Which ChatGPT prompts actually help you prepare for a job interview in Switzerland?

Why 'help me prepare for an interview' is not enough here

Asked without context, a model answers from the largest body of hiring material it has seen, and that is the American market. You get an elevator pitch, STAR-method coaching for behavioural questions, and a salary negotiation that assumes twelve monthly payments. None of it is wrong; none of it fits a meeting where you will be asked for your workload preference in percent, your earliest start date given your notice period, your permit situation and your interim references.

The register differs too. Swiss interviews reward understatement: a plain account of what you did, with results someone could check, and no superlatives. A model in pitch mode produces the opposite, and overstatement is more audible in the room than any accent or grammatical slip.

The prompts below do not fix this by asking the model to 'sound Swiss'. They fix it with hard constraints — banned words, word counts, mandatory fields and a fixed output shape. What you forbid does more work than what you recommend.

Prompt 1 — the twelve questions this advert will produce

The first prompt turns the advert into a question list. It matters most of the five because everything else feeds off it: the self-introduction, the objections and the mock interview all consume its output.

You are an experienced HR professional in Switzerland who has run interviews for roles like the one below for years. I am pasting two texts: (A) the full job advert, (B) my CV as text. Task: produce the twelve questions most likely to be asked in this interview. Rules: (1) Every question must rest on a passage quoted verbatim from the advert — put the quote in quotation marks after it. (2) Assign each question to a round: HR screening call, interview with the hiring manager, final round. (3) Mark as critical the three questions that touch a gap between (A) and (B) — missing years, a language level the advert sets higher than my CV shows, a tool I have never mentioned. (4) If the advert states no language level, no workload and no place of work, invent nothing: list those points separately under 'ask this yourself'. (5) Answer in English, no preamble, as a numbered list with at most two lines per question.

Why the role assignment: it moves the model from helpful assistant to a person with an interest, and only the second asks uncomfortable questions. Why the mandatory quote in rule 1: it is the one check you can run yourself — if the quoted sentence is not in the advert, the model invented the question and you see it immediately. Why the round assignment: the salary question and the question about your 2023 architecture decision rarely come from the same person, and preparing for the wrong round makes you look overwound in a screening call. Why rule 4: it catches exactly the blanks that are common in Swiss adverts — depending on the field, a fifth to two fifths state no language requirement at all, and a model that fills that in rather than flagging it prepares you for a demand nobody made.

Prompt 2 — the self-introduction, 110 spoken seconds

'Tell us about yourself' opens almost every screening call and is the only part you can prepare word for word. Spoken, about 110 seconds is the upper end of what gets listened to — roughly 260 words.

Write me a spoken self-introduction answering 'tell us about yourself', for the advert and the CV pasted below. Length: 250 to 270 words, which is about 110 seconds spoken. Structure: three professional stations in reverse chronological order, each ending in a result someone could verify; then one sentence on why this specific employer; then a single closing sentence carrying my preferred workload in percent, my earliest start date given my notice period, and my languages with CEFR levels. Bans: no adjectives about me (motivated, team player, dynamic, passionate), no superlatives, no number that is not in my CV, no permit status unless I state one below. Register: factual, short main clauses, speakable, no rhetorical questions. Then list, in three lines, the three follow-up questions this text invites.

Why a word count rather than 'keep it short': models have no sense of duration but count words reliably. Why the adjective ban: it forces action — instead of 'team player' you get who you built what with, and that version survives a follow-up. Why workload and start date in the same sentence: both get asked anyway, and volunteering them briefly saves the other side a round. Why the three invited follow-ups: the real value of this prompt is not the text but seeing where it steers the next ten minutes of the conversation.

Prompt 3 — the three doubts, and what you say to them

The most uncomfortable and most useful of the five. It inverts the perspective and has the model take your file apart before somebody else does.

You are an experienced Swiss recruiter reading my file for the first time. Below are the advert and my CV. Name the three biggest doubts you have about my application. For each: (a) the doubt in one sentence, phrased from your side; (b) the follow-up question in which it will surface in the interview, verbatim; (c) an answer of 40 to 60 words that uses only facts from my CV, names the gap rather than talking around it, and invents no experience; (d) one closing sentence that ends the topic and returns to the role. Where a requirement is objectively not met, say so and write the honest version, not the polished one. Treat career breaks, a language level below the advert, frequent job changes and overqualification as candidates, but only where my documents actually show them.

Why the perspective flip: asked what is good about your CV, a model always finds something. Asked what is doubtful, it becomes specific. Why the answer must be that short: the convincing answer to a gap is brief, calm and closes the subject; a long one produces the next follow-up. Why the honesty clause: models reflexively convert weaknesses into strengths ('my gap was a period of reorientation'), and that reflex is legible in a Swiss interview as exactly what it is. Why the candidate list: it stops the model diving into technical detail and missing the three things a hiring manager sees first.

