly encoded. There is nothing to forget, because nothing was stored.
Now put that finding in a conference room. The moment before you speak in a meeting is, by definition, a moment when someone else is speaking. If you have a point queued up — and in any contentious discussion, most people do — you are spending exactly the seconds that matter composing your opening line. Worse, the decision moments are the high-participation moments. When the group finally converges, everyone is leaning forward with one last thing to add. Which means everyone is rehearsing. Which means almost no one is encoding the very sentence they will all later claim to remember.
Everyone assumed someone else had it
In the 1980s, Daniel Wegner noticed something about long-term couples: they remember as a unit. One partner holds the birthdays and the names; the other holds the directions and the account numbers. Together they store far more than either could alone — but only because each knows, implicitly, who holds what. Wegner called this a transactive memory system, and well-functioning teams build them too. The senior engineer is the memory for why the database looks the way it does; the ops lead is the memory for what broke last quarter.
Meetings quietly corrupt this system. Transactive memory works when responsibility for remembering is assigned, even tacitly. In most meetings it is assigned to no one, so everyone makes the same comfortable assumption: the group has it. This is diffusion of responsibility applied to remembering — the mnemonic version of a crowd watching someone struggle and each person assuming somebody else will step in.
Modern meetings add a second leak. In 2011, Betsy Sparrow, Jenny Liu, and Wegner published research in Science showing that when people believe information will be stored externally — that they can look it up later — they remember the information itself less well, retaining instead a memory of where to find it. The finding became known as the Google effect. In meeting terms: the moment someone says “this is being recorded” or “there's a notes doc,” everyone's encoding relaxes. The belief in the record substitutes for the memory. And if the recording is never rewatched, and the notes turn out to be three ambiguous bullet points typed by someone who was also arguing — the external memory is empty, and the internal one was never made. The room ends up with less memory than if no one had promised a record at all.
Memory is a rewrite, not a recording
Even the parts people do encode don't stay put. In 1932, Frederic Bartlett had English students read an unfamiliar Native American folk tale, “The War of the Ghosts,” and retell it over increasing intervals. The retellings didn't just shrink — they warped, systematically, toward what the tellers already expected stories to be. Strange details were dropped or replaced with familiar ones. Bartlett's conclusion reshaped the field: remembering is not playback. We store the gist and reconstruct the details at recall, using our existing schemas as scaffolding.
This is why the engineer and the product manager can both pass a lie detector about the same decision. The engineer reconstructs it through a schema of architecture and risk; what got stored was “we protected the deadline,” and the details fill in from there. The product manager reconstructs through the roadmap; what got stored was “export survives in some form.” Neither is lying. Each is doing what human memory always does — and the meeting, by ending at the moment of maximum verbal agreement, handed the actual interpretation over to be reconstructed separately, privately, and differently overnight.
Making a room remember
The fix is not longer minutes or a better recording bot. Each mechanism above has a specific counter.
Write the decision in the room, in exact words. Not after the meeting — during it, on a shared screen, while the people who made the decision are still together. The sentence has to survive contact with the group before the group disperses. This is the only moment when five private reconstructions can be collapsed into one artifact, because the schemas haven't had a night to diverge yet.
Read it back aloud. Psychologists studying what's called the production effect — Colin MacLeod and colleagues have examined it extensively — find that material read aloud is remembered better than material read silently; the act of producing it makes it distinctive in memory. Reading the decision back also runs a final integrity check: a sentence that sounded fine in the flow of conversation often reveals its vagueness when spoken alone, and objections surface while they're still cheap.
Assign the remembering. Name a single owner of the decision log for the meeting. This converts a broken, implicit transactive memory system into a working, explicit one — and it kills the someone-else-has-it assumption, because now someone specific does.
Don't make your most active speaker the note-taker. The facilitator or the person driving the argument is the most rehearsal-loaded mind in the room — the next-in-line effect guarantees they are encoding the least. Give the notes to someone whose entire job, for that hour, is to listen.
Treat “it's recorded” as a warning, not a comfort. The Google effect means an external record weakens internal memory whether or not the record is ever consulted. A recording is a backup for a written decision, never a replacement for one.
The meeting after the meeting
A forgotten decision is never just forgotten. It comes back as a clarification thread, a redone estimate, and eventually a second meeting to re-decide what the first one decided — the most expensive kind of meeting there is, because it produces nothing new. That afterlife is what meetingmortem was built to examine. It treats every meeting as something worth a small autopsy: what it cost in real salary terms, what it actually decided, and whether those decisions survived the week — so the pattern of decisions evaporating shows up as a pattern, not as a series of isolated mix-ups. If you'd like to see what your meetings are really leaving behind, you can start at meetingmortem.lumenlabs.works.