REYX Labs · Research Report
The Science of Forgetting:
Why Conversations Are Lost,
and What Changes That
REYX Research · Updated 2026 · 6 sections, 6 references
Abstract
Human memory was not built to preserve conversation. This brief walks through four well-established findings in memory research: the forgetting curve, the limits of working memory, the serial position effect, and the reconstructive nature of recall. Each one specifically undermines the ability to retain what's said in meetings and calls. It then examines what changes when the burden of retention moves from biological memory to an external, always-on record, using REYX as the applied example.
Contents
1. The forgetting curve
In 1885, psychologist Hermann Ebbinghaus ran the first rigorous experiments on how fast newly learned information decays without reinforcement. Working alone, he memorized lists of meaningless syllables and tested his own recall at increasing intervals (an hour later, a day later, a week later) to isolate pure forgetting from the influence of prior knowledge.
The pattern he found, a steep drop within the first hours followed by a long, shallow tail, has been replicated in memory research ever since, including with meaningful material rather than nonsense syllables (decay is slower for content that connects to something you already know, but the same basic shape holds). Exact figures vary by study, material, and individual, but the direction is remarkably consistent:
These are the commonly cited approximations from forgetting-curve research (see References, §6), not a guarantee about any one conversation. But the shape is the point: without reinforcement, most of what you hear is gone fast, and almost all of it is gone within days.
2. Why conversations are especially vulnerable
The forgetting curve describes decay in general. Conversation is a worst case for it: four separate, well-documented mechanisms compound against retention at once.
Working memory has a hard ceiling
Psychologist George Miller's classic 1956 paper argued that working memory holds roughly seven items at once; Nelson Cowan's later reconsideration put the real number closer to four. Either way, the ceiling is fixed and low. A single meeting can produce a dozen names, figures, and commitments inside twenty minutes, far more than working memory can hold at once. Items past that ceiling don't get half-remembered. They mostly don't get encoded at all, because encoding requires holding something in working memory long enough to move it toward longer-term storage.
You can listen, or you can write, rarely both well
Attention is a limited resource that doesn't split cleanly between tasks. Mueller and Oppenheimer's 2014 study on note-taking found that the physical act of writing something down pulls cognitive resources away from actually processing what's being said, and that this trade-off holds even when the notes themselves are accurate. The more diligently you try to capture a conversation by hand in real time, the less of it you genuinely absorb as it happens, and the less you retain afterward, regardless of what's sitting in the notebook.
The middle disappears first
The serial position effect, documented experimentally by Bennet Murdock in 1962 and replicated many times since, shows that recall is strongest for the first and last items in a sequence and weakest for the middle. Applied to a conversation, that means the opening pleasantries and the closing summary tend to stick, while the specific number, objection, or commitment raised at minute 22 of a 40-minute call is disproportionately likely to be lost, regardless of how important it was.
Every act of recall quietly rewrites the memory
Memory isn't played back like a recording. It's reconstructed each time it's retrieved. Elizabeth Loftus and John Palmer's 1974 research on eyewitness testimony showed that even the specific wording of a question can measurably alter what someone reports remembering afterward. Applied to work, this means two people can leave the same meeting genuinely believing different, incompatible things were agreed. Not because either is being dishonest, but because both are reconstructing an approximation, not replaying an unaltered original.
3. What this costs in practice
None of this stays abstract once it plays out at work. The same four mechanisms from §2 show up as concrete, familiar friction:
Repeated context-setting
The same background gets re-explained meeting after meeting, because no single person fully retained it the first time it was covered. Time that should go toward new work goes toward re-establishing what was already discussed.
Disputes over what was actually agreed
Both sides are usually sincere, and both sides are usually wrong in different directions, because both are working from a reconstruction rather than a record. Resolving the disagreement takes longer than the original conversation did.
Cold handoffs
When someone leaves a role, the relationship history they were holding in their head leaves with them. Whoever inherits the account, the client, or the project starts from a lower baseline than the person before them ever had.
Slower decisions
Confirming what was actually said takes a round of messages instead of a search. Multiply that by every open question in a given week, and a meaningful share of a team's time goes to re-establishing facts that were already stated once, out loud, in the room.
4. An illustrative scenario
REYX is pre-launch. This is a composite walkthrough of a typical call, not a real customer account, used to make the research above concrete.
5. What external memory changes
Each mechanism from §2 has a direct counterpart once retention no longer depends on biological memory alone:
No ceiling to hit
REYX doesn't select what's worth remembering in real time. It captures everything. Working memory's four-item limit stops being the bottleneck for what gets preserved, because nothing has to pass through that bottleneck in the first place.
Listening and capture are decoupled
Nothing needs to be written down as it happens, so nothing competes with actually following the conversation. The record happens in the background; attention stays where it should be, on the person talking.
The middle is as retrievable as the ends
There's no primacy or recency bias in a transcript. Minute 22 is exactly as searchable as minute 1 or minute 40. Asking what was said doesn't depend on where in the conversation it happened.
The record doesn't reconstruct itself
A transcript doesn't quietly rewrite itself the way biological memory does. What was actually said stays what was actually said, so two people checking back on a commitment are checking the same source, not two independent reconstructions.
6. References
- Ebbinghaus, H. (1885). Über das Gedächtnis: Untersuchungen zur experimentellen Psychologie [Memory: A Contribution to Experimental Psychology]. Leipzig: Duncker & Humblot.
- Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97.
- Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114.
- Murdock, B. B. (1962). The serial position effect of free recall. Journal of Experimental Psychology, 64(5), 482–488.
- Mueller, P. A., & Oppenheimer, D. M. (2014). The pen is mightier than the keyboard: Advantages of longhand over laptop note taking. Psychological Science, 25(6), 1159–1168.
- Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13(5), 585–589.
Citations are provided to the best of available knowledge for orientation and further reading. Verify against the original source before formal citation elsewhere.
REYX
Memory shouldn't be the bottleneck.
REYX captures what your brain was never built to hold onto, so the conversation stays exactly as it happened, not as anyone remembers it.
Join the waitlist