The forgetting curve, 140 years on — what Ebbinghaus still gets right

You read a long-form article last month. At the time, you nodded along. The argument was clean, the examples were good, you closed the tab feeling sharper than when you opened it. If a friend asked you about it today, you'd manage a sentence. Maybe two. The shape of the thing is gone.

It's the most ordinary feeling in adult learning, and it shouldn't surprise anyone. The decay you're experiencing was first measured in 1885, has been replicated continuously since, and is, by some accounts, the single most reliably reproduced finding in psychology. 140 years we've known about it. Most online learning is still designed as though we don't.

A man, a metronome, and 2,300 nonsense syllables

Hermann Ebbinghaus was a German philosopher who, in the early 1880s, decided to do something nobody had really tried: measure memory experimentally. He had no laboratory, no funding, and exactly one subject — himself. He wanted to study memory uncontaminated by meaning, so he invented his materials: lists of three-letter "nonsense syllables" (zof, kel, dax), strings of consonant-vowel-consonant that didn't already mean anything to him.

He memorised list after list. He timed himself with a metronome to keep his recitation pace constant. Then he waited (twenty minutes, an hour, a day, six days, a month) and re-tested. The metric he tracked was savings: how much faster could he relearn a list compared to learning it the first time. Savings is a clever measure because it captures partial memory you can no longer recall on demand but that is still doing some quiet work under the hood.

In 1885 he published the results in a slim book titled Über das Gedächtnis — On Memory. The headline finding was a curve. After learning, retention dropped fast. Roughly half within an hour, around two-thirds by the end of the first day. Then the rate of forgetting slowed sharply. By a week the curve was nearly flat. There was always some residue left, but the bulk of what you'd just acquired was gone within 24 hours.

That's the forgetting curve, and it's almost certainly the most famous chart in the history of psychology. Worth pausing on is the modesty of the experiment behind it. One man, alone, with a metronome and a notebook, sitting at a table reciting bok zif lan. That's it. No control group, no fMRI, no statistical software. Just a careful person measuring himself.

The 130-year replication

A study with one subject and no controls would, today, be ignored. Ebbinghaus's findings survived because they kept replicating across subjects, materials, languages, decades.

The most direct test came in 2015, when Jaap Murre and Joeri Dros, at the University of Amsterdam, ran a faithful 130-year replication of the original protocol. Same metronome-paced learning of nonsense syllables. Same retention intervals. Same savings score. They wrote: "the original 1880 results of Ebbinghaus could be reproduced to a large extent." The curves overlay each other almost cleanly.

What 140 years of follow-up has changed is not the existence of the curve but our understanding of what shifts it:

• Meaningful material decays slower than nonsense. A poem you understand decays slower than syllables. A skill you've practised decays slower than a list of facts. Ebbinghaus chose nonsense specifically because it was the worst case; that's why his numbers feel so brutal.
• Prior knowledge slows decay. New material that hooks into things you already know forgets less aggressively than material that floats free.
• Emotional salience helps. Things you cared about when you encountered them stick longer than things you didn't.
• Sleep matters. Forgetting curves measured across a night of sleep are gentler than curves measured across an equivalent waking interval.

These adjust the steepness, the floor, the exact percentages. They don't change the shape. The shape (fast initial drop, slow asymptotic floor) is the part that holds up across material, age, language, and method. Not a quirk of nonsense syllables. It's how memory works when you don't intervene.

Why this matters for online courses

With that in hand, look at the standard online course shape: a sequence of video lectures, perhaps an end-of-module quiz, a final assessment, a certificate. The structure assumes that going through the material once is the work, and that the artefact at the end (the certificate, the completion percentage) represents what the learner now knows.

The forgetting curve makes this assumption a fiction. By the time a learner finishes a six-week course, the material from week one has been sitting unrehearsed for about 35 days. According to every replication of Ebbinghaus, including Murre and Dros 2015, the bulk of it is gone. Not because the course was bad, or the learner inattentive — because that's what unrehearsed memory does.

The certificate at the end measures completion. It does not measure retention. Different variables. Most platforms are honest about the first while pretending it implies the second.

The only known antidote that scales

There is exactly one intervention with strong, decades-deep evidence for actually flattening the forgetting curve: spacing your encounters with the material out over time. Encounter it. Wait. Encounter it again, ideally by trying to retrieve it before re-reading. Wait longer. Encounter it again. Each successful retrieval pushes the next forgetting curve out further; each one bends the slope a little flatter.

Call it the spacing effect, the close cousin of the testing effect. Cepeda and colleagues' 2008 paper is the clearest demonstration of how the optimal interval relates to how long you want to remember something. Their finding, simplified: the gap between reviews should be roughly 10–20% of the horizon at which you want to retain the material. If you want to remember something for a year, reviews two months apart are about right. For a week, reviews of about a day. There's no sharp peak; any spacing beats massed practice. The principle holds up.

Pair spacing with retrieval (making yourself try to recall before being shown the answer) and the effect compounds. Roediger and Karpicke's 2006 paper is the canonical demonstration that retrieval crushes re-reading at long horizons, even when the re-reading group studies for longer. There's a forthcoming post on the testing effect specifically; the short version is that recalling is itself an act of memory consolidation in a way that re-reading is not.

What you can actually do, today

You don't need an app. Even rough adherence to a review schedule beats the default of one-pass learning. A practical, low-tech version:

1. When you finish a lesson, close it and write down the three or four things you'd want to still know in a month. Phrase them as questions, not summaries.
2. The next day, look at the questions only. Try to answer them from memory. Then check.
3. Repeat at roughly day 3, day 7, day 14, day 30. The exact numbers don't matter much — what matters is that you're spacing them and trying to retrieve before checking.
4. Mix subjects. If you're learning two things, alternate between them on the same day. Blocked practice feels easier and produces less durable knowledge.

That's the whole technique. It is unsexy. It works.

The reason it isn't more widespread is not that people don't know about it. It has been popularised in dozens of books and apps over the last decade. The reason is that it requires upkeep. You have to maintain the questions, set the schedule, keep showing up. The systems that make this easy work brilliantly for the small subset of people who stick with them. Most people don't, and that's not a moral failing — it's a design problem.

What we do at Strive

When you generate a course on Strive, each lesson seeds a handful of small retrieval-practice cards automatically. They're phrased as questions, grounded in the actual lesson content, paraphrased so you can't pattern-match your way to the answer. They land in a finite daily queue at expanding intervals: tomorrow, three days, a week, two weeks, a month. You don't make the cards. You don't set the intervals. You answer questions when they come due.

It is, deliberately, the unflashy thing the research has been pointing at for 140 years, with the upkeep removed. The only known antidote to the forgetting curve is the one you don't have to remember to apply.

---

If you want to read deeper: Murre and Dros's 2015 replication is short, well-written, and reproduces Ebbinghaus's curves with eerie fidelity. Cepeda et al.'s 2008 paper on optimal spacing is the clearest treatment of how review intervals scale with retention horizon. Ebbinghaus's original 1885 monograph is in the public domain — it's also surprisingly readable.