Which tool to choose
You don't need a dedicated tool: the two assistants you already use do role-play better than the paid apps. The choice depends on the device and the type of practice.
You want to speak out loud and simulate the pressure of being face to face → ChatGPT with voice mode. It offers practice with adaptive questions in real time, helps you improve on-the-spot reasoning and refine your answers. It's the most natural one for those who want to get used to the rhythm of a real conversation.
You're on Android and want a free session without tight limits → Gemini Live. It's useful for conversational interview role-play: you give it the role, the company context, and the type of interview, then you ask it to be the interviewer and give structured feedback; it works well for recruiter screening, behavioral questions, elevator pitches, and follow-up questions.
You only want to prepare and refine written answers, without speaking → the text chat of either of the two. Slower as a simulation, but useful for building the outline before moving on to voice.
For most people ChatGPT in voice mode is the right choice. Both modes are free to try before moving to a paid plan: that's enough to figure out whether it's for you.
How to do it
The principle is one: you don't ask "ask me some questions," you assign a precise role to the AI and give it the rules of the game before starting. Neither ChatGPT nor Gemini are platforms dedicated to interviews: they give their best when you bring the structure, that is, a clear prompt, a job description, and an evaluation grid.
Open the up-to-date app or site and look for the microphone or headset icon in the input bar: that's where the voice conversation starts, identical on phone and on computer.
Before moving to voice, in a text chat paste the prompt below. It serves to "load" the context, because voice mode has a shorter memory.
Paste into the prompt your target role, the company, and — if you have it — the job posting or the slides of the presentation.
Start voice mode and say "let's begin." Answer by speaking, as you really would.
At the end, ask for feedback. If you set the rules well, the AI returns a point-by-point evaluation, not a generic "well done."
The operating syntax for an interview:
You are an expert recruiter. Conduct a simulated job interview with me
for the position of [role] at [company/sector]. Here is my resume and the posting:
[paste text].
Rules:
- Ask one question at a time and wait for my answer before the next.
- Start with an opening question, then alternate behavioral and
technical questions, gradually raising the difficulty.
- React to my answers with realistic follow-up questions,
as a real interviewer would.
- Don't give me feedback during the interview: hold it for the end.
- After 8 questions, stop and give me a structured evaluation: a score
from 0 to 100, three strengths, three weaknesses with examples from my
answers, and one concrete piece of advice for each weakness.
Begin now with the first question.
For a presentation, the syntax changes its objective:
You are the audience of a presentation: a mix of a skeptical executive, a
technical colleague, and an inexpert client. I'm about to present [topic] to you.
I'll paste the outline to you: [text].
Rules:
- Listen to my delivery (I'll read it or say it out loud).
- Interrupt me with the objections and tough questions that this audience
would really ask, one at a time.
- Test me on timing, clarity, and weak points in the reasoning.
- At the end, evaluate: clarity of the message, handling of objections,
filler words I used, and what you would cut to stay within the time.
I'm ready, ask me the first context question and then let me start.
A detail that comes in handy later: when you've finished, close the session and the full transcript of the interview — questions, your answers, and feedback — stays saved in the chat history, to review whenever you want.
A concrete example
Marta has to do an interview for a project manager role at a software house. The evening before, she opens ChatGPT on her phone, pastes the first prompt with her resume and the posting into the chat, then turns on voice and says "let's begin."
The AI starts with "Tell me about a project you managed under pressure." Marta answers, and here the mechanism that matters kicks in: instead of moving on, the AI presses. When she describes the project, she's asked to explain how she handled the supplier's delay. Marta realizes she gave a vague answer and rephrases it with numbers: "three weeks of delay recovered by reassigning two resources."
After eight questions the AI stops and gives her the score: 72/100. Weakness number one: she tends to recount the context for too long before getting to the result. Saying the answers out loud without being able to edit them forces her to organize her thoughts efficiently. The next day, at the real interview, she structures each answer starting from the result. It comes naturally to her: she had already "gotten it wrong" the evening before.
When it does NOT work (and how to fix it)
If the AI invents details about your sector or the company
In voice mode the AI can confidently state false things, so verify the important information. Fix: use the role-play to practice the form of the answers (clarity, structure, handling of objections), not as a source of technical facts. Verify those facts separately, on the company's official sources.
If it interrupts you constantly while you speak
It's the most frustrating problem with voice. The interruption handling of advanced voice mode is aggressive: it treats any sound — a cough, a pause, background noise — as a possible interruption. Fix: use headphones with a good microphone, reduce background noise, and turn on "push to talk" mode if you're in a noisy environment. Alternatively, speak in shorter, more decisive sentences.
If the free trial runs out halfway through the interview
The free limits on advanced voice are tight. You get a warning when 3 minutes of audio remain, and the conversation closes when the limit is reached. Fix: when advanced voice ends, you're not stranded. At the free daily limit you're automatically brought back to standard voice mode until the next reset. Standard is less fluid but it's perfectly fine for questions and answers. For logged-in Free users, voice is powered by GPT-4o mini with a limit of 2 hours per day.
If it "forgets" the role and goes back to being the polite assistant
Voice mode tends to lose the thread of the character, because it doesn't keep the context of previous voice sessions. Fix: when it slips, correct it by voice — "remember: you're the recruiter, don't interrupt the interview to explain things to me." For long sessions, reload the role prompt at the start of each new session.
A tip from someone who really uses it
The value isn't in the "perfect" interview, it's in asking the AI to be unpleasant. Don't practice only on smooth scenarios: ask the AI to interrupt you, ask unexpected questions, or challenge your assumptions, because real situations are chaotic and the thing that matters is practicing recovery, that is, how you react when you lose the thread or have to clarify a point. Add a line to the prompt like "every now and then ask me a surprise question or a blunt objection": that's where you really train.
And don't ask it to replace a human eye when the stakes are high. The AI is excellent for the volume of repetitions, but it doesn't catch the nuanced feedback of an expert: a coach notices communication habits, body language, and delivery problems that AI tools don't evaluate. Use the AI for daily repetitions and a person for the decisive feedback.
Frequently asked questions
Can I do it for free or do I have to pay?
Free, yes. Standard voice mode is accessible to everyone, and even the advanced version has a daily preview. Advanced voice mode — the more natural and emotionally responsive one — is available in the free version with a short daily preview, and with much higher limits on Plus ($20/month) and Pro ($200/month). To practice for a few days before an interview, the free plan is enough.
In which language should I practice?
In the language of the real interview. Voice supports Italian, but sometimes it needs to be steered: if voice input doesn't accurately recognize what you say, correct it by voice, asking it to stay in the language you want. If you have to do the interview in English, do the simulation in English: it also serves to train pronunciation and quickness in the right language.
Does it also work to simulate difficult clients or negotiations, not just interviews?
Yes, it's the same mechanism. You can role-play scenarios with difficult clients and refine your communication skills thanks to the AI's feedback. You only change the role in the prompt: "you are an angry client asking for a refund," "you are a supplier who doesn't want to lower the price." The structure — role, rules, final feedback — stays identical.
The AI doesn't understand body language: so is voice role-play really useful?
Yes, but knowing what it gives you and what it doesn't. It's true that it doesn't see posture, gestures, and gaze, and for those you need to record yourself or have a human eye. But voice role-play trains what text can't: saying the answers out loud without being able to edit them forces you to organize your thoughts efficiently and to feel comfortable with unprepared answers. It's the gym for content and rhythm; body language you add in front of the mirror or the webcam.