Sources

Checked many times

• Studying a little every day gets more into your head than spending the same total time in one sitting.checked many times
  Donovan & Radosevich, 1999, Journal of Applied Psychology 84(5) · 63 studies compared, difference d = 0.46 — a sizeable figure for a learning effect.
• This holds not only in experiments but in ordinary classrooms, with real students.checked many times
  Mawson & Kang, 2025, Behavioral Sciences 15(6):771 · 22 studies, d = 0.54 (confidence interval 0.31–0.77). The authors warn that individual studies disagree widely.
• The longer you want to keep a word, the rarer its repetitions should be. To hold it for a week, review every day; to hold it for a year, about once a month.checked many times
  Cepeda et al., 2008, Psychological Science 19(11) · Over 1,350 participants, tested after delays of up to a year. The best gap is about 20% of the target interval on a horizon of weeks, and 5–10% on a horizon of a year.
• This has been tested on foreign words too. The difference shows up most when you test yourself a few days later rather than straight away.checked many times
  Kim & Webb, 2022, Language Learning 72(2) · 48 experiments, 3,411 participants, 98 comparisons.
• Recalling beats re-reading. Covering the translation and trying to remember it does more than reading it ten more times.checked many times
  Adesope, Trevisan & Sundararajan, 2017, Review of Educational Research 87(3) · 188 experiments, 272 comparisons. g = 0.61; against plain re-reading 0.51, against passive review 0.93.
• It is better to test yourself the next day than at the end of the same session. That is why a word comes back to you tomorrow.checked many times
  Adesope, Trevisan & Sundararajan, 2017, Review of Educational Research 87(3) · g = 0.82 when tested a day or two later, against 0.56 on the same day.
• An answer you worked out yourself sticks better than one handed to you.checked many times
  Bertsch et al., 2007, Memory & Cognition 35(2) · 86 studies, 445 comparisons, d = 0.40.
• Mixing different topics inside one session works better than drilling a single one for a long time. Not always, though: on connected texts there is no difference.checked many times
  Brunmair & Richter, 2019, Psychological Bulletin 145(11) · g = 0.42 on average; strongest on topics that resemble each other, and not statistically significant on connected texts.
• To follow about 95% of ordinary conversation you need around 3,000 words. But 95% means one unknown word in twenty — not “I understand nearly everything”.checked many times
  Adolphs & Schmitt, 2003, Applied Linguistics 24(4) · Counted across two large collections of recorded speech: CANCODE (5 million words) and the BNC. Word families were counted, not each form separately.
• Reading a book without a dictionary takes 8,000–9,000 words; following speech by ear just as freely takes 6,000–7,000.checked many times
  Nation, 2006, Canadian Modern Language Review 63(1); Hu & Nation, 2000 · The threshold for independent reading is 98% of words known, not 95%. Word families were counted.
• The people who keep going longest are the ones who find it interesting or genuinely need the language. “I'd be ashamed to quit” and “there is something in it for me” do not help at all.checked many times
  Alamer et al., 2025, Educational Psychology Review · Correlation with success: interest and importance r = 0.23; guilt −0.06; external reward −0.01.
• Games and points do help you take the material in. That they also add any desire to study rests on much thinner evidence than people assume.checked many times
  Sailer & Homner, 2020, Educational Psychology Review 32 · g = 0.49 for learning, 0.36 for motivation, 0.25 for behaviour. Among the most rigorous studies the effect on motivation did not hold up.
• Pay someone to study and they study less, not more. Praise and a clear explanation of their mistakes do the opposite — they help.checked many times
  Deci, Koestner & Ryan, 1999, Psychological Bulletin 125(6) · 128 experiments. Material reward: d = −0.40 for taking part, −0.36 for finishing. Praise and feedback: +0.33.
• Deciding exactly when and where helps more than deciding to study at all. “After my morning coffee, in the kitchen, ten minutes” works; “I really should get to it” does not.checked many times
  Gollwitzer & Sheeran, 2006, Advances in Experimental Social Psychology 38 · 94 trials, more than 8,000 participants, d = 0.65.
• Fear of making mistakes gets in the way of learning a language, and it gets in the way badly. It is one of the most reliably measured findings in the entire field.checked many times
  Teimouri, Goetze & Plonsky, 2019, Studies in Second Language Acquisition 41(2) · 19,933 participants from 23 countries; correlation with results r = −0.36.
• That fear can be dealt with — the methods have been tested and they work.checked many times
  Xiong et al., 2024, Journal of Language and Social Psychology · 37 studies, anxiety reduced by −0.61. Of the seven conditions examined, only age turned out to matter.

Checked little, or the result is disputed

• A word said out loud is remembered slightly better than one read in silence. The difference is small.checked little
  Fawcett, 2013, Acta Psychologica · g = 0.37. Part of the effect is not “out loud” winning but “in silence” losing.
• Getting by on simple subjects usually takes 180–200 hours of study. That is a very rough estimate: it works out differently for different people.checked little
  Cambridge English, Guided learning hours · A guideline from the Cambridge exam board for level A2; up to B1 it is 350–400 hours. It is the board's estimate, not a research finding.
• A habit takes about two months to settle: two weeks for some people, half a year for others. One missed day spoils nothing.checked little
  Lally et al., 2010, European Journal of Social Psychology 40(6) · 66 days on average, ranging from 18 to 254. Missing a single day had no noticeable effect on habit formation.
• People abandon something they have already started less often than something they have not begun. That is why the first steps are marked straight away.checked little
  Nunes & Drèze, 2006, Journal of Consumer Research 32(4) · A card with ten stamps, two of them already given, got 34% of participants to the end; a card with eight stamps from scratch got 19%. The work required is the same either way.
• It is not too late for an adult to learn a language. It gets a little harder with the years, but there is no age after which the door closes.checked little
  Hartshorne, Tenenbaum & Pinker, 2018, Cognition 177; criticism — van der Slik et al., 2022 · Data on 669,498 people; the authors put the boundary at around 17. A re-analysis of the same data says there is no sharp boundary at all and the decline is gradual.
• 225 adults learned a language from scratch with an app alone and reached a level comparable to two years of university classes.checked little
  Jiang et al., 2021, Foreign Language Annals 54(4) · 225 participants, beginner to low-intermediate reading and listening — comparable to four semesters of a university course. Some of the authors work for the service under study, which is worth keeping in mind.

Claims you will not find here

Each of these circulates in this niche, sounds convincing, and does not survive a check against the primary source.

• “21 days to form a habit” — contradicted by the data: the median is 66 days, ranging from 18 to 254.
• “Short sessions improve memory by 20–30%” — nobody has properly tested this, and different authors call anything from one to fifteen minutes “short”.
• “2000 words and you understand 95% of conversations” — a bait-and-switch: 95% refers to text coverage, not to the share of conversations you understand.
• “There is no critical period” — the data shows a decline; the argument is about its shape, not its existence.
• “Streaks are proven to help learning” — there are no peer-reviewed trials, and the nearest data points the other way.