Nanny Agency JobsUse AI as a practice questioner, then answer from your own experience. A useful rehearsal gives you room to explain, notice gaps and try again. It should not supply a more impressive employment history or train you to repeat a story that never happened.
Set the session up so the model asks questions rather than writing your performance. This guide concerns rehearsal before an interview, not assistance during an assessment or live conversation. Follow the agency's instructions about any task you have been asked to complete independently.
Decide what this rehearsal should improve
Choose one skill for the session: answering a follow-up directly, explaining your contribution, or stating a schedule boundary clearly. “Make me excellent at interviews” is too broad to review. A small goal lets you judge whether the practice helped.
Prepare a few factual notes outside the tool. Identify an experience you can discuss, what you personally did and what remains uncertain. Avoid uploading your full employment file or confidential family material. The AI privacy guide helps you make a minimal practice brief.
Use the agency interview preparation guide for the appointment, documents and questions you need. The AI session should work on how you explain your experience, not replace those preparations.
Decide in advance when to stop. One opening question and two follow-ups may be enough to reveal the part you need to clarify. More rounds are useful only if they address an identified difficulty.
Ask for questions before feedback
Here is an original NAJ practice instruction: “Help me rehearse an agency interview. Ask one question at a time about the experience notes I provide. Wait for my answer. Ask a follow-up where my responsibility is unclear. Do not invent my answer or supply achievements. After three questions, identify unclear wording and unsupported claims in my responses. Do not predict a hiring outcome.”
OpenAI recommends giving clear context and refining a request after reviewing the response. That supports treating the instruction as something you may need to adjust, rather than a guaranteed script. OpenAI: Prompt engineering best practices for ChatGPT
If the model immediately provides an ideal answer, redirect it: “Please ask the question and wait. I want to produce my own answer.” If it continues to write for you, end that attempt and practice from your own notes.
Answer with facts, including limits
Speak your answer before typing a summary if that helps you rehearse conversation. Keep the account grounded in a real task. Explain the setting generally, your action and what you know about the result. Do not add a successful ending merely because an interview story seems to require one.
Use an honest boundary when needed: “I assisted with that part,” “I would need to check the date,” or “I have not held that responsibility independently.” A practice session is a place to become comfortable with accurate limits.
If a question concerns a situation you have not encountered, distinguish a hypothetical response from past experience. You can discuss what you would clarify without claiming to have handled the event before.
Do not use generated childcare instructions as a training substitute. If a rehearsal exposes a knowledge gap, identify the topic and use appropriate current training or authoritative guidance to address it separately. The interview practice itself does not establish competence.
Audit the feedback instead of accepting a score
OpenAI notes that ChatGPT can produce incorrect information even when it sounds confident. Treat feedback as a suggestion to inspect, not an assessment of your employability. OpenAI: Does ChatGPT tell the truth?
Useful feedback points to something visible in your answer: you did not explain who planned the activity, you skipped the question about availability, or you used a phrase whose meaning was unclear. You can verify those observations by rereading your response.
Less useful feedback predicts that recruiters will love an answer, assigns a hiring probability or tells you to claim a stronger result. Do not convert those statements into evidence about an agency's decision.
Make your own two-column note: “Change I will try” and “Why it stays accurate.” If the suggested improvement has no factual basis, reject it. A clearer answer should be easier to defend, not merely more persuasive.
Test the method with a fictional rehearsal
Original fictional example: Taylor says, “I organized the whole school-holiday program.” A follow-up asks which parts Taylor planned personally. Taylor realizes that a parent selected the activities while Taylor prepared materials and followed the schedule.
Taylor revises the account: “The parent chose the holiday activities. I prepared materials, checked the day's schedule and told the parent when a booking detail needed clarification.” The revision is less sweeping but more precise.
The model then suggests adding that the program improved the children's confidence. Taylor has no basis for that claim and removes it. The useful result is a clearer description of responsibility, not a bigger achievement.
For the next attempt, Taylor looks away from the text and answers the question again using the same facts in different words. This checks whether the account is understood rather than memorized. The fictional experience belongs only to this exercise.
Transfer the practice into an unaided conversation
Close the chat and explain the same example aloud. Keep only a few factual prompts beside you. If you can answer only by repeating the generated wording, return to the underlying facts and shorten the answer.
Use this original completion checklist: I answered the actual question; I can identify my contribution; I did not invent an outcome; I can state a limitation; and I can explain the example without the tool open.
Record one useful improvement and one remaining preparation task. You might need to verify a date, clarify the interview format or review a relevant course record. Those are concrete next steps. A model's encouraging score is not a reason to stop checking the information you will present.
Sources
OpenAI: Prompt engineering best practices for ChatGPT and Does ChatGPT tell the truth?, checked October 27, 2026. The session controls, prompt, checklist and fictional rehearsal are original NAJ editorial tools.