Do language apps work? What the makers' own studies measured

Learners in the makers' studies usually improved: 18 of the 20 documents tested language, and those that tested before and after found most learners scoring higher. But only one compared app users with people who did not use the app, 17 tested Spanish learners, and the maker paid for 10 of the 20. They show that people improve while using an app, not that an app works better than a class, a tutor or another app.

20 studies from 7 makers (Duolingo 7, Babbel 5, Rosetta Stone 3, Busuu 2, Lingvist 1, Mango Languages 1 and Pimsleur 1), each read from its own document on 11 Oct 2026.

On this page

Five questions to ask of any study
Every study at a glance
Study by study
Makers with no published study
What this means when you choose

Five questions to ask of any study

A maker's study is not worthless, and it is not proof either. Five questions sort out what one can tell you.

Were learners tested before and after? A single test at the end shows where learners stand, not how far they came. 11 of the 20 documents tested the same learners before and after; others tested once and compared the result with a different group, or with university students in other studies.
Against what? Without a group that learned another way, or not at all, a gain could come from the effort of volunteers who knew they were being tested. One study had such a group: Babbel's 2018 study of Swedish learners, and its groups were small.
Written, spoken, or both? A written placement test (WebCAPE, used by US colleges) checks grammar, vocabulary and reading; a speaking test such as the ACTFL OPIc checks what you can say. Several studies used only one of the two.
How many finished? Volunteers drop out, and those who stay are often the keenest. Some reports state it: Pimsleur's study had 24 of 82 learners finish Level 1, and Babbel's 2018 Spanish study tested 54 of 85 starters.
Who paid, and what were learners given? 10 of the 20 documents say the maker paid, and participants were usually given a free subscription, a gift card or both.

Every study at a glance

Newest first. "Comparison" is the group the app's learners were measured against, if any.

App	Year	Learners	Language	Test	Before and after?	Comparison	Paid by
Busuu	2025	1,205	Six languages: English, French, German, Italian, Spanish, Japanese	Written placement test, speaking test	Yes	Free against paid versions of the app	Busuu
Duolingo	2024	257	Spanish and French	Reading, listening, writing and speaking test	No	None	Duolingo
Duolingo	2024	140	English, Spanish and French	Short conversation tasks scored by raters	No	None	Not stated
Duolingo	2022	240	Spanish and French	Reading and listening tests	No	Learners at an earlier unit, and university students in other studies	Not stated
Duolingo	2021	258	Spanish and French	Speaking test	No	None	Not stated
Duolingo	2021	340	Spanish and French	Reading and listening tests	No	University students in other studies	Not stated
Duolingo	2020	225	Spanish and French	Reading and listening tests	No	University students in other studies	Not stated
Babbel	2019	117	Spanish only	Speaking test	No	None	Babbel
Rosetta Stone	2019	143	Spanish only	Written placement test, speaking test	Yes	None	Rosetta Stone
Pimsleur	2019	82 (24 finished Level 1)	Spanish only	Speaking test	Yes	None	Pimsleur
Mango Languages	2019	95	Spanish only	Written placement test, speaking test	Yes	None	Mango Languages
Lingvist	2019	38	Mixed	Questionnaire on what learners felt; no language test	No test	None	Not stated
Babbel	2018	54 of 85 who started	Spanish only	Speaking test, grammar and vocabulary tests	Yes	None	Babbel
Babbel	2018	24 with the app, 14 without	Swedish	Vocabulary and pronunciation test	Yes	Yes: a group that took the same class without the app	Not stated
Babbel	2016	325	Spanish only	Written placement test	Yes	None	Babbel
Busuu	2016	196	Spanish only	Written placement test, speaking test	Yes	None	Busuu
Duolingo	2012	88	Spanish	Written placement test	Yes	None	Duolingo
Rosetta Stone	2009	135 of 176 who started	Spanish only	Written placement test, speaking test	Yes	Two groups, both using the app (at a centre or at home)	Not stated
Rosetta Stone	2009	89	Spanish only	Speaking test	Yes	None	Not stated
Babbel	Undated	7 interviewed, 52 surveyed	Mainly English	Questionnaire on what learners felt; no language test	No test	None	Not stated

Study by study

Each study's own facts, from its document. The link opens the document itself.

Duolingo

Duolingo Path Meets Expectations for Proficiency Outcomes (Duolingo Research Report DRR-24-04) (2024; Xiangying Jiang; Elise Hopman; Ben Reuveni; Audrey Kittredge, Duolingo, Inc.)

