AI language learning statistics

AI Language Learning Statistics 2026: 25+ Verified Facts

Artificial intelligence has fundamentally changed how people learn languages. From adaptive flashcard apps to real-time pronunciation coaches that never sleep, AI has moved language learning out of classrooms and into earbuds, commutes, and lunch breaks worldwide.

But how big is this shift, really? And does the research back up the hype?

This page collects AI language learning statistics for 2026, covering market size, adoption, learner outcomes and platform performance. Every figure was checked against its original source: company filings, peer-reviewed studies and official reports. Figures we could not trace to a primary source were removed.

Last verified: 18 October 2026

Key AI Language Learning Statistics

Top Statistics

  • 58.7 million people used Duolingo every day in Q2 2026, up 23% on a year earlier
  • Across 17 research projects and 8,282 learners, AI-guided language learning beat normal instruction with an effect size of d = 0.39, a moderate gain
  • In a 1,000-student field experiment, students who practised with a general-purpose AI tool scored 17% lower once the tool was taken away
  • Speak users had spoken more than 1 billion sentences to its AI tutor by December 2024

1. The AI Language Learning Market: Size & Growth

Market size depends heavily on what each research firm counts, so read these figures as estimates rather than settled numbers.

$21.06 billion: online language learning market in 2025, forecast to reach $50.82 billion by 2031, a 15.83% annual growth rate from 2026 (Mordor Intelligence, updated September 2026)

$14 billion: total language learning market in 2025, forecast to reach $46.2 billion by 2035, about 12.7% a year from 2026 (GM Insights, October 2026)

$22.1 billion: online language learning market in 2024, with self-learning apps holding 64.2% of revenue (Grand View Research)

$2.06 billion: AI language learning apps market in 2025, forecast to reach $10.92 billion by 2033, 23.5% a year from 2026 (Future Data Stats)

Notice the spread. GM Insights now puts the whole language learning market below Mordor Intelligence’s figure for online learning alone, and an earlier GM Insights edition valued the same market at $85.1 billion. The firms define the market differently and revise often, so use these numbers for scale, not precision. If the Future Data Stats forecast holds, AI language apps would grow more than fivefold by 2033.

Source: Mordor Intelligence (2026), GM Insights (2026), Grand View Research, Future Data Stats

2. How Many People Are Using AI to Learn Languages?

The most reliable adoption data comes from the companies themselves, in figures they report to investors.

58.7 million: Duolingo daily active users in Q2 2026, up 23% year on year (Duolingo Q2 2026 shareholder letter)

140.6 million: Duolingo monthly active users in Q2 2026, up 10% (Duolingo)

12.7 million: paying Duolingo subscribers at the end of June 2026, up 17% (Duolingo)

50 million+: daily active users and more than $1 billion in bookings, Duolingo’s milestones for full-year 2025 (Duolingo Q4 2025 results)

10 million+: downloads of Speak, the AI conversation app, as of December 2024. Speak does not disclose its active users (TechCrunch)

No independent survey has yet measured what share of all language learners use AI tools. Figures such as “52% of learners use AI apps” circulate widely, but we could not trace them to any published survey, so they are not included here.

Source: Duolingo (2026), TechCrunch (2024)

3. Does AI Actually Help You Learn? What the Research Says

The research is positive, but more modest than app marketing suggests, and the gains are uneven across skills.

3a. General Learning Outcomes

  • A 2024 meta-analysis in Language Learning & Technology pooled 17 research projects with 8,282 learners. AI-guided learning outperformed business-as-usual instruction with an effect size of d = 0.39, a moderate gain, and systems using machine learning did better than rule-based ones (Lee & Lee, 2024). That fits how adaptive learning algorithms personalise lessons and adjust difficulty in real-time.
  • A 2025 meta-analysis of 15 studies and 2,156 participants found a larger effect (d = 1.17), but the gains were stronger for receptive skills (reading and listening) than for productive skills (speaking and writing), and strongest of all for vocabulary (Xu, Yu & Liu, 2025). Vocabulary is also where spaced repetition apps for vocabulary building have the strongest evidence.
  • In a 10-week study of 60 university students in China, the group learning with Duolingo scored 73.9 on the post-test against 61.1 for the traditionally taught group, and reported higher motivation and self-regulated learning (Wei, 2023). It is a small, single-site study, so treat it as an example rather than proof.

Two cautions. Small classroom studies often report inflated effect sizes, and a meta-analysis is only as strong as the studies it pools. And short-term gains are not the same as lasting learning (see section 7).

Source: Language Learning & Technology (2024), International Review of Applied Linguistics (2025), Frontiers in Psychology (2023)

3b. AI, Speaking and Pronunciation

Speech recognition has made instant spoken feedback possible, from text-to-speech tools that model native pronunciation patterns to dedicated AI coaches. The evidence for speaking is thinner than for vocabulary, though.

