AI Video Practice
Use AI as a low-pressure practice partner for behaviors people avoid in public.
In practice: Duolingo
AI video practice moves the scariest part of language learning, speaking, into a private low-stakes call with a familiar character. It stays short and safe to fail, so fluency builds before there is a human on the other end.
Calling a familiar character
The Video Call screen invites the learner to phone an established character like Lily for XP. Practicing with a known personality feels like part of the product world, not a cold utility bolted on.
A real conversation, zero risk
During the call the character responds in real time, so the learner can stumble, repeat, and retry without a raised eyebrow from a real listener.
Practice beside the lessons
The conversation hub offers Video Call and Roleplay alongside regular skill practice, so speaking rehearsal is one option among many rather than a high-stakes test.
Why it works
When criticism feels possible, the brain protects instead of learns. A judgment-free AI partner keeps learners in the receptive state where mistakes become information, and short bounded sessions keep practice frequent instead of intimidating.
How Duolingo does it
Private Rehearsal
Up to 70% of language learners report speaking anxiety. A call with an AI character lets them stumble, repeat, and retry without a raised eyebrow from a real listener.
Short, Bounded Sessions
Calls are capped at about three minutes and pay out XP, which keeps practice frequent and low-commitment instead of a long test of nerve.
Behavioral principles
Psychological Safety
A judgment-free partner keeps learners in the state where mistakes become information rather than embarrassment.
Optimal Arousal
Real conversation with no social risk lands in the sweet spot: engaged enough to attend, not stressed enough to shut down.
Character Framing
A familiar character makes the AI feel like part of the product world, so learners pick a partner that matches how they like to be pushed.
Situated Memory
Roleplay puts words into scenarios, giving learning a narrative context that sticks better than isolated facts.
Design considerations
Explain what the AI can and cannot evaluate so learners calibrate trust in its corrections, and design graceful repair moments for mishearing and awkward replies, or the safe space stops feeling safe.
Implementation prompt
Use the pattern above as the reference, then prototype the same behavior in your own product surface.
I want to incorporate AI practice sessions into my product, for a skill people avoid practicing in front of others. The goal is to move the intimidating part of the behavior into a private, low-stakes rehearsal where failing costs nothing. Please suggest how to implement this in the product: • Identify the moment users avoid because someone might be watching or judging • Offer that rehearsal as a short, bounded session with a clear end, not an open-ended test • Give the AI partner a consistent character and register, so it feels like part of the product rather than a utility bolted on • Let users stumble, repeat, and retry with no score attached to the attempt itself • Place the rehearsal beside normal practice as one option among several, never as a gate For implementation, propose: • Where the session belongs in the learning or practice flow • Session length, entry points, and what a completion is worth • Repair states for mishearing, silence, awkward replies, and repeated failure • Copy that states plainly what the AI can and cannot evaluate, so users calibrate trust in its feedback • How to measure whether rehearsal actually transfers to the real, unassisted behavior.