The challenge teachers face when AI tutors are everywhere: students who have unlimited access to AI ESL tutors still ask for real conversation partners. They could use ChatGPT for hours a day. They could pair with an AI voice tool any time they want. They still email their teacher asking about language exchanges, conversation clubs, and in-person speaking practice.
This isn't because the AI tutors are bad. It's because something is happening in human conversation that AI conversation can't replicate, and students notice it even when they can't articulate it. This post unpacks what that something is and what it means for designing ESL lessons.
YapYapGo is a classroom speaking practice tool for ESL teachers, designed around the in-person pair conversation that students keep asking for. This post is about what makes human conversation preferable to AI conversation in the eyes of motivated learners.What students actually say
In conversation with ESL students who've used both AI tutors and real conversation partners, the same themes keep coming up:
- "I want to know if they actually understand me." Students notice that the AI says it understands but doesn't really react. A real person's face tells them whether the meaning landed.
- "I want to make my friend laugh in English." The social goal of humour in the target language is impossible to satisfy with AI. The AI doesn't laugh genuinely; the student knows it.
- "I want to talk about something we both care about." Shared interests, current events, the show both students just watched - real human interlocutors share contexts that AI can't.
- "I get bored with the AI after a while." Even patient students saturate. The lack of social variation makes AI conversations feel repetitive even when the topics vary.
- "I want to be recognised by my classmates." The status game of being known as the student who speaks well, who's funny, who's confident - this only exists in social contexts.
These are not whims. They're motivational signals pointing at the things AI conversations structurally can't deliver.
Why social reward matters for sustained practice
Self-Determination Theory (Deci and Ryan) identifies three psychological needs that drive intrinsic motivation: autonomy, competence, and relatedness. We covered this in the post on gamification and SDT. The relatedness need is the social one - feeling connected to other people through the activity.
For most learners, relatedness is the durable engine of language learning. The student studies because they want to talk to their host family, their international friends, their potential colleagues. The end state of the work is social. The work itself, if it's going to sustain, has to be social too.
AI conversations don't satisfy relatedness. They satisfy competence (the student feels they're making progress) and weakly autonomy (the student chooses what to practise). Relatedness is structurally absent because the AI is not a relatable agent. Even when the AI is friendly, the student knows the warmth is performed by software.
Over time, the absence of relatedness causes drop-off. Students who exclusively practise with AI tutors often peter out within months. Students who have human conversation partners persist longer because the relatedness need keeps pulling them back.
This is the same mechanism behind the Willingness to Communicate research - social presence is what gets students to speak in the moment. It's also what keeps them coming back across years.
The peer modelling effect
A real classroom has another effect AI can't replicate: peer modelling. Students see what other students at their level can do. They notice when a classmate uses a structure they didn't know. They overhear better answers and absorb them. They see what fluent ESL students sound like - not the idealised AI voice, but the realistic version of "where I could be in six months".
This is enormously motivating. The AI tutor presents a target that's always slightly out of reach (the AI's fluent English) and never shows the intermediate steps. The peer in pair work is exactly the intermediate step. Students learn from their classmates what realistic progress looks like.
This works particularly well when classes are mixed-level. Students in the lower half of the class see what students in the upper half can do. The vision of achievable progress becomes concrete instead of theoretical.
The status game
This is less discussed than the other motivational levers, but it's real: students compete for status within their class through their English ability. Speaking well in front of classmates is a social achievement. Being funny in the second language carries higher status than being funny in the first. The student who can crack a good joke in English is a class figure in a way that no AI conversation could create.
The status game is not always healthy - it can suppress speech from shy students who fear ranking low. But for the median student, it's a powerful motivator. The desire to perform well in front of peers drives preparation, practice, and engagement.
AI tutors are zero-status environments. Nobody cares how you did. The student's performance is private. The lack of social stake reduces motivation. (The companion piece on Willingness to Communicate covers how team formats can preserve the motivational benefits while protecting the shy students.)
What this means for ESL programmes
A few design implications:
- Maximise pair-time. Every minute of pair work is a minute of relatedness, peer modelling, and status-game-playing. The parallel/serial speaking maths is also a relatedness argument.
- Vary pairs across sessions. Different partners means different relationships, different status dynamics, different peer-modelling exposure. The Team Maker makes this trivial.
- Use teams for shared goals. Team competition gives students something to win together. Shared goals create the relatedness condition more powerfully than individual practice.
- Acknowledge social achievement. When a student speaks well, name it. The status game runs better with explicit recognition.
- Place AI in support roles, not as the main partner. The WTC research and the team-sport framing both point at AI being supplement, not centre.
You can structure all of this with the Team Maker, Topic Generator, and Classroom Timer. The setup is minimal; the social benefits compound across the term.
The bottom line
Students who use AI tutors still ask for human conversation partners because AI can't satisfy the relatedness need that drives sustained language learning. The social reward of being understood, of making a classmate laugh, of holding status through English ability - none of these exist in AI conversation. Real classroom pair work provides them automatically. Listen when your best students ask for more conversation partners; they're telling you what their motivation needs.
Sources:
- Deci, E. L., & Ryan, R. M. (2000). Intrinsic and extrinsic motivations: Classic definitions and new directions. Contemporary Educational Psychology.
- Dörnyei, Z., & Ryan, S. (2015). The Psychology of the Language Learner Revisited. Routledge.
- Ushioda, E. (2009). A person-in-context relational view of emergent motivation, self and identity. Motivation, Language Identity and the L2 Self.