Who took part: 257 adult Duolingo learners (165 Spanish, 92 French) who reached the end of A2 content (about the end of the beginner level) after the 2022 'path' redesign; little prior proficiency; IP address outside countries where the language is widely spoken; no other courses or apps
Language: Spanish and French (for English speakers)
How it was tested: Single group tested once at the end of A2 content with the STAMP 4S test (reading, listening, writing, speaking); outcomes also compared with earlier results from the older 'tree' home-screen design and related to self-efficacy; no control group
How long: On average about 200 hours of time spent learning to complete the beginner-level content, over a learning interval of about 276 days (about nine months)
What it found: Spanish learners: overall ACTFL Intermediate Mid (reading Intermediate High, listening Intermediate Mid, writing Intermediate Low, speaking approaching Intermediate Low). French learners: overall Intermediate Low. The path took longer to complete than the old tree and led to better outcomes.
Who paid: Duolingo (the test was paid for by Duolingo; participants received $100)

Duolingo learners can start a conversation after 4-6 weeks of app use (Duolingo Research Report DRR-24-01) (2024; Audrey Kittredge; Ryan Peters; Farrah Neumann; Xiangying Jiang, Duolingo, Inc.)

Who took part: 140 active Duolingo learners (70 who started 4 to 6 weeks earlier and 70 who started 6 to 8 weeks earlier), very little prior proficiency, no other learning tools
Language: English (for Portuguese or Spanish speakers), Spanish and French (for English speakers)
How it was tested: Single assessment per learner: short spoken and written conversation tasks scored by raters on a 0 to 2 scale; no pre-test and no control group
How long: Learners had used the app for 4 to 8 weeks before the single assessment; data collected September 25-26 and October 25-27, 2023
What it found: Participants had an average accuracy of 90% across the tasks and 84% passed every task with at least 60% accuracy, described as skills at the A1 level; spoken tasks were harder for some learners.
Who paid: Not stated (authors are Duolingo, Inc. employees; participants received $15 but the payer is not named)

Developing reading and listening proficiency with Duolingo: A cross-sectional analysis (Duolingo Research Report DRR-22-11) (2022; Xiangying Jiang; Lucy Portnoff; Luke Plonsky; Bozena Pajak, Duolingo, Inc.; Northern Arizona University (Plonsky))

Who took part: 240 adult Duolingo learners with US IP addresses who had completed Unit 7 (144 Spanish, 96 French), little or no prior proficiency, Duolingo as only tool; compared with an earlier Unit 5 group
Language: Spanish and French (for English speakers)
How it was tested: Cross-sectional: learners tested once at the end of Unit 7 (ACTFL reading and listening) and compared with a different group tested at the end of Unit 5, and with published university results by semester; not the same learners followed over time
How long: Hours not stated in the report; learners were tested shortly after completing Unit 5 or Unit 7
What it found: In Spanish the Unit 7 group scored about one ACTFL sublevel above the Unit 5 group in reading and listening (reading Intermediate Low to Intermediate Mid; listening Novice Mid to Novice High); Unit 7 results comparable with five university semesters, except Spanish listening (four semesters).
Who paid: Not stated (authors are Duolingo, Inc. employees except one co-author at Northern Arizona University; participants were paid $100 but the payer is not named)

How well does Duolingo teach speaking skills? (Duolingo Research Report DRR-21-02) (2021; Xiangying Jiang; Joseph Rollinson; Haoyu Chen; Ben Reuveni; Erin Gustafson; Luke Plonsky; Bozena Pajak, Duolingo, Inc.; Northern Arizona University (Plonsky))

Who took part: 258 adult Duolingo learners (156 Spanish, 102 French) who reached the end of Unit 5, with little or no prior proficiency, using Duolingo as their only tool, with IP addresses outside countries where the language is widely spoken
Language: Spanish and French (for English speakers)
How it was tested: Single group tested once with the Pearson Versant Spanish or French speaking test after finishing the beginning-level content, results mapped to CEFR levels; no pre-test and no control group. The report says only about 40% of the learners eligible for the test attempted to take it.
How long: Hours on Duolingo not stated; learners were tested on reaching the end of Unit 5 (completion of the beginning-level content) within the data collection window
What it found: Mean Versant speaking score 40.97 (Spanish) and 36.72 (French); most participants reached CEFR A2 or above (66% of Spanish and 53% of French learners per the discussion); French pronunciation was weak.
Who paid: Not stated (authors are Duolingo, Inc. employees except one co-author at Northern Arizona University; participants took the test free and received $20, payer not named)

