How to Choose an AI Tutor That Actually Improves Your Speaking

Key takeaways

• To choose an AI tutor, judge it on one thing first: how much you speak and how specific the correction is. Feature lists don't predict progress. Speaking time and feedback quality do.
• AI tutoring works, and the size of the effect is measured. In VanLehn's 2011 review in Educational Psychologist, intelligent tutoring systems produced an effect size of d = 0.76 compared with no tutoring, and human tutoring produced d = 0.79 — close to the same.
• General correction is worthless. "Good job" teaches nothing. Derwing, Munro, and Wiebe (1998) found gains in real conversation only when instruction targeted specific speech habits rather than vague practice.
• AI removes the biggest blocker for adults: fear. Horwitz, Horwitz, and Cope (1986) identified foreign language anxiety as a distinct condition, driven partly by fear of negative evaluation. An AI tutor has no audience to embarrass you.
• Test before you pay. Ten free minutes tells you more than 10 reviews. The questions below are written so you can answer them during a trial.

Search "AI tutor" and you'll get 40 tools promising fluency. They all list the same features: personalized lessons, instant feedback, 24/7 access. After 20 minutes of comparing, most people give up and pick whichever one had the nicest website.

That's a bad way to spend money. AI tutors differ enormously in one respect that marketing pages rarely mention: how much you actually talk, and how precisely the tool tells you what to fix.

This guide gives you 10 questions to ask, a comparison of AI tutors against other options, the red flags worth walking away from, and a realistic price range. You can work through it during a free trial.

How we built these criteria: the 10 questions map to factors with published evidence behind them — tutoring-system effectiveness (VanLehn, 2011), feedback specificity (Derwing, Munro, and Wiebe, 1998), level alignment (CEFR descriptors, Council of Europe), and language anxiety (Horwitz et al., 1986). Features with no evidence behind them, like streak animations, are not on the list.

What is an AI tutor?

An AI tutor is software powered by artificial intelligence that teaches you through interaction — it asks, listens, corrects, and adapts — instead of just showing you content to study. In language learning, these systems are designed to support learners through back-and-forth practice, usually with instant feedback on pronunciation, grammar, and word choice.

The category has three common shapes:

1. Voice conversation tutors — you speak, the tutor responds and corrects. Best for fluency and confidence.
2. Structured course tutors — an AI layer on top of a graded syllabus, where guided lessons and feedback can improve the learning experience as practice follows a real curriculum.
3. General chatbots repurposed for practice — flexible, but with no curriculum, no level tracking, and no pronunciation scoring.

For a fuller definition and how the underlying technology works, see the Promova guide to AI-powered language learning.

Do AI tutors actually work?

Yes, and the evidence is more specific than most marketing suggests. VanLehn's 2011 review of experiments comparing human tutoring, computer tutoring, and no tutoring found an effect size of d = 0.76 for intelligent tutoring systems versus d = 0.79 for human tutors. Both were far above no tutoring at all.

Two caveats matter when you're choosing:

1. The finding applies to systems that interact step by step, not to tools that just mark an answer right or wrong. In the same review, answer-based systems performed much worse than step-based ones.
2. The research predates conversational AI. Modern voice tutors weren't tested in that review, so treat d = 0.76 as evidence that interactive tutoring software can work — not as a promise about any specific app. Evidence also suggests combining AI with human tutoring can improve learning outcomes more than relying on either one alone.

There's a second, quieter benefit. Horwitz, Horwitz, and Cope (1986) showed that foreign language anxiety is a distinct construct, measured on a 33-item scale, and that fear of negative evaluation is a core part of it. An AI tutor has no face, no impatience, and no memory of how badly you said the sentence last time. For a lot of adults, that's the difference between practicing and not practicing.

How do you choose an AI tutor for your learning style? 10 questions to ask

Answer these during a free trial so you can compare the key features that actually matter. Each one takes under two minutes to check.

1. How many minutes did I actually speak? Open the tool, run one session, and time your own speaking. If you spent most of it reading, tapping, or listening, it's a course with an AI feature, not an AI tutor.
2. Does the feedback name the mistake? "Try again" is useless. You want "you stressed the second syllable; it should be the first." Specific correction is what the 1998 pronunciation study found transfers to real conversation.
3. Does it score pronunciation, or only grammar? Many tools check your words and ignore how you said them. If speaking is your goal, pronunciation feedback isn't optional. Test it by saying a word wrong on purpose.
4. Is it tied to a real level system? Look for CEFR levels — A1 through C2 — not vague labels like "beginner" and "advanced." Without a framework, you can't tell whether you're progressing or circling.
5. Can I choose the topic? A job interview, a doctor's visit, a rental dispute. Generic small talk is comfortable and mostly useless. Relevance is what makes you keep going.
6. Does it remember me between sessions? A good tutor knows the words you learned last week and the sound you keep missing, and that memory should reflect your progress, current knowledge, and learning goals instead of resetting each time. If every session starts from zero, you're paying for a demo.
7. Does it interrupt me or wait? Constant correction kills fluency; zero correction kills accuracy. The better tools let you finish the thought, then give feedback. Check which mode you get, and whether you can switch.
8. What happens when I say something wrong in a way it doesn't expect? Say something deliberately garbled. A weak tool praises you or invents a correction. A good one asks you to repeat.
9. What does it do with my voice data? Voice recordings are personal data, so privacy checks matter because these systems may process student data or other personal learning data. Check whether the privacy policy explains retention and whether the company is subject to rules like the GDPR. "We use your data to improve our models" with no opt-out is a reason to look elsewhere.
10. What's the real price after the trial? Note the monthly rate, the annual rate, what's locked behind premium, and whether speaking minutes are capped. Capped speaking is the single most common hidden limit.