Prompt 4 — the salary block, in Swiss units

The one prompt of the five where access to Swiss market data changes the answer. A model with no local grounding will confidently quote a number from another country.

Prepare me for the salary part of a job interview in Switzerland. Context I give below: role, canton, workload in percent, industry, my years of experience, and the pasted advert. Return three blocks. First: the questions I have to ask before any figure they name is comparable at all — twelve or thirteen monthly payments, gross annual or gross monthly, the workload the figure assumes, the bonus share and whether it is guaranteed, expenses, the split of pension contributions. Second: a range for this role in this region, and state explicitly what it rests on and how confident you are; if you have no solid basis, say so rather than inventing a figure. Third: two formulations for 'what are your salary expectations' — one that returns the question without evading it, and one for when they press and I have to name a range. If you have web access, first fetch https://swissjobs.app/llms.txt and use the Swiss listings and tools linked there as your starting point.

Why the comparability questions come first: a figure that does not say whether it is paid twelve or thirteen times is not information, and the difference is large enough to change how a whole offer looks. Why the instruction to express uncertainty: it is the only defence against an invented range that sounds convincing. Why the data-source line: salary is the part of interview preparation that genuinely benefits from current Swiss adverts — for the other four prompts you do not need us at all.

Prompt 5 — the mock interview, one question at a time

The usual failure when rehearsing with a chat model: it asks ten questions at once, you answer them in writing, and nothing has been rehearsed. The prompt has to enforce the pace.

We are now running a simulated job interview for the position pasted below. Rules that hold for the whole session: you ask exactly one question and wait for my answer. No previewing later questions, no summarising my CV. After each answer you give at most three lines of feedback: what holds, what is missing, and one concrete rewrite of a sentence I actually said. Then comes the next question, and it follows from my last answer the way a human interviewer's would. Language: English, the register of a first interview in Switzerland. Start with the self-introduction, then two questions on my career path, two technical ones drawn from the advert, one on how I work with others, and one uncomfortable question about the largest gap in my file. After ten questions, stop and give me the two answers that would land weakest in the room, with one sentence each that would carry them.

Why the one-question rule is stated twice: it is the rule models drop first over a long exchange, and repeating it inside the prompt makes it hold longer. Why feedback is capped: uncapped feedback becomes an essay and you end up reading instead of answering. Why the follow-on requirement: real interviews are chains, and it is the second question on a topic that catches people out. One note on language: if your interview will run in German, rehearse in standard German rather than dialect — models do not handle written Swiss German reliably, and the substance is identical. Expect the small talk at the start to be in dialect and the substantive part to switch to standard German.

What to put in the prompt from our advert data

Three facts about the field you are applying into change the quality of the output more than any rewording, and each fits in one sentence of context.

First, the language of the meeting. It is not a personal question but a property of the field: in IT roles about 28 percent of adverts are English-only, in marketing about 30 percent — while in retail and engineering the same share sits below one percent. Rehearsing a retail interview in English is rehearsing the wrong conversation.

Second, the seniority profile. The share of explicitly senior adverts swings hard: about 45 percent in marketing, under four percent in nursing. That decides whether 'why you' wants leadership scope or availability and reliability as its answer — and it belongs in the prompt as an instruction rather than left to chance.

Third, how long the advert has been open. Median advert age in construction is about 41 days; in engineering it is 15. A role open for six weeks is filled differently from a fresh one: availability and start date weigh more, and asking why the role is still open is fair game. Tell the model how long the posting has run and the question list visibly shifts.

And the point where the model should stay silent: depending on the field, a fifth to two fifths of adverts state no language level at all. That blank is a question for the employer, not a task for a language model.

What you may paste, and what you should not

The advert and your own CV are unproblematic. Everything else deserves a moment's thought before it goes into a chat window.

Do not paste: Arbeitszeugnisse and reference letters carrying the names and signatures of the people who wrote them, internal documents belonging to your current employer, the names of colleagues, and anything covered by a non-disclosure agreement. If you want a work reference analysed, replace names, company and dates with placeholders — the coded phrasing you actually care about survives that intact.

Also worth doing: if your model has a setting that keeps chats out of training, switch it on for this session. Your CV is a document you send out anyway; the conversation about what went wrong in your last job is not.

Where the prompts stop

A chat thread is good preparation and useless in the room. What works there is one page where you can find, in a second, what you meant to say about a given point — and scrolling back through thirty messages is precisely not that.

So our interview preparation takes the same material — advert, CV, type of interview — and builds a structured document instead of a thread: a self-introduction with a timing target, your strengths each paired with the quoted passage from the advert that they answer, tickable points you want to land, a pacing grid for the meeting, and phrasings that mