Large language models can produce fluent tutoring dialogue, but whether they actually help students learn remains unclear.
We present a process-centered synthesis of LLM tutoring that follows the path from learner-state inference to pedagogical action, tutor training, evaluation, and student simulation.
This view exposes a central gap: implementing a tutoring component does not establish that it works or that its outputs meaningfully affect later stages.
A structured study-level audit shows that pedagogical evaluation is common, while evidence becomes much thinner for learner-conditioned intervention, human learning gains, and especially transfer or delayed retention.
Human studies nevertheless provide a conditional positive result: LLM tutors can improve learning, but benefits depend on pedagogical structure, learner engagement, and how learning is measured.
We argue that progress requires validating the interfaces between tutoring stages, not only improving individual components.