If you'd rather start from a shortlist than a checklist, the Promova roundup of the best AI tutors compares specific tools by feature.

2

AI tutor vs. human tutoring vs. chatbot: what's the difference?

All three can help. They fail in different places.

Some AI tools also help teachers by handling repetitive tasks, though that matters less if you're choosing a speaking tutor for yourself.

The practical answer for most adults: use an AI tutor for volume — daily reps, low friction, no scheduling — and add human feedback when you hit something an algorithm can't judge, like tone in a difficult work conversation.

How much should an AI tutor with key features cost?

As of mid-2026, dedicated AI language tutors generally sit between $8 and $25 per month, with annual plans usually 40–60% cheaper per month than rolling monthly ones. Some platforms also use a free plan with basic features, while richer limits or usage unlock in premium tiers. Anything far above that range is normally bundled with human lessons.

Three things to check before paying:

1. Are speaking minutes capped? A cheap plan with 10 minutes a day may be fine, or may be the whole product.
2. Is the annual discount real? Compare what is included for free versus paid before judging value, then compare the annual total against 12 monthly payments, not against the "was" price.
3. Can you cancel inside the trial? Set a calendar reminder for the day before it converts.

Free versions are worth using for exactly one purpose: answering the 10 questions above before you commit money.

Who benefits most from an AI tutor?

AI tutors suit some situations much better than others.

1. People with unpredictable schedules. If you can't commit to a fixed weekly slot, a tool with no booking wins by default.
2. Learners who freeze in front of people. This is the anxiety case described by Horwitz and colleagues — and the clearest use for a judgment-free partner.
3. Intermediate learners who understand but can't produce. You have the words. What you lack is reps. AI is very good at reps, especially for independent learners who want extra practice between classes.
4. People preparing for one specific conversation. An interview, a presentation, a visa appointment. Role-play the exact scenario until it stops feeling strange.
5. Anyone rebuilding a habit. Ten minutes with no scheduling friction is easier to repeat than a 60-minute commitment, and it can support the wider learning journey by letting you revisit concepts at your own pace.

It's a weaker fit if you need certified exam preparation, detailed written essay feedback, or a language with very little training data behind it, and fit also depends on your learning style.

What are the red flags when choosing an AI tutor?

Walk away from any of these.

1. Fluency promises with a timeline. "Fluent in three months" is a marketing claim, not a learning outcome, and vague promises about education results should not replace evidence. Progress depends on hours and consistency.
2. Praise instead of correction. A tutor that never tells you you're wrong isn't teaching. Test this in the first session.
3. No level framework. If the app can't tell you where you sit on a recognized scale, it can't show you progress.
4. Text-only "speaking" practice. Typing is not speaking. Your mouth needs the reps.
5. No pronunciation feedback at all. Then it's a conversation partner, not a tutor, and it should guide students through mistakes instead of only giving answers.
6. A privacy policy you can't find. Voice data is sensitive. Silence is an answer.
7. Streaks as the main feature. Gamification supports habits, but a long streak is not the same as a higher level.
8. Reviews that only praise the interface. Look for reviews that mention feedback quality, deeper understanding, or problem solving, not just praise for design or specific improvements.

How does Promova's AI Tutor score on these 10 questions?

1. Speaking time — the AI Tutor is voice-first: sessions are conversations, not reading exercises.
2. Specific feedback — the pronunciation system highlights which part of a word was wrong, rather than returning a pass or fail. You can hear a model recording, record yourself, and compare at adjustable playback speed. It also uses Socratic questioning to guide you with prompts instead of jumping straight to answers, which helps build critical thinking.
3. Pronunciation scoring — yes, built into lessons and speaking practice. Background reading: English pronunciation practice.
4. Level framework — courses are CEFR-based, A1 to C1, and there's a level test to place you.
5. Topic choice — Speaking practice with AI covers role-plays like job interviews, doctor's visits, and ordering food; Practice with AI builds sessions around vocabulary you've just studied.
6. Memory between sessions — the learning plan adjusts to weak points and pulls previously learned words back into practice. According to Promova's own platform data, 79% of learners say AI recommendations helped them learn faster.
7. Correction timing — feedback arrives after the exchange, so you can finish a thought before being corrected.
8. Handling unexpected input — role-play stays in character and asks for clarification rather than accepting garbled input.
9. Data handling — see the Promova privacy policy for voice data retention. (Add direct link at publish.)
10. Price — free to download, with several tools free permanently, including the pronunciation tool. Premium plans start at $9.99 per month and unlock unlimited speaking practice plus 4,000+ lessons.

Promova's own usage data indicates 84% of learners feel more confident in conversations after practicing with AI.

For a routine that uses these tools together, see how to learn a language efficiently with Promova and language learning with AI. Learning Spanish instead? Start with how to learn Spanish with AI.

Want to run the 10 questions yourself? Download the Promova language learning app and test it in one session.

Final Thoughts

Choosing an AI tutor comes down to a test you can run in 10 minutes: did I speak, and did it tell me exactly what to fix?

Everything else — streaks, avatars, lesson counts — is decoration. The research is reasonably clear that interactive tutoring software works when it engages you step by step and corrects you specifically. Tools that only show you content, or only tell you "great job," aren't doing the thing that produces progress.

Run the 10 questions on two or three options during their free trials. You'll know within a session, and you'll spend your money once instead of three times.