{"course":{"slug":"ai-careers","title":"Working in AI","url":"https://addaly.com/learn/ai-careers","level":"beginner","track":"work"},"lesson":{"slug":"practising-out-loud","title":"Rehearsing with a model that wants to please you","url":"https://addaly.com/learn/ai-careers/practising-out-loud","position":48,"of":85,"minutes":8,"answer":"Interviews are lost in production rather than recognition, so rehearsal must be out loud, one question at a time, with the model instructed to press on vagueness rather than encourage.","keyIdea":"Interviews are lost in production rather than recognition, so rehearsal must be out loud, one question at a time, with the model instructed to press on vagueness rather than encourage."},"body":{"format":"markdown","text":"## Recognition is not production\n\nYou read an answer to \"tell me about a time you disagreed with a colleague\" and it makes sense. That feeling is recognition, and it is not the skill being tested. The skill is producing a clear answer, out loud, in ninety seconds, while nervous.\n\nThe gap between the two is where almost all interview failure lives, and it closes with production practice only. There is no version of this that happens in your head.\n\n## A rehearsal that is worth the hour\n\nGive the model the advert and your CV, then:\n\n> \"Conduct a 25-minute first-round screen for this role. Ask one question at a time and wait for my answer. Do not give feedback or hints during the interview. Ask at least two follow-up questions when an answer is vague. At the end, tell me the two weakest answers and exactly why.\"\n\nThree details matter and all three are usually skipped.\n\n**One question at a time.** A model given a free hand produces a list of twelve questions, which you then read. That is recognition again.\n\n**No feedback during.** Encouragement after every answer trains you for an interview that does not exist.\n\n**Follow-ups on vagueness.** The follow-up is where real interviews are lost (\"you said you improved the process, by how much?\"), and it is the part you cannot rehearse alone.\n\nAnswer out loud, not in the chat box. Typing is a different skill and you will not be typing.\n\n## Making it stop flattering you\n\nModels are agreeable by default, which is the single thing that makes them poor practice partners. Instruct against it explicitly:\n\n> \"Now play a sceptical hiring manager who thinks I am too junior for this role. Give me the three objections you would raise, in the words you would use.\"\n\n> \"For each of my answers, name the follow-up question I would least like to be asked.\"\n\n> \"Which of my claims would you not believe without evidence, and what evidence would satisfy you?\"\n\nThe last one is the most useful prompt in this lesson. It surfaces the claims on your CV that are load-bearing and unevidenced, which is exactly what a good interviewer probes.\n\n## The free thing that beats every tool\n\n**Record yourself on your phone and watch it back.** It is unpleasant, it takes fifteen minutes, and it is more useful than any coaching product sold to job seekers.\n\nYou will discover, in order: that you take forty seconds to reach the point; that you say \"basically\" eleven times; that you trail off at the end of answers instead of stopping; and that a story you thought was two minutes is five. All four are fixable within a week, and none of them would have been visible from the inside.\n\nA friend or a study partner is better still, because a person reacts and a recording does not. Peer practice costs nothing and both sides improve, since watching somebody else answer badly teaches you more about structure than answering well does.\n\n## Timing, concretely\n\n- **Behavioural answer: 90 seconds.** Two minutes is the ceiling.\n- **\"Tell me about yourself\": 60 to 90 seconds.** Present, past, why here. Nothing before university unless asked.\n- **Technical explanation: 2 minutes**, then stop and ask whether they want more depth.\n\nTime three answers with an actual timer once. Your internal estimate is wrong by roughly a factor of two, in the direction you would expect.\n\n## Two honest limitations\n\n**A model does not know the company's real bar.** It invents plausible questions, not their questions. It cannot tell you whether your answer would pass at that firm, and if it says it can, it is guessing.\n\n**Over-rehearsal is visible.