Finishing half of B1 on Duolingo comparable to five university semesters in reading and listening (Duolingo Research Report DRR-21-03) (2021; Xiangying Jiang; Haoyu Chen; Lucy Portnoff; Erin Gustafson; Joseph Rollinson; Luke Plonsky; Bozena Pajak, Duolingo, Inc.; Northern Arizona University (Plonsky))

Who took part: 340 adult Duolingo learners worldwide (208 Spanish, 132 French; 58 countries) who finished the first seven units (first half of B1 content), little or no prior proficiency, Duolingo as only tool; average age 42 (Spanish) and 38 (French)
Language: Spanish and French (for English speakers)
How it was tested: Single group tested once (ACTFL reading and listening) at the end of Unit 7, compared with published results for university students after five semesters (Tschirner 2016); no pre-test; Speaking and writing not tested
How long: Median 203 hours of Duolingo time to complete the first seven units (Spanish 227, French 171 per the results); calendar time not given
What it found: Learners reached ACTFL Intermediate Mid/High in reading and Intermediate Low in listening; scores comparable with fifth-semester US university students (Tschirner 2016).
Who paid: Not stated (authors are Duolingo, Inc. employees except one co-author at Northern Arizona University; participants were paid $100 but the payer is not named)

Duolingo efficacy study: Beginning-level courses equivalent to four university semesters (Duolingo Research Report DRR-20-04) (2020; Xiangying Jiang; Joseph Rollinson; Luke Plonsky; Bozena Pajak, Duolingo, Inc.; Northern Arizona University (Plonsky))

Who took part: 225 adult Duolingo learners in the US (135 Spanish, 90 French) who had finished the beginning-level course (Checkpoint 5); ages 18 to 83, mean about 44
Language: Spanish and French (for English speakers)
How it was tested: Single group tested once (ACTFL reading and listening tests) on finishing the beginning-level course, then compared with published results for university students after four semesters (Tschirner 2016; Rubio and Hacking 2019); no pre-test, and the comparison is across different studies and populations. Speaking and writing were not tested.
How long: Median 112 hours of Duolingo sessions to finish the beginning-level content (Spanish 125, French 99); calendar time not given
What it found: Spanish learners reached ACTFL Intermediate Low in reading and approached Novice High in listening; French learners approached Intermediate Mid in reading and reached Novice High in listening; scores comparable with 4th-semester US university students in most comparisons (Duolingo significantly higher or lower in a few).
Who paid: Not stated (authors are Duolingo, Inc. employees except one co-author at Northern Arizona University; participants were paid $100 but the payer is not named)

Duolingo Effectiveness Study: Final Report (2012; Roumen Vesselinov; John Grego, Queens College, City University of New York; University of South Carolina)

Who took part: 88 Duolingo users in the final sample (196 randomly drawn initially): adults 18+, native English speakers, not of Hispanic origin, not advanced in Spanish, living in the US; people who answered an ad on Duolingo and Google
Language: Spanish (for English speakers)
How it was tested: Pre/post test with the same college placement test (WebCAPE); no control group described; participants who studied under 2 hours did not complete the study
How long: About eight weeks, September to November 2012
What it found: Average gain of 91.4 WebCAPE points (statistically significant); about 8.1 points per hour of study; estimated 26 to 49 hours (34 on average) for a person with no Spanish to cover the first college semester.
Who paid: Duolingo (the report says data collection and analysis were independent)

Babbel

Measuring Babbel's Efficacy in Developing Oral Proficiency (2019; Nelleke Van Deusen-Scholl; Mary Jo Lubrano; Zachary Sporn (Babbel), Yale University, Center for Language Study; third author is Babbel's Senior Communications Manager: Research and Pedagogy)

Who took part: 117 U.S.-based Babbel users and sales leads who met all requirements (out of 700+ respondents, 351 consented); self-reported absolute beginners in Spanish; 52% of the 117 were over 55
Language: Spanish only
How it was tested: Single-group, post-course-only speaking test (ACTFL OPIc); no pre-test and no control group, baseline assumed from self-reported beginner status
How long: Approximately 12 weeks; participants averaged 48 hours on Babbel (median 42; range 10.3 to 274 hours)
What it found: Spoken proficiency on the ACTFL OPIc after 12 weeks: 79.5% (93 of 117) scored in the Novice range and 20.5% (24 of 117) in the Intermediate range; 91% said they enjoyed the app and 75% felt they had met their goals. There was no baseline test, so gain is not measured.
Who paid: Babbel (the report says the researchers analysed the data independently)

