An AI interview assistant does not decide on its own whether a question wants code or an explanation; you tell it, and the telling takes one keypress. With Coding mode on (Ctrl+Shift+C), every question returns complete code in your chosen language with an example run. When the interviewer asks something conceptual, such as "why did you pick a hash map?" or "what is the time complexity?", you press 🗣 Talk answer (Ctrl+Shift+E) and the next question is answered in words while Coding mode stays on for the question after it.
That is the whole mechanism, and it is deliberately simple. A model that tried to infer "code or words" from the question alone would get it wrong often enough to hurt you at the worst moment. A toggle you control never does. The rest of this page explains what each mode produces, how the screen-reading features fit in, and how a typical round plays out.
Why the tool does not guess
Coding rounds mix two kinds of question that sound alike. "How would you find duplicates in this list?" might want a function, or it might want you to talk through the trade-offs between sorting and hashing. The interviewer knows which; the audio alone does not say. If an assistant guessed wrong and showed you a block of code while the interviewer wanted a sentence about complexity, you would lose time reading something irrelevant.
So CoPilot Interview treats it as a state you set. Coding on or off is the round-level setting. Talk answer is the one-question override. You stay in control of the shape of every answer, which matters because you are the one who has to say it out loud and defend it.
What Coding mode produces
With Coding mode on, a captured question returns:
- Complete code in your chosen language. The picker lists 50+ programming languages. Pick the one the interview uses before the call starts.
- An example run showing the function called on sample input, so you can talk through what it does.
- Optional comments. The 📝 Note option adds a 2-4 line approach and complexity comment at the top, useful when you want to narrate your plan before typing. 🚫 No comments gives clean code when you would rather explain aloud.
Everything appears as text in the app's own window on your screen. You still type the code into the shared editor yourself, and you should type it in your own style, variable names included, because you will be asked to change it.
What Talk answer produces
Press 🗣 Talk answer (Ctrl+Shift+E) and the next question is answered in words instead of code: key points first, then detail, the same structure as a behavioral answer. Typical conceptual follow-ups in a coding round and the kind of answer you get (examples we wrote):
- "Why a hash map here?" One-sentence reason (constant-time average lookup), the trade-off (extra memory, no ordering), and when you would choose something else.
- "What is the complexity?" Time and space stated separately, with the one line of reasoning that justifies each. Our Big-O complexity visualizer is a good thing to skim the night before so these answers sound like yours.
- "How would this behave with a million items?" Where it scales, where it breaks, and the first change you would make.
After that one answer, the mode returns to code. If the interviewer asks two conceptual questions in a row, press Ctrl+Shift+E again.
Reading the shared editor and slides
Sometimes the question is not spoken at all; the interviewer pastes a problem statement or a buggy function into the shared coding editor and says "take a look." Three features cover that:
- Read screen (Ctrl+Shift+S) captures what is on the screen, a shared editor or a slide, and answers what is shown.
- Pin screen keeps a captured region as context so later answers refer back to it.
- 🐞 Debug mode (Ctrl+Shift+D) finds bugs in code captured from the screen, which is the usual shape of a "what is wrong with this?" question.
Global hotkeys work while your cursor stays in the editor, so you do not have to click into the app window. Turn on click-safe (Ctrl+Shift+F) and the window never takes focus from the app you are typing in. Our coding interview copilot page goes into these screen features in more depth.
Which button for which question
| Interviewer says | What you press | What appears |
|---|---|---|
| "Write a function that returns the k most frequent words." | Nothing (Coding mode already on) or ⚡ Answer | Complete code in your language, example run |
| "Walk me through your approach before you code." | 🗣 Talk answer (Ctrl+Shift+E), or turn on 📝 Note | Words: approach and complexity, or a 2-4 line comment at the top of the code |
| "Why a heap instead of sorting?" | 🗣 Talk answer (Ctrl+Shift+E) | A spoken-style explanation, key points first |
| "What is the complexity?" | 🗣 Talk answer (Ctrl+Shift+E) | Time and space, each with one line of reasoning |
| "Here is some code. What is wrong with it?" | 🐞 Debug (Ctrl+Shift+D) then Read screen (Ctrl+Shift+S) | The bugs found in the captured code |
| "Now handle the case where the input is empty." | Ctrl+Shift+Q for a new question, then ⚡ Answer | Revised code |
| "Tell me about a project where you used this." | Ctrl+Shift+C to turn Coding off, or 🗣 Talk answer | A behavioral answer referencing your pasted résumé |
A worked round (example we wrote)
Coding mode is on, Python selected. The interviewer asks for a function that merges overlapping intervals. The code appears in about 2 seconds with a sample call. You say "I will sort by start, then sweep and merge," and type it in your own words. The interviewer asks "what is the complexity?" You press Ctrl+Shift+E; the answer says O(n log n) for the sort, O(n) for the sweep, O(n) extra space for the output, and you repeat that in a sentence. The interviewer pastes a version with an off-by-one bug into the editor and asks what is wrong. You press Ctrl+Shift+D, then Ctrl+Shift+S; the app points at the comparison. You say where the bug is and fix it on screen. Finally they ask about a time you debugged something under pressure; you press Ctrl+Shift+C to leave Coding mode and answer from your own history.
Nothing in that round required clicking into the assistant. The real-time interview copilot sat in a separate desktop window on your own screen the entire time, and everything you said was in your words.
What it cannot do for you
It cannot type for you, and it cannot tell what the interviewer meant by an ambiguous question. If "how would you find duplicates?" is really asking for a conversation, code on your screen is a distraction; ask "do you want me to code this or talk through it first?" That question is normal in a coding round and interviewers like it. The assistant also does not know your interviewer's preferred language or style; pick the language before the call and expect to adapt variable names and structure. Practicing the split in the AI mock interview, which asks both coding and conceptual questions from your pasted job description, is the fastest way to make the two keys automatic. If you are choosing a tool, compare AI interview assistants on exactly this: whether code and words are a mode you control or a guess the model makes. The broader AI interview assistant handles the behavioral rounds too, with the same key-points-first structure.
Follow the rules of the employer or assessment you're in — many people use it only to practice, which works on the free plan.
Practice the code-or-words switch before it counts
The AI mock interview asks coding and conceptual questions from your own job description. Try the two keys on the Free plan, 15 minutes a day, no signup.
Start a mock interviewFAQ
Does the AI interview assistant pick the programming language itself?
No. You choose the language in the picker before or during the session, and there are 50+ programming languages to choose from. Pick the one the interview will use so the code matches what you type into the shared editor.
What happens if I forget to press Talk answer for a conceptual question?
With Coding mode on, the answer comes back as code. Press Ctrl+Shift+R to redo after pressing Ctrl+Shift+E, or type the question yourself in the app. Nothing is lost; it costs you a few seconds.
Can it explain the complexity of code I already wrote?
Yes. Use Read screen (Ctrl+Shift+S) to capture the editor, then press Talk answer (Ctrl+Shift+E) and ask about complexity. The answer states time and space separately with a line of reasoning for each.
Does Debug mode fix the code or just find the bug?
It finds bugs in code captured from the screen and explains them. You make the change in the shared editor yourself, which is what the interviewer wants to see anyway.
Can I use Coding mode on the free plan?
Yes. The Free plan includes all modes, including Coding mode, Talk answer, Read screen and Debug, for 15 minutes a day in sessions of up to 15 minutes with the free AI models.
Does the assistant see what the interviewer typed in the shared editor?
Only when you ask it to. Read screen (Ctrl+Shift+S) captures what is currently shown on your screen, and Pin screen keeps that capture as context for later answers. It does not watch the screen continuously.