Does learning English with AI really work?

Yes, it works — and science confirms it. The meta-analysis by Zou et al. (2023), which reviewed 67 empirical studies on AI in language teaching, found a measurable positive effect compared to traditional methods. The effect size varies by skill: larger in vocabulary and grammar, medium in listening and speaking, smaller in writing. AI does not replace the correct method, but accelerates when used properly.

Is AI better than an English teacher?

No — not yet. Research shows that AI is an excellent complement but does not fully replace a qualified human teacher. What AI does better: 24/7 availability, massive practice volume, instant correction, low cost, zero embarrassment. What a human teacher does better: adapting to emotional context, personal motivation, culture, social nuances, and fine diagnosis of error causes.

What does research say about conversational chatbots?

The study by Belda-Medina & Calvo-Ferrer (2022) with 129 students showed that conversational chatbots reduce language anxiety (affective filter, Krashen's concept) and increase the volume of speaking practice. The effect is greater in introverted students or those with social anxiety — who typically do not practice with a human teacher. The average weekly practice volume doubles when a chatbot is available.

Does AI help with pronunciation correction?

Yes. Models like OpenAI's Whisper (Radford et al., 2022) achieve a word-error rate below 5% in non-native English speakers, which is comparable to human transcription. This allows for real-time pronunciation feedback — showing the student which phonemes are deviating from the native standard. Specific studies on AI-based pronunciation tutors show gains of 15-25% in intelligibility after 8 weeks of daily use.

What is the risk of relying solely on AI?

Three main risks: (1) AI can generate plausible but grammatically incorrect text in rare cases (hallucination), so corrections need to be checked against reliable sources; (2) chatbots lack real cultural context — they don't know how to explain social nuances, taboos, informal registers; (3) passive dependence can reduce motivation to speak with real humans, which is where the language is truly tested. Therefore, AI should be a complement, not a substitute.

How long does it take to see results with AI?

It depends on the intensity and method. Studies with 30 minutes a day of AI (conversation + SRS + pronunciation) show measurable gains in active vocabulary after 4 weeks, in pronunciation after 6-8 weeks, and in overall fluency after 3-6 months. Results are comparable to those obtained with in-person private lessons twice a week, but at a cost 10-20 times lower.

Does AI work for beginners or only for intermediates?

It works better for intermediates (A2-B2). Total beginners (A0-A1) first need structured instruction — with explicit rules, basic vocabulary, and introductory pronunciation — before AI becomes useful. Once the student has a minimal foundation, AI accelerates significantly: they can formulate sentences, make mistakes, receive corrections, and iterate. For beginners, the recommendation is to use AI for vocabulary and listening, but grammar and initial speaking still benefit from a teacher or structured course.

Does Learning English with AI Work?

Q: What is the most important research on AI and language learning?

The meta-analysis by Zou, Huang, Wang, and Xie (2023) published in Computer Assisted Language Learning is the most cited reference. It aggregates results from 67 studies between 2015 and 2022, covering chatbots, voice recognition, machine translation, and text generation. Conclusion: AI has a significant positive effect on learning, especially in environments where the student lacks access to natural input of the language.

TL;DR · What science says

Learning English with AI works — and there's strong science validating it. The meta-analysis by Zou et al. (2023), which reviewed 67 studies published between 2015 and 2022, found a measurable positive effect compared to traditional methods. AI is especially good at vocabulary, pronunciation, and volume of speaking practice. It still doesn’t replace a human teacher in cultural context and motivation. In this post, I show what each study found, where AI shines, where it still fails, and how to use it the right way.

Look, if you're thinking, "Does learning English with AI is just another hype wave or does it really work?" that’s the right question. And there’s a validated answer — the science of language learning has been studying this seriously since 2015. This post is about what the research really shows: where AI works, where it doesn’t, and how to use it to maximize results.

I’ll cite 5 main peer-reviewed studies, show a graph comparing AI vs traditional methods, and at the end, I’ll give you a data-driven recommendation — not just an opinion.

The reference meta-analysis: Zou et al. (2023)

The most cited study in 2024 and 2025 about AI and language learning is the meta-analysis by Zou, Huang, Wang, and Xie published in Computer Assisted Language Learning. They aggregated results from 67 empirical studies published between 2015 and 2022 — covering chatbots, voice recognition, machine translation, text generation, and personalized tutoring.

The conclusion: AI has a significant positive effect on learning compared to traditional methods. But the effect size varies by skill. In passive skills (reading, vocabulary), the effect is large. In productive skills (speaking, writing), the effect is moderate — still positive, but smaller. And the gain crucially depends on the method of how AI is used, not just the fact of using AI.

Where AI shines: the skills graph

Efficacy of AI by skill · meta-analysis Zou et al. (2023)

Effect size (Cohen's d) comparing groups with AI vs without AI — 67 studies aggregated

AI wins where feedback is measurable. Loses where cultural context is needed.

The reading of the graph is straightforward: AI works very well where feedback is objectively measurable (is the word right or wrong, is the pronunciation close or far from the standard). And it works poorly where feedback is subjective and cultural (is it polite? is it appropriate for the context? does it sound natural?).

5 areas where AI really works

1. Vocabulary and SRS (d=0.82)

The area where AI has the strongest effect. The reason is simple: AI can apply spaced repetition (SM-2 by Wozniak) automatically, generate contextualized examples on demand, and adapt difficulty to the user. The famous study by Vesselinov & Grego (2012) showed that 34 hours of Duolingo were equivalent to 1 semester of university classes in terms of vocabulary gain.

2. Pronunciation correction (d=0.75)

The Whisper by OpenAI (Radford et al., 2022) has a word-error rate below 5% in non-native English speakers — equivalent to human transcription. This allows AI to show the student, in real-time, which phonemes they are getting wrong. Specific studies on AI-based pronunciation tutors show gains of 15-25% in intelligibility after 8 weeks of use.

3. Volume of speaking practice (d=0.52)

It’s not the largest effect overall, but it’s the most transformative in day-to-day life. Belda-Medina & Calvo-Ferrer (2022) showed that conversational chatbots doubled the weekly volume of oral practice for students compared to just having lessons with a teacher. And they reduce language anxiety (Krashen's concept), which is the main enemy of introverted students.

4. Tailored comprehensible input (d=0.68 in listening)

Stephen Krashen proposed in 1985 that language is acquired when the student receives input slightly above their current level (i+1). The problem has always been that finding material at the right level is difficult. AI solves this — it generates texts and audio adapted to the student's level or simplifies existing material. Recent research (Godwin-Jones, 2022) calls this "partnering with AI" for personalized input.

5. Instant feedback (applicable to everything)

Hattie & Timperley (2007), in one of the most cited education papers in history, showed that instant feedback is one of the biggest predictors of effective learning in any domain. AI delivers this 24/7. You miss a word, get correction in 200ms, and practice again immediately. With a human teacher, this loop takes hours or days.

3 areas where AI still fails

1. Cultural context (d=0.12 — insignificant)

AI doesn’t know when something is appropriate or offensive in real social context. It doesn’t know the difference in register between "whassup" and "how do you do". It doesn’t understand specific cultural humor. This is the territory of a human teacher or real immersion.

2. Emotional motivation

AI doesn’t hug you when you’re about to give up. It doesn’t know you had a bad day. It doesn’t cheer when you progress. Students who only use AI tend to give up sooner than students with a human tutor involved — because they lack the emotional bond.

3. Fine diagnosis of causes

When a student makes a mistake, AI usually corrects the form but doesn’t identify the root cause of the error. An experienced teacher notices: "oh, you’re translating from the structure of Portuguese." AI hasn’t reached this level of meta-diagnosis on most platforms yet.

How to use AI the right way (based on evidence)

Use AI as a volume engine. It frees the human teacher to focus on what they do best (diagnosis, motivation, culture). 5h of AI + 1h with a teacher > 6h only with a teacher.
Prioritize the skills where AI wins. Vocabulary with SRS (d=0.82), pronunciation with feedback (d=0.75), conversation without social pressure (d=0.52). Don’t waste time trying to learn culture or creative writing only with AI.
Combine with real input. Series, podcasts, books, news. AI does not replace exposure to the language "in real life".
Do not ignore explicit grammar at the beginning. If you are a beginner (A0-A1), take a quick structured course before relying solely on AI. After A2, AI accelerates.
Measure your real hours. The FSI table still applies: 600-750 hours for B2/C1. AI cuts the time per hour, but does not erase the hours.

The data-driven recommendation

Learning English with AI works — especially for Brazilian adults who work, don’t have time for in-person classes, and need measurable results. Studies clearly support the use of AI as a primary tool, provided it is combined with: (1) the correct method based on comprehensible input, (2) daily active production, (3) SRS for vocabulary, (4) consumption of real content in English.

Those who rely only on AI and ignore real content or active production won’t go far. Those who use AI as a volume engine and combine it with scientifically validated principles learn faster and at a lower cost than ever before. That’s the thesis behind Lanna.

Frequently Asked Questions

Will AI replace English teachers?

Not completely, but it changes the role. The teacher becomes more of a coach, less of a content transmitter.

What is the strongest study on AI and languages?

The meta-analysis by Zou et al. (2023), which aggregated 67 studies.

Does AI work for children too?

It works, but children learn better through natural immersion. For children, AI is a complement; for adults, it can be the base.

How long will it take to see results?

In 4-6 weeks with consistent use (30 min/day), measurable gains are already possible.

What is the biggest risk?

Passive dependence — using AI as a crutch and never testing in real situations with humans.

Start today with method

If you’re convinced that AI is worth it (science agrees), the next question is how to use. The research-based recommendation: 30 minutes a day combining voice conversation (d=0.52), vocabulary SRS (d=0.82), pronunciation with feedback (d=0.75), and tailored comprehensible input (d=0.68). In 6 weeks, you’ll feel the difference.

Research-based AI, not hype

Lanna combines everything the meta-analysis identified as effective: SRS + pronunciation tutor + speaking with AI + tailored comprehensible input. All in one flow.

Try Lanna for free

Does Learning English with AI Work? What Science Says in 2026