Learning Spanish with Babbel: Report on the findings of an efficacy study (listed by Babbel as 'Learning Spanish with Babbel: Oral Proficiency Outcomes in App-Based Foreign Language Learning') (2018; Shawn Loewen; Daniel R. Isbell; Zachary Sporn (Babbel), Michigan State University (Second Language Studies); third author is Babbel's Senior Communications Manager: Research and Pedagogy)

Who took part: 85 university-community participants recruited at Michigan State University began; 54 completed all procedures (final sample average age about 24; English native speakers)
Language: Spanish only
How it was tested: Within-subjects quasi-experimental pre/post design (ACTFL OPIc oral interview, grammar test, vocabulary test); one group, no separate control group described
How long: Roughly three months (about 12 weeks) of autonomous study with Babbel; participants were asked to study a minimum of 15 minutes a day on average
What it found: Oral proficiency on the ACTFL OPIc rose by a median of one sublevel (mean 0.70); 70% of those who studied at least 6 hours and 78% of those who studied at least 15 hours gained one or more sublevel; 11% of the total sample gained two or more; grammar and vocabulary test scores also rose. Only 54 of 85 starters were tested at the end.
Who paid: Babbel (financial and technical support, including the cost of OPIc testing; the report says recruitment and analysis were independent)

Mobile language learning for integration - A case study of Babbel as a complementary tool: Final report (2018; Linda Bradley; Khaled Al-Sabbagh; Sylvi Vigmo, University of Gothenburg (Department of Education, Communication and Learning); Chalmers University of Technology)

Who took part: 24 highly educated adult migrants with 19 different first languages used Babbel (experimental group); 14 in the control group; all in a beginner Swedish (SFI) evening course in Sweden; non-native English speakers needed at least B2 English
Language: Swedish (Swedish as a Foreign Language course for migrants), with English as the Babbel display language for non-native English speakers
How it was tested: Experimental group (Babbel plus the SFI class) versus control group (SFI class only), randomly selected from 52 course participants; same vocabulary and pronunciation test before and after
How long: 12 weeks, in parallel with one term of the SFI course
What it found: Both groups improved on the Swedish vocabulary test after 12 weeks; the improvement was larger in the Babbel group; heavy Babbel users improved intonation and flow in spoken Swedish. Very small groups (24 and 14).
Who paid: Not stated (study set up in collaboration with Babbel, which supplied free app codes and usage data; funder of the research not named)

The Babbel Efficacy Study: Final Report (2016; Roumen Vesselinov; John Grego, Queens College, City University of New York (Economics); University of South Carolina (Statistics))

Who took part: 325 Babbel users, described as a random representative sample (taking the tests in Berlin or New York offices, or online at home)
Language: Spanish only
How it was tested: Pre/post test with the same college placement test (WebCAPE, Spanish); participants studied only with Babbel; the report describes no control or comparison group
How long: About 8 weeks (two months), June to August 2016; recommended goal 16 hours of study, absolute minimum 2 hours
What it found: Written Spanish placement-test (WebCAPE) gain: 92% of participants improved; on average 21 hours of study in two months to cover the requirements of one college semester of Spanish (15 hours for truly novice users). Written test only, no oral test.
Who paid: Babbel (the report says the researchers collected and analysed the data independently)

The impact of Babbel's virtual-classroom solution on workplace communication (undated; Milene Mendes de Oliveira (University of Potsdam); Zachary Sporn (Babbel); Lea Kliemann (Babbel); Alexandra Borschke and Meike Meyering (University of Potsdam), University of Potsdam, with Babbel staff as co-authors)

Who took part: 7 interviewees (German speakers learning English with Babbel Intensive) and 52 Babbel Intensive survey respondents (Babbel for Business users, ages 21 to 59)
Language: Mainly English (67.3% German speakers learning English); other languages for the rest
How it was tested: Sequential mixed-method: interviews, then a survey; measures learners' perceptions only, with no language test and no control group
How long: Data collection August 2020 to January 2021 (no fixed study period for learners)
What it found: Self-reported perceptions, not measured language gain: 86.5% of survey respondents felt learning with Babbel gives them more confidence to use a foreign language in real-life situations at work; 96.2% found the live classes enjoyable.
Who paid: Not stated

