From prepared answers to real responses
Speaking became the skill I explored most carefully. I noticed that memorising a polished answer could create a false sense of confidence. Once the question changed, the prepared sentences might no longer fit. I wanted to respond to a person, rather than search my memory for the nearest script.
Studying the speaking assessment criteria helped me give my practice a clearer direction. I worked on developing a main point with a reason and an example. I also used phone recordings and AI feedback. These tools gave me more chances to notice problems and try again, without the pressure I associated with classroom performance.
The emotional difference mattered. Speaking to AI felt less like being judged by another person, so I could hesitate or repeat an answer more freely. This did not make my English perfect, but it made practice easier to begin. I was gradually learning to treat an unfinished sentence as something I could improve, rather than evidence that I lacked ability.
A more focused routine to develop
Reflecting on these experiences, I have designed a more focused routine that I want to develop further. The diagram below summarises it. It is a plan built on my experience with recordings and AI, rather than a record of a completed experiment.
First speak without a model answer
I would begin with one familiar question, such as whether I prefer studying alone or with others. I would record a spontaneous answer for about one minute without reading a script. My intensive course and later independent study give me real material to discuss. I could explain that independent study allowed me to replay difficult material and adjust my pace.
A prompt to try
This prompt keeps the conversation connected with my own meaning. If AI supplies a polished answer before I speak, I may simply borrow its ideas. Speaking first allows me to discover what I can express independently and what I need help with.
Then review a small number of problems
Next, I would listen to the recording and check any automatic transcript against the audio. A recognition error should not become a mistake I believe I made. I would ask AI to identify one repeated expression, one unclear idea and one grammar issue that affects meaning, with a brief Chinese explanation and small English corrections.
Limiting the feedback gives me something manageable to work on. A completely rewritten paragraph may be fluent but use language I cannot reproduce. Chinese helps me understand the explanation, while English gives me wording to practise. Using both languages here supports learning rather than separating it into two competing systems.
Add a real example and record again
If an answer needs more detail, I would ask AI to question me about my actual experience before suggesting changes. For the training camp, it could ask about the timetable and why it did not suit me. The detail already belongs to my story: classes began at eight, and studying often continued until two the next morning.
After reviewing the feedback, I would close the suggested answer and record again using only three keywords. Then I would try a related question with different content. This could show whether I can use the structure flexibly, rather than memorise one improved response. A clearer main point and a more relevant example would matter more to me than speed alone.
Keeping responsibility for my voice
For pronunciation, I would work on one short sentence, listen to a reliable audio example and record myself repeating it. If a tool receives only a transcript, it cannot assess how I actually pronounced the sentence. Even when audio feedback is available, I need to consider whether the advice is clear and useful.
AI feedback also has limits. An estimated score is not an official result, and a confident correction may need checking. I still need human conversations, where someone can interrupt, misunderstand or respond emotionally. Those exchanges involve listening and relationships in ways that a controlled exercise cannot fully reproduce.
For me, technology works best when it supports a cycle of attempting, noticing and adjusting. I want to sound understandable and express my experience, rather than perform a voice that does not feel like mine. That growing confidence connects my identity as a learner with my later decision to teach.