Someone asks how last month went and you answer before you’ve really thought about it. Rough, honestly. Or fine, quiet. The answer arrives so fast it feels like a lookup — like you opened a file.

You didn’t. You built that answer just now, out of whatever was nearest to hand.

I think we are well advised to keep on nodding terms with the people we used to be, whether we find them attractive company or not.

— Joan Didion, “On Keeping a Notebook,” Slouching Towards Bethlehem (1968)

Didion was making the case for a notebook. Decades of research have been making the same case with data.

You don’t retrieve a feeling — you rebuild it

In 2016, Robin Kaplan, Linda Levine, Heather Lench and Martin Safer built a study around two American presidential elections. A few days after the 2008 and 2012 votes, they asked people how they felt about the outcome — an emotion, attached to an event — and how they had been feeling in general, which is a mood. A month later they asked the same people to recall both answers.

Memory didn’t simply fade. It bent, in two directions at once. People underestimated how intensely they had felt about the election, and overestimated how intense their general mood had been. What predicted the shrinkage was how important the election still seemed a month on: as the importance faded, so did the remembered feeling. Meanwhile the strength of their reaction bled outward and coloured the whole month around it.

So it isn’t that you exaggerate your past. You shrink the sharp specific thing and inflate the vague background — roughly the opposite of what most of us assume we do.

Confidence is not accuracy

The obvious objection is that this applies to forgettable feelings, not the ones seared in. Those are the ones to worry about.

On 12 September 2001, Jennifer Talarico and David Rubin asked 54 Duke students to write down their memory of hearing about the attacks the day before, alongside a memory of an ordinary recent event, then brought them back 1, 6 or 32 weeks later.

The 9/11 memories decayed — and decayed at the same rate as the everyday ones. What set them apart was that ratings of vividness, recollection and belief in their own accuracy stayed high, while falling away for ordinary events.

Their conclusion, in their own words:

Flashbulb memories are not special in their accuracy, as previously claimed, but only in their perceived accuracy.

Your certainty and your accuracy came apart, and nothing announced it.

What fills the gap

Michael Robinson and Gerald Clore, reviewing the literature on how people report their own feelings, draw the distinction that explains this. An emotion is episodic — experiential, tied to a moment and its context. A belief about your emotions is semantic: general, conceptual, detached from any particular day. While the episode is still accessible you report from it directly. Once it isn’t, you don’t get an error message. You answer from the belief instead — what someone like me would have felt, in a month like that.

There’s a second edit upstream of that one. Donald Redelmeier and Daniel Kahneman recorded pain minute by minute for 154 colonoscopy and 133 lithotripsy patients, then asked each of them to rate the total pain afterwards. Those ratings tracked the worst moment and the last three minutes; how long the procedure went on barely registered. That’s remembered pain rather than remembered emotion, so it isn’t the same finding — but it’s the same architecture. What you keep isn’t a summary of the whole thing. It’s a couple of vivid moments standing in for everything else.

The drift has a direction, and it’s mostly kind

None of this is a malfunction, and stopping here would leave out the best part.

Across a program of research gathered into a 2003 review, Richard Walker, John Skowronski and Charles Thompson named the fading affect bias: the unpleasantness attached to a negative memory fades faster than the pleasantness attached to a positive one. Call up a bad week from three years ago and it has less grip on you than the good week sitting beside it. Their reading — pressed further in a later paper that asks bluntly what the bias is even for — is that this is healthy coping doing its job.

Your memory isn’t a broken instrument, then. It’s tuned for living forwards rather than for keeping the record, and it’s rather good at the first job.

And you trust the wrong one

In 2020, Linda Levine and colleagues asked 957 people to predict how they would feel about the 2016 US election result, then how they actually felt, then — later — to recall it. People remembered their responses more accurately than they predicted them, but judged their predictions the more accurate of the two. What explained that was the greater importance and vividness of anticipated feeling: vividness is what your sense of accuracy actually tracks.

The study also included 33 people with Highly Superior Autobiographical Memory, the rare individuals who can tell you what they did on a particular Tuesday in 1994. They were no better at this than anyone else.

“I have a good memory for how things felt” is not an exit.

What to do about it

Mostly this calls for writing things down nearer to when they happen.

  1. Record close to the time. Robinson and Clore are clear about where the trustworthy report lives: while the episode is still accessible. A week later you’re interviewing a witness who has already started summarising.
  2. Write the specific, not the summary. “Work was stressful” is already a belief about a whole month. “My hands were shaking before the 3 p.m. call” is a record. Only one survives the edit.
  3. Keep the date and the original wording. Both are the evidence.
  4. Expect the fade. You’re not supposed to still feel last year’s bad week at full volume.
  5. Be most suspicious of your most confident memories. Vividness isn’t a quality signal.

One caveat the research earns: a note written at the time isn’t the objective truth of that day either — it was shaped by the mood you were in, and by what you chose to write down at all. What it is is a record made while the episode was still accessible, rather than a reconstruction assembled later out of beliefs about yourself.

Where this shows up for us

Two parts of Smart Journaling exist for this gap. Echoes keeps sentences you actually said and brings them back later, sometimes in your own recorded voice, anchored to the moment you said them. Letters sends one line from a month or a year ago, with its date on it and a tap through to the entry behind it.

To be straight about the limits: nothing in this literature says that reading old entries back makes you happier, and nobody has tested whether this app does. The narrower claim is the one we’d stand behind — a sentence you said at the time is a different kind of evidence from your memory of how you felt, and it’s the only version that doesn’t get quietly revised.

If keeping better company with who you used to be sounds worth it, you can join the early-access list.

Sources

The studies referenced here are peer-reviewed. Effects are real but modest and differ between people — this is a description of the evidence, not medical advice.