** An answer delivered word-for-word sounds recited, and interviewers discount it. Rehearse the *structure* and the *facts* (the situation, the number, the decision, the outcome) and let the sentences be new each time. Aim to have said each story out loud four or five times, not twenty."},"check":{"question":"Why does asking a model for \"a list of likely interview questions for this role\" do so little for a candidate's performance?","options":[{"text":"The questions it generates are usually inaccurate for the specific company.","correct":false,"tempting":"True and worth knowing, but a candidate who practised badly on accurate questions would still fail for the same underlying reason."},{"text":"It rehearses recognition, reading answers that seem sensible, while the interview tests production of a structured spoken answer under pressure, including the follow-up when an answer is vague.","correct":true},{"text":"Reading questions in advance makes answers sound rehearsed and interviewers discount them.","correct":false,"tempting":"Identifies a real risk of over-rehearsal, which comes from repeating fixed wording rather than from having seen the questions."},{"text":"Interview questions are confidential, so any published list is out of date.","correct":false,"tempting":"Implies the problem is the freshness of the list, when the format of the practice is what fails regardless of which questions are used."}]},"prev":{"slug":"the-ai-screened-interview","title":"Recorded and machine-scored interviews","url":"https://addaly.com/learn/ai-careers/the-ai-screened-interview"},"next":{"slug":"building-with-an-assistant","title":"Building a project you can still defend","url":"https://addaly.com/learn/ai-careers/building-with-an-assistant"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Rehearsing with a model that wants to please you', in Working in AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-careers/practising-out-loud (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Rehearsing with a model that wants to please you","courseTitle":"Working in AI","module":{"index":5,"slug":"searching-with-ai","title":"Using AI in your own search"},"answer":null,"keyIdea":"Interviews are lost in production rather than recognition, so rehearsal must be out loud, one question at a time, with the model instructed to press on vagueness rather than encourage.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"recognition-is-not-production","runs":[{"text":"Recognition is not production"}]},{"t":"p","runs":[{"text":"You read an answer to “tell me about a time you disagreed with a colleague” and it makes sense. That feeling is recognition, and it is not the skill being tested. The skill is producing a clear answer, out loud, in ninety seconds, while nervous."}]},{"t":"p","runs":[{"text":"The gap between the two is where almost all interview failure lives, and it closes with production practice only. There is no version of this that happens in your head."}]},{"t":"h2","id":"a-rehearsal-that-is-worth-the-hour","runs":[{"text":"A rehearsal that is worth the hour"}]},{"t":"p","runs":[{"text":"Give the model the advert and your CV, then:"}]},{"t":"quote","runs":[{"text":"“Conduct a 25-minute first-round screen for this role. Ask one question at a time and wait for my answer. Do not give feedback or hints during the interview. Ask at least two follow-up questions when an answer is vague. At the end, tell me the two weakest answers and exactly why.”"}]},{"t":"p","runs":[{"text":"Three details matter and all three are usually skipped."}]},{"t":"p","runs":[{"text":"One question at a time.","b":true},{"text":" A model given a free hand produces a list of twelve questions, which you then read. That is recognition again."}]},{"t":"p","runs":[{"text":"No feedback during.","b":true},{"text":" Encouragement after every answer trains you for an interview that does not exist."}]},{"t":"p","runs":[{"text":"Follow-ups on vagueness.","b":true},{"text":" The follow-up is where real interviews are lost (“you said you improved the process, by how much?”), and it is the part you cannot rehearse alone."}]},{"t":"p","runs":[{"text":"Answer out loud, not in the chat box. Typing is a different skill and you will not be typing."}]},{"t":"h2","id":"making-it-stop-flattering-you","runs":[{"text":"Making it stop flattering you"}]},{"t":"p","runs":[{"text":"Models are agreeable by default, which is the single thing that makes them poor practice partners. Instruct against it explicitly:"}]},{"t":"quote","runs":[{"text":"“Now play a sceptical hiring manager who thinks I am too junior for this role. Give me the three objections you would raise, in the words you would use.”"