You read a fact on Monday, feel it click, and cannot produce it on Friday. That is not a broken memory. It is the normal shape of forgetting, measured precisely 140 years ago and replicated almost point for point in 2015. The curve is steep at first and then flattens — which means when you meet a fact again matters as much as how well you learned it the first time. This article is about timing: what the forgetting curve actually measured (it is widely misquoted), what a modern replication found, and what the largest spacing experiments — including a 2026 systematic review — say about scheduling reviews for facts you want to keep.
It builds on, and is different from, our explainer on why trivia sticks, which covers retrieval practice itself. Here the question is narrower: given that recall works, when should the next recall happen?
In 1879–1880, Hermann Ebbinghaus taught himself lists of nonsense syllables — three-letter strings chosen to mean nothing, so that prior knowledge could not help him — and then relearned the same lists after delays of 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days, and 31 days. He published the results in 1885. His measure was savings: how much less time relearning took, as a share of the original learning time. If a list took 20 rehearsals to learn and 13 to relearn, the saving was 35 percent.
His savings figures are the real forgetting curve:
Two things follow. First, the popular version of this chart — "you forget 50 percent in an hour, 70 percent in a day" — mislabels the data. Savings is relearning effort avoided, not the share of a list he could recall, and Ebbinghaus never published a recall percentage curve. Second, look at where the loss happens: savings fell 24.5 points between 20 minutes and one day, then only 12.6 points more across the next thirty days. The first day is the cliff. Everything after is a slope.
And one limit matters for trivia players: he chose meaningless material on purpose. Facts with hooks — a story, a surprise, a connection to something you know — are forgotten more slowly than nonsense syllables. The curve gives you the shape of forgetting, not a personal expiry date for any given fact.
Sceptics had 130 years to wonder whether one man's self-experiment could be trusted. In 2015, Jaap Murre and Joeri Dros at the University of Amsterdam repeated it as faithfully as they could, publishing in PLOS ONE. Dros, then 22, spent about 70 hours learning and relearning syllable lists at the same intervals. The authors' conclusion: the forgetting curve "has indeed been replicated," with savings of 47.2 percent at 20 minutes, 37.3 percent at 1 hour, and 31.7 percent at 1 day — close to Ebbinghaus throughout the first day.
Two honest caveats. The replication's 31-day saving was just 4.1 percent against Ebbinghaus's 21.1 percent — the one large divergence, which the authors link partly to Dros's learning times creeping up over the long experiment (interference or fatigue); correcting for that drift puts the figure nearer 13.7 percent. And the curve was not perfectly smooth in either dataset: savings appear to tick up at the 24-hour point, a bump the authors note sleep research would predict but say remains unproven for this kind of task. It is also, still, a single-subject result on nonsense material. Nobody has produced one universal curve for all learners and all content — anyone who shows you an exact personal forgetting percentage is decorating, not measuring.
Ebbinghaus also found the remedy in his own data: relearn something and the next decline is slower. Modern work turned that observation into a scheduling science.
A 2006 quantitative synthesis by Cepeda and colleagues in Psychological Bulletin pooled 839 assessments from 317 experiments and confirmed that practice spread over time reliably beats the same practice massed together. Their 2008 follow-up in Psychological Science is the study trivia learners should know, because its material was trivia: more than 1,350 people learned a set of obscure facts, reviewed them after a gap of up to 3.5 months, and were tested up to a year later. The best gap between learning and review depended on how long the memory had to last:
As a share of the wait, the optimal gap shrinks — roughly 20–40 percent of a one-week delay, down to about 5–10 percent of a one-year delay. Two practical morals. First, there is an inverted U: too short a gap wastes the review, too long a gap lets the fact go entirely, and the peak moves with your deadline. Second, the payoff is large: averaged across delays, the best gap improved final recall by about 64 percent compared with reviewing immediately. Reviewing a fact the same evening you learned it is close to the least efficient time to do it.
Does this still hold up in current evidence? A 2026 systematic review by Luu and colleagues in Cureus gathered the spaced-repetition and retrieval-practice literature in undergraduate medical education: of 314 records, 21 studies met the inclusion criteria. Most controlled studies — where the schedule was imposed and outcomes measured — reported better examination performance or retention after spaced retrieval, with greater benefit from sustained, long-term engagement. The observational studies, mostly of students using flashcard apps on their own, were mixed: some found positive associations, others, including ones that measured app use objectively instead of by self-report, found none, and most carried a serious risk of bias.
Read fairly, that is a mature evidence picture, not a hype one. The spacing effect itself is among the most replicated findings in memory research. What is not established is that any particular app's default schedule is optimal for you, or that hours logged in an app — as opposed to successful, effortful recalls at widening gaps — are what does the work. The 2026 review's split between controlled and observational results points the same way as Cepeda: the schedule and the retrieval are the active ingredients, not the tool.
Translated into a habit a daily-question player can actually run:
No. Ebbinghaus deliberately used meaningless material to strip out prior knowledge. Meaningful, connected, surprising facts — most good trivia — decay more slowly. The curve's shape (steep, then flat) generalises; its exact numbers do not.
They are savings: the share of original learning time you no longer need when relearning. Ebbinghaus saved 58.2 percent of the effort after 20 minutes and 33.7 percent after a day. They are not the percentage of material recalled, despite how the chart is usually captioned.
It depends on when you need it. In the largest test (Cepeda et al., 2008), the best single gap was about 1 day for a test a week later and about 3 weeks for tests months to a year later — roughly 20–40 percent of a short wait, shrinking to 5–10 percent of a long one.
Controlled studies of spaced retrieval show real benefits; observational studies of self-directed app use are mixed and bias-prone (Luu et al., 2026). The app is a scheduler. The benefit comes from effortful recalls at widening gaps, which an app can arrange but cannot do for you.
This article is the starting point. Copy any of these prompts into your favorite AI assistant to learn more about the forgetting curve and spacing your reviews:
🔎 Build a review ladder for facts you keep losing
Here are 15 trivia facts I learned today and want to still know in [3 months]: [paste your facts]. Using the spacing pattern from Cepeda et al. (2008) — best gap roughly 20–40% of a one-week wait, shrinking to 5–10% of a one-year wait — build me an expanding review calendar (day 1, then widening gaps), turn each fact into a question I must answer from memory before seeing the answer, and tell me what to do with any fact I miss at a review: which rung of the ladder it restarts from, and why.
🧭 Audit a forgetting-curve chart before you believe it
I found a chart claiming people forget [X]% of new information within [time period], citing Ebbinghaus. Check it against what Ebbinghaus actually measured in 1885 (savings on relearning nonsense syllables: 58.2% saved at 20 minutes, 33.7% at 1 day, 21.1% at 31 days) and against Murre & Dros's 2015 replication. Tell me: is the chart reporting savings or recall, does it mention the material was meaningless syllables learned by one person, and what would an honest caption for the same data say?
💡 Run a two-week spacing experiment on yourself
Design a fair self-experiment for me: 40 trivia facts I don't know, split into two matched sets of 20. Set A I review once, the next day. Set B I reread three times tonight. Give me the full protocol — how to match the sets for difficulty, exact review times, and a blind recall test on day 14 — plus how to score it and what result would (and would not) let me conclude spacing beat rereading for me, given a sample size of one.
Tip: replace the bracketed parts with your own situation — the more specific your prompt, the more useful the answer.
Ebbinghaus's forgetting curve (1885) tracks savings — relearning effort avoided — not recall percentages: 58.2 percent saved at 20 minutes, 33.7 percent at one day, 21.1 percent at 31 days, with most of the loss inside the first day. Murre and Dros's 2015 single-subject replication matched the shape closely, including a possible sleep-related bump at 24 hours. Cepeda and colleagues showed across 317 experiments that spaced practice beats massed practice, and in a 1,350-person study of learned facts that the best review gap scales with the deadline — about 1 day for a test in a week, about 3 weeks for a test in a year — worth roughly 64 percent better recall than reviewing immediately. A 2026 Cureus systematic review of 21 studies found controlled spaced-retrieval interventions mostly helped, while self-directed app use showed mixed, bias-prone results.
The single most important takeaway: forgetting is a scheduling problem before it is a memory problem — put one effortful recall inside the first day, then widen the gaps to match when you actually need the fact.
One sharper AI prompt + one new daily trivia topic every week. Free forever, unsubscribe anytime.
We never share your email. One email a week, that's the deal.
Published November 14, 2026