Loading post…
Monthly notes on comprehension, accessibility, and AI tutoring. No spam, no upsells, unsubscribe anytime.
Search "forgetting curve" and you get a tidy graph: a line that plunges from 100 percent to almost nothing within a week, usually decorated with a claim that you forget 70 percent of what you read within 24 hours. The graph is real, the number is invented, and the story most people tell about it gets both the shape and the point wrong. The forgetting curve is one of the most cited results in psychology and one of the most misquoted. It is worth knowing what the original experiment actually showed, because the accurate version is more useful than the meme.
Hermann Ebbinghaus was his own only subject. Starting in the early 1880s he memorized thousands of nonsense syllables, consonant-vowel-consonant strings like WID and ZOF, chosen precisely because they carried no meaning he could lean on. He learned a list to perfect recitation, waited a set interval, then relearned it, and measured how much faster the second learning went. That saving in relearning time was his index of memory. His 1885 monograph, later translated as Memory (opens in new tab), reported that savings fell steeply at first and then leveled off, dropping fast in the first hour and much more slowly across the following days.
Two things about the method matter for reading the results correctly. First, he measured savings, not recall. Savings can be high even when you cannot consciously reproduce a single item, because relearning traces on a memory too faint to retrieve directly. So the curve tracks something closer to latent retention than to "how much can you recite right now." Second, nonsense syllables were the point, not a flaw. Ebbinghaus wanted material stripped of prior knowledge so that meaning could not inflate his memory. Real reading is the opposite case, dense with meaning and connection, which is one reason the raw curve understates how well you hold onto a well-understood argument.
The remarkable thing is how well the finding has held up. In 2015, Jaap Murre and Joeri Dros ran a careful replication, with one subject spending 70 hours learning and relearning syllable lists at intervals from 20 minutes to 31 days. Their replication in PLOS ONE (opens in new tab) reproduced the shape of the original curve closely. Savings dropped to roughly a third after a single day and toward a fifth after a month, and the overall trajectory matched Ebbinghaus's numbers from 130 years earlier. Forgetting is fast, it is front-loaded, and it is stable enough across a century of cultural change to count as a genuine fact about human memory. On the headline claim, that memory for new material decays sharply and predictably if you do nothing, Ebbinghaus was right.
The trouble is in the details that got flattened on the way to the internet.
He guessed the wrong equation. Ebbinghaus fit his data with a logarithmic function, and for a century textbooks drew the curve as a smooth exponential decay. Neither is the best description. John Wixted and Ebbe Ebbesen tested six candidate functions against forgetting data from words, faces, and even pigeons, and their 1991 paper On the Form of Forgetting (opens in new tab) found that a simple power function beat exponential, logarithmic, hyperbolic, and linear fits in every case. When they reanalyzed Ebbinghaus's own savings data, it too declined as a power law. The practical meaning of a power curve is that forgetting decelerates: the rate of loss keeps slowing, so a memory that survives a week is far more durable than its first-day decay rate would predict. The exponential meme makes forgetting look like a lost cause. The power law says the survivors stick.
The curve is not even smooth. Murre and Dros noticed a bump: retention actually ticked upward around the 24-hour mark rather than continuing straight down. They linked it to a night of sleep, during which the brain consolidates recently learned material. That single wrinkle undercuts the clean plummeting graph entirely and points at how sleep locks in what you read. The curve you were shown in a productivity thread has the sleep step sanded off.
And the biggest distortion is one of framing rather than math. The popular curve describes what happens when you encounter something once and never touch it again. Almost nobody quoting it mentions that the whole reason the curve is interesting is that you can interrupt it.
Ebbinghaus studied a second effect in the same monograph, and it is the one that turns his gloomy graph into a plan. Each time you relearn material, the next round of forgetting is slower. Review resets the curve and flattens it, so the line falls less steeply after every well-timed encounter. A century of research on spacing has since put numbers on this. A 2006 meta-analysis of 317 experiments (opens in new tab) found that spreading the same study time across separate sessions reliably beats massing it into one, and that the gains grow the longer you need to remember something. The forgetting curve is not a sentence. It is a schedule waiting to be written.
Modern spaced repetition software is built directly on this. Algorithms like FSRS model each memory with three quantities: how difficult the item is, how retrievable it is right now, and its stability, defined as the number of days it would take for your chance of recall to fall from 100 to 90 percent. Every successful review increases stability, which is the same thing as saying it flattens your personal forgetting curve for that item. Trained on Anki review histories, the open spaced repetition benchmark (opens in new tab) evaluates these schedulers against hundreds of millions of real reviews, and the three-variable memory model predicts recall far more accurately than the fixed intervals older systems used. Ebbinghaus measured the curve. FSRS measures yours, and reschedules around it.
Ebbinghaus gave us the curve. His real gift was showing that it bends.