}]},{"t":"quote","runs":[{"text":"“For each of my answers, name the follow-up question I would least like to be asked.”"}]},{"t":"quote","runs":[{"text":"“Which of my claims would you not believe without evidence, and what evidence would satisfy you?”"}]},{"t":"p","runs":[{"text":"The last one is the most useful prompt in this lesson. It surfaces the claims on your CV that are load-bearing and unevidenced, which is exactly what a good interviewer probes."}]},{"t":"h2","id":"the-free-thing-that-beats-every-tool","runs":[{"text":"The free thing that beats every tool"}]},{"t":"p","runs":[{"text":"Record yourself on your phone and watch it back.","b":true},{"text":" It is unpleasant, it takes fifteen minutes, and it is more useful than any coaching product sold to job seekers."}]},{"t":"p","runs":[{"text":"You will discover, in order: that you take forty seconds to reach the point; that you say “basically” eleven times; that you trail off at the end of answers instead of stopping; and that a story you thought was two minutes is five. All four are fixable within a week, and none of them would have been visible from the inside."}]},{"t":"p","runs":[{"text":"A friend or a study partner is better still, because a person reacts and a recording does not. Peer practice costs nothing and both sides improve, since watching somebody else answer badly teaches you more about structure than answering well does."}]},{"t":"h2","id":"timing-concretely","runs":[{"text":"Timing, concretely"}]},{"t":"ul","items":[[{"text":"Behavioural answer: 90 seconds.","b":true},{"text":" Two minutes is the ceiling."}],[{"text":"”Tell me about yourself”: 60 to 90 seconds.","b":true},{"text":" Present, past, why here. Nothing before university unless asked."}],[{"text":"Technical explanation: 2 minutes","b":true},{"text":", then stop and ask whether they want more depth."}]]},{"t":"p","runs":[{"text":"Time three answers with an actual timer once. Your internal estimate is wrong by roughly a factor of two, in the direction you would expect."}]},{"t":"h2","id":"two-honest-limitations","runs":[{"text":"Two honest limitations"}]},{"t":"p","runs":[{"text":"A model does not know the company’s real bar.","b":true},{"text":" It invents plausible questions, not their questions. It cannot tell you whether your answer would pass at that firm, and if it says it can, it is guessing."}]},{"t":"p","runs":[{"text":"Over-rehearsal is visible.","b":true},{"text":" An answer delivered word-for-word sounds recited, and interviewers discount it. Rehearse the "},{"text":"structure","i":true},{"text":" and the "},{"text":"facts","i":true},{"text":" (the situation, the number, the decision, the outcome) and let the sentences be new each time. Aim to have said each story out loud four or five times, not twenty."}]}],"check":{"question":[{"text":"Why does asking a model for “a list of likely interview questions for this role” do so little for a candidate’s performance?"}],"options":[{"text":[{"text":"The questions it generates are usually inaccurate for the specific company."}],"correct":false,"tempting":[{"text":"True and worth knowing, but a candidate who practised badly on accurate questions would still fail for the same underlying reason."}]},{"text":[{"text":"It rehearses recognition, reading answers that seem sensible, while the interview tests production of a structured spoken answer under pressure, including the follow-up when an answer is vague."}],"correct":true,"tempting":null},{"text":[{"text":"Reading questions in advance makes answers sound rehearsed and interviewers discount them."}],"correct":false,"tempting":[{"text":"Identifies a real risk of over-rehearsal, which comes from repeating fixed wording rather than from having seen the questions."}]},{"text":[{"text":"Interview questions are confidential, so any published list is out of date."}],"correct":false,"tempting":[{"text":"Implies the problem is the freshness of the list, when the format of the practice is what fails regardless of which questions are used."}]}],"order":[2,1,3,0],"why":null},"prev":{"slug":"the-ai-screened-interview","title":"Recorded and machine-scored interviews","minutes":9},"next":{"slug":"building-with-an-assistant","title":"Building a project you can still defend","minutes":9}}}