  • In a 6-week study of 131 Chinese university students, the group practising with an AI speaking assistant showed a notable drop in foreign language anxiety, while the control group showed no significant change (Zhang et al., 2024, reviewed in Frontiers in Psychology, 2025).
  • AI support helps productive skills such as speaking less than receptive skills such as listening (Xu, Yu & Liu, 2025).
  • Speak users had spoken more than 1 billion sentences to its AI tutor by December 2024 (Speak).

There is no published, independent standard for how accurately language apps detect pronunciation errors. Accuracy claims such as “91%” or “95%” usually come from vendors or market reports without a stated test. That is why we are testing it ourselves: Heklin is running the same scripted error and pronunciation test on six AI speaking apps, and the results will be published here. For how the technology works day to day, see our guide to AI speech recognition as a 24/7 pronunciation coach.

Source: Frontiers in Psychology (2025), International Review of Applied Linguistics (2025), Speak (2024)

3c. AI Tutors and Chatbots

Conversational AI tutors (chatbots that simulate real-world conversation practice) have become one of the most studied areas in language learning research.

  • A 2025 systematic review of 37 studies on AI chatbots in language learning, published between 2014 and January 2025, found that most research measures cognitive outcomes such as test scores (43% of studies), followed by emotional outcomes (38%), behavioural outcomes (15%) and learner agency (4%) (Li et al., 2025).
  • In practice, that means far more is known about whether chatbots raise test scores than about whether learners keep practising with them.
  • Duolingo’s AI Video Call feature is now included for most new Super Duolingo subscribers, with access extending to existing subscribers later in 2026 (Duolingo Q2 2026 shareholder letter).

Source: Smart Learning Environments (2025), Duolingo (2026)

4. Platform Performance: The Numbers Behind the Big Names

Two of the largest AI-driven platforms report their own figures. Here is what they have disclosed:

Duolingo

$748 million: Duolingo’s revenue in 2024, up 41% (Duolingo annual report)

$298.5 million: Duolingo’s revenue in Q2 2026, up 18% year on year (Duolingo)

148: new language courses launched in April 2025, built with generative AI (Duolingo)

250+: language courses on offer at the end of 2025 (Duolingo Form 10-K)

Speak

$1 billion: valuation reached in December 2024 with a $78 million Series C round, bringing total funding to $162 million (Speak)

200+: business customers using Speak for Business (Speak, December 2024)

AI is also changing how progress is measured. A new generation of AI-powered language assessment apps that track progress with skill diagnostics and personalised insights is replacing end-of-unit tests with continuous, adaptive measurement.

Source: Duolingo (2025, 2026), Speak (2024)

5. Retention: The Engagement Problem

Keeping learners active is the hard part, and it is where gamification earns its place.

84%: Duolingo’s current user retention rate in Q2 2026, its own measure of how many recent learners stay active and an all-time high (Duolingo)

15.4 million: learners revived a lost streak during Duolingo’s June 2026 Streak Revival event, including nearly 8 million with no active streak when it began (Duolingo)

>60%: first-week churn for many freemium language apps, according to Mordor Intelligence, which does not publish the underlying data

Duolingo’s numbers show how gamification makes language learning habit-forming: streaks, reward loops and social pressure combined into one experience. Even excellent AI is not enough on its own. The platforms that keep learners are the ones that use AI to personalise the experience, not just automate the content.

Source: Duolingo (2026), Mordor Intelligence (2026)

6. Who Is Learning Languages with AI, and Why

  • Self-learning apps held 64.2% of online language learning revenue in 2024, the largest share of any learning format (Grand View Research).
  • 1,177,766 international students studied in the United States in the 2024/25 academic year, up 5%, a large market for English language platforms. A fall 2025 snapshot showed a 1% decline (Institute of International Education, Open Doors 2025).
  • Business demand is real: Speak alone reports more than 200 corporate customers. Many companies are pairing language training with AI translation for business communication as a parallel capability.

Source: Grand View Research, Institute of International Education (2025), Speak (2024)

7. Where AI Language Learning Is Heading: Trends to Watch

Here are the developments with the clearest evidence behind them:

Generative AI and Conversational Fluency

Large language models are enabling platforms that can sustain open-ended, context-aware conversation in many languages. The shift is from scripted exercises toward genuinely unpredictable dialogue, closer to what you get with a human tutor. Speak’s billion-sentence milestone and Duolingo’s AI Video Call feature are early signals of where the mainstream is heading.

Voice-First Learning

Speaking practice is the fastest-growing category of AI language tool, yet it is where the research is weakest: current meta-analyses find smaller gains for speaking than for vocabulary, reading and listening. Expect this gap to be the main test for voice-first apps over the next few years.

AI for Minority and Less-Resourced Languages

Most current AI language tools are heavily skewed toward major languages. Startups are beginning to address this gap. Lingawa raised $1.1 million in pre-seed funding in November 2024 to build native-language learning for the African diaspora, a signal that the next wave of platform investment will look beyond English, Spanish and Mandarin.

The OECD’s Warning on Lasting Learning

The OECD’s Digital Education Outlook 2026, published in January 2026, warns that higher short-term performance does not automatically mean durable learning. A field experiment with about 1,000 high-school students in Türkiye, published in PNAS in 2025, shows why. Students with a general-purpose GPT tool scored 48% higher on practice problems, and a GPT tutor designed to give hints instead of answers raised practice scores by 127%. Once access was removed, students who had used the general-purpose tool scored 17% lower than those who never had AI, while the tutor group avoided that drop.

The study was in mathematics, not languages, but the lesson carries over: an app that hands you answers can make practice feel easier without making you better. Look for tools that make you produce the language yourself.

Source: OECD Digital Education Outlook (2026), Bastani et al., PNAS (2025)

Key Takeaways

If you only take one thing from each section:

  • Market size: Estimates for 2025 range from $14 billion to over $21 billion depending on what is counted, and every firm forecasts double-digit growth.
  • Adoption: The hardest number available is Duolingo’s: 58.7 million people a day in Q2 2026, up 23%.
  • Effectiveness: AI helps, by a moderate margin on average, and most of all for vocabulary, reading and listening.
  • Speaking: The evidence is thinnest here. Lower speaking anxiety is the clearest benefit so far.
  • Lasting learning: Tools that give you answers can lift short-term scores and still leave you worse off. Choose apps that make you speak and write.

Taken together, the evidence points to a sector that is growing fast and delivering real but uneven results. The best AI tools help most when they make you do the work.

How we verify: every statistic on this page was checked against its original source on 18 October 2026. We prefer company filings, peer-reviewed studies and official reports. Market research estimates are included only where the publisher states the figure on its own site, and we note when firms disagree. Figures we could not trace to a primary source were removed.

Sources & Methodology

  1. Duolingo: Q2 2026 Shareholder Letter (2026)
  2. Duolingo: Fourth Quarter and Full Year 2025 Results (2026)
  3. Duolingo: Form 10-K for 2025 (2026)
  4. Duolingo: Form 10-K for 2024 (2025)
  5. Duolingo: Duolingo Launches 148 New Language Courses (2025)
  6. Speak: Raising $78M Series C at a $1B Valuation (2024)
  7. TechCrunch: OpenAI-backed Speak raises $78M at $1B valuation (2024)
  8. Lee & Lee: The effects of AI-guided individualized language learning, a meta-analysis. Language Learning & Technology (2024)
  9. Xu, Yu & Liu: A meta-analysis examining AI-assisted L2 learning. International Review of Applied Linguistics (2025)
  10. Wei: Artificial intelligence in language instruction. Frontiers in Psychology (2023)
  11. Frontiers in Psychology: Learner emotions in AI-assisted language learning (2025)
  12. Li et al.: Design language learning with AI chatbots based on activity theory. Smart Learning Environments (2025)
  13. Bastani et al.: Generative AI without guardrails can harm learning. PNAS (2025)
  14. OECD: Digital Education Outlook 2026 (2026)
  15. Institute of International Education: Open Doors 2025 (2025)
  16. Mordor Intelligence: Online Language Learning Market (2026)
  17. GM Insights: Language Learning Market (2026)
  18. Grand View Research: Online Language Learning Market (2025)
  19. Future Data Stats: AI in Language Learning Apps Market (2026)
  20. Launch Base Africa: Lingawa pre-seed funding (2024)

Frequently Asked Questions

How big is the language learning market in 2025?

Estimates depend on what each firm counts. Mordor Intelligence valued the online language learning market at $21.06 billion in 2025, while GM Insights puts the wider language learning market at $14 billion. Future Data Stats values AI language learning apps specifically at $2.06 billion in 2025, forecast to reach $10.92 billion by 2033.

What percentage of language learners use AI tools?

No independent survey has measured this reliably yet. The clearest data comes from company filings: Duolingo, which builds AI into its lessons and conversation features, had 58.7 million daily and 140.6 million monthly active users in Q2 2026.

Is AI better than a human language tutor?

The research does not support replacing one with the other. Across 17 research projects, AI-guided learning beat normal instruction by a moderate margin (d = 0.39), with the clearest gains in vocabulary, reading and listening. Human tutors remain stronger for nuanced conversation, cultural context, accountability and spontaneous interaction. The most effective learners typically combine both.

How accurate is AI pronunciation feedback?

There is no independent, published accuracy standard for language apps, and the accuracy figures often quoted come from vendors without a stated test method. Heklin is testing six AI speaking apps with the same scripted pronunciation errors and will publish the results on this page.

Why do so many people quit language learning apps?

Mordor Intelligence estimates first-week churn above 60% for many freemium language apps. Common reasons are unclear progression, no accountability and the gap between app exercises and real conversation. Retention can be improved: Duolingo reported its highest-ever current user retention rate, 84%, in Q2 2026.

Similar Posts