Rosetta Stone

The 2019 Rosetta Stone Efficacy Study: Final Report (2019; Roumen Vesselinov; John Grego; Steven J. Sacco; Mila Tasseva-Kurktchieva, Queens College, City University of New York; University of South Carolina; San Diego State University (Emeritus)/Sacco Global Consulting; University of South Carolina Linguistics Program)

Who took part: 143 Rosetta Stone users, a random representative sample of existing or new users residing in the US, 18+, novice or beginner learners of Spanish
Language: Spanish only
How it was tested: Pre/post test with the same college placement test (WebCAPE, written only); participants studied only with Rosetta Stone; no control group described; no oral test
How long: About 8 weeks (two months), June to August 2018
What it found: Gain of about 21 WebCAPE points per hour of study (95% CI 16 to 26); on average 13 study hours in two months (10 to 17) to cover the first college semester of Spanish.
Who paid: Rosetta Stone (the report says data collection and analysis were independent)

Measuring the Effectiveness of Rosetta Stone (final report) (2009; Roumen Vesselinov, Queens College, City University of New York)

Who took part: 135 adults finished (176 started) of a random selection from a pool of 6409 respondents to an advertisement in the Washington DC metro area; ages 19 to 70; mean age 39; people with advanced Spanish or Hispanic origin excluded
Language: Spanish only
How it was tested: Initial and final tests (WebCAPE written test and ACTFL OPIc oral test); participants randomly assigned to study at a Rosetta Stone facility (n=70) or at home (n=65); no group without Rosetta Stone
How long: 55 hours of study with Rosetta Stone, controlled by the investigators; calendar time not stated
What it found: After 55 hours: significant WebCAPE improvement; a share of students (56% to 72%) raised their oral proficiency by at least one level; 80% to 90% of users rated the software easy to use, helpful and enjoyable.
Who paid: Not stated

Rosetta Stone Evaluation Report (Rockman et al) (2009; Rockman et al (REA), San Francisco, Rockman et al, an independent evaluation, research and consulting firm)

Who took part: 89 adults: a stratified random sample selected by Rockman et al from a group of adults whom Rosetta Stone staff had recruited
Language: Spanish only
How it was tested: Pre/post test with the ACTFL OPI-C oral test, usage data and post-surveys gathered by Rosetta Stone staff; one group, no control group described
How long: Study run in May 2009; participants took six hours of Rosetta Studio sessions and averaged 64.5 hours of work with the software
What it found: ACTFL OPI-C oral proficiency: the report says all participants had significant gains and the majority advanced 1 to 2 levels in Spanish; participants rated the software highly. Oral test only.
Who paid: Not stated

Busuu

Busuu Six-Language Efficacy Study 2025 (2025; Roumen Vesselinov; John Grego; Mila Tasseva-Kurktchieva, Research team of comparelanguageapps.com (affiliations of the team in earlier reports: Queens College CUNY, University of South Carolina))

Who took part: 1,205 Busuu users (final sample), from 97 countries, 18+, living outside the countries where the study language is spoken; random selection from a pool of volunteers; levels ranged from beginner to advanced (Busuu's blog: 'complete beginners to advanced learners')
Language: Six languages: English, French, German, Italian, Spanish, Japanese; three Busuu versions compared (Free, Premium only, Premium and Live Lessons)
How it was tested: Pre/post tests (WebCAPE vocabulary/grammar and TrueNorth oral test), with groups compared by Busuu version (Free versus Premium); no group that did not use Busuu
How long: Approximately 8 weeks (two months), July to December 2024
What it found: 63.1% of all participants improved vocabulary/grammar; 72.5% of those with 8+ study hours improved oral proficiency; 88.7% improved on one or both tests; novice users need on average 15.6 hours over two months to cover first college-semester placement requirements.
Who paid: Busuu (cost covered by Busuu; the report says data collection and analysis were independent)

The busuu Efficacy Study: Final Report (2016; Roumen Vesselinov; John Grego, Queens College, City University of New York; University of South Carolina)

Who took part: 196 busuu users, a random representative sample of users living in or around London or New York, 18+, not of Hispanic origin, not advanced learners of Spanish
Language: Spanish only (beginner level)
How it was tested: Pre/post tests: WebCAPE written placement test and an OPIc oral test, both proctored; participants studied only with busuu; no control group described
How long: Two months (about 8 weeks), February to April 2016
What it found: Written test: 84% improved; on average 22.5 hours of study in two months to cover one college semester of Spanish. Oral test (OPIc): over 75% of the oral-test takers rose at least one level.
Who paid: Busuu (the report says data collection and analysis were independent)

