Picture the night before an English interview. Your résumé is polished, your answers are written down, and there is nobody to say them to. You try talking to the mirror, but you can't tell how your English would land on a real person, and the evening slips by.
You can open Gemini Live, say "I'd like to practice for an interview," and a conversation starts. Left like that, though, the other side stays a friendly assistant, and it may simply praise everything you say. What I'd like to say up front is that the role has to be handed over in your first sentence. The reason is one line in the official Help page.
You can't save the role ahead of time
Google's Help page for Gemini Live lists a limitation: Gems can't be used with Gemini Live. A Gem is a saved set of instructions you can call up later. It's tempting to save an interviewer as a Gem and be done, but you can't call it from inside Live.
So the role is set in a spoken sentence, every session. If you assume it can be saved, you end up hunting through the Gems screen for something that isn't there; reading the limitations section first saves that time.
It also means you can change the difficulty every time you practice.
Three things to put in the first sentence
The pattern I'd suggest is to cover three things at once: the role, the pacing, and what to do when you can't understand.
- Role: what kind of company and job the interviewer represents
- Pacing: one question at a time, no evaluation until you ask for it
- When you can't catch something: rephrase it more simply instead of repeating it
I'm an indie developer who writes English store descriptions for my apps, and I've learned that reading a text silently rarely reveals how unnatural it sounds; saying it aloud does. Interview practice is the same kind of work.
Live is something you speak into your phone, so the setup has to be short enough to say out loud. Something like this:
I'd like to practice an English job interview.
You are an interviewer at a web app company.
Ask one question at a time, in English.
Don't evaluate my answers until I say "finished," and then tell me
what went well and what to fix, in Japanese.
If I can't follow you, rephrase in simpler English rather than repeating.You can give the whole thing in English. I'd still keep the setup in your native language and let only the conversation itself be in English, because it seems to reduce the chance of the role getting mixed up.
Why "no evaluation until I say finished"? If feedback lands after every answer, the flow of the interview breaks. A real interview doesn't grade you midway, so run it to the end first and ask for the review once.
You can cut in when you can't follow
When the interviewer's English is fast, many people just let it pass. The Help page says you can interrupt Gemini by speaking while it talks. Even if you've turned off "Interrupt Live responses" in settings, tapping the screen still interrupts.
So don't hesitate to stop it:
- Cut in with "Sorry, could you say that more simply?"
- Ask about a single word: "What does 'cross-functional' mean?"
- Ask for just that sentence in Japanese, then return to English
Asking again is part of the practice. "Could you repeat that?" is not rude in a real interview either.
Read the transcript afterward
According to the Help page, tapping End closes the session and lets you review a transcript of the exchange. That is the main advantage over practicing alone.
Don't try to fix everything when you read it. Three things are enough to look for:
- Sentences where the subject and verb don't line up
- A phrase you repeat many times ("I think," for example)
- Answers where the conclusion only appears at the very end
Fix one thing per session, ideally just where the conclusion sits.
One fix per practice session. I think that rule is what keeps rereading from becoming a chore.
Limits to know before you start
Here is what the Help page says, in the range I could confirm.
| What to check | What the Help page says |
|---|---|
| Devices | In this version of the guide, Android phones and tablets. Not available in the Gemini web app |
| Account | A personal Google Account, or a work or school account with access. Work or school accounts are limited to Google Calendar, Keep, and Tasks as connected apps |
| Age | 13 or older, or the applicable age in your country |
| Camera and screen sharing | Stop when Live is put on hold or the screen locks, and don't resume automatically |
I should also be plain about what the page doesn't say. It gives no retention period or deletion steps for Live conversations, only a link to a privacy hub. If you plan to mention your work history or the company you're applying to, check that first. For how stored context is handled, see the note on chats deleted while context remained.
Next step
Tonight, take five minutes. Read the short setup above aloud, answer a single question, tap End, and pick one point from the transcript.
Gemini's answers can contain mistakes too. If it teaches you a rephrased interview expression, have a look at how to read the orange sentences before sending. That habit stays the same even when the practice partner is an AI.