Lingvist

Lingvist users rate their learning experience (blog post reporting a user-perception survey; not an efficacy test) (2019; Not named on the page (the writer says she was doing a master's degree in e-education), not stated)

Who took part: 38 Lingvist users who answered every question (14 female, 24 male, average age 30; 57.9% Lingvist Free and 42.1% Unlimited)
Language: Mixed; French most common, then German
How it was tested: Online questionnaire on user perceptions, recruited through social media; measures reaction only, no language test, no control group
How long: Survey ran in July 2019; half of the respondents had used Lingvist for more than a year
What it found: Satisfaction only: over 80% of respondents agreed or strongly agreed that Lingvist has a beneficial impact and leads to a positive learning experience; the writer states she cannot say anything about effectiveness.
Who paid: Not stated

Mango Languages

Mango Languages Efficacy Study: Final Report (2019; Roumen Vesselinov; John Grego, Queens College, City University of New York; University of South Carolina)

Who took part: 95 novice Spanish learners, a random representative sample; Mango's page says 95 adults in the United States aged 21 to 66
Language: Spanish only
How it was tested: Pre/post tests: WebCAPE written placement test and TrueNorth (TNT) oral test; participants studied only with Mango; no control group described
How long: Two months, July to September 2019
What it found: Written test: 82% improved and novice users need on average 15 hours in two months to cover one college semester of Spanish. Oral (TrueNorth): 77% of all participants increased their oral score; the 85% and 74% headline figures apply only to the 46 who studied 8 hours or more.
Who paid: Mango Languages (the report says data collection and analysis were independent)

Pimsleur

Pimsleur Efficacy Study: Final Report (2019; Roumen Vesselinov; John Grego, Queens College, City University of New York; University of South Carolina)

Who took part: 82 novice users of Spanish, a random representative sample of existing or new Pimsleur users residing in the US, 18+
Language: Spanish only (Pimsleur Premium Spanish)
How it was tested: Pre/post oral test (OPIc) only, no written test; participants studied only with Pimsleur; no control group described
How long: Two months, February to April 2019
What it found: 83% of users who completed Pimsleur Level 1 (30 lessons) rose at least one oral proficiency level (95% CI 63% to 94%); 73% of users with at least eight hours of study did so (CI 61% to 82%). Only 24 of the 82 completed Level 1.
Who paid: Pimsleur (the report says data collection and analysis were independent)

Makers with no published study

We looked on their own sites for a study of their app and found none: Memrise, Mondly, Rocket Languages and Speak. That does not mean the app does not work; it means its maker has not published a test of it. Independent researchers have studied some of these apps, and those papers are not listed here, because this page compares what each maker itself publishes.

What this means when you choose

The studies agree on one thing: hours matter. Several give their best figures for learners who studied at least eight hours, and in Babbel's 2018 Spanish study 70% of those who studied at least 6 hours gained a speaking sublevel, against 78% of those who studied at least 15. So the app you will open most days is a better bet than the one with the best-sounding study, and the free tier or trial is the cheapest way to find out which that is.

A course's stated level (a Spanish course "to C1", for example) describes what its lessons cover. It is not a result, and none of these studies tested learners at the top of a course. To see what a level means, and how far each app's courses say they go, read language levels explained.

Questions people ask

Do language learning apps actually work?

Learners in the makers' own studies usually scored higher after some weeks of study than before, on written placement tests and on speaking tests. But only one of the 20 studies compared app users with people who did not use the app, 17 tested Spanish learners, and the maker paid for 10 of the 20. They show that learners can improve while using an app; they do not show that an app beats a class, a tutor or another app.

Is Duolingo effective?

Duolingo's own reports, mostly written by its staff, tested learners who finished parts of its Spanish and French courses: those finishing the beginner content scored at levels comparable with US university students after four semesters in reading and listening. Its reports since 2020 each tested learners once, with no group of non-users (its 2012 study tested the same learners before and after, on a written test), and speaking and writing were tested in only some of them.

Are these studies independent?

Not in the usual sense. 10 of the 20 documents say the maker paid; most say the researchers collected and analysed the data on their own. Several Duolingo reports name no payer but are written by Duolingo employees. Participants were usually given a free subscription, a gift card or both for taking part.

Can an app make you fluent?

None of these studies claims it. The longest tested about 200 hours of study and found learners at intermediate levels in reading and listening. A course that says it goes to B2 or C1 describes what its lessons cover, not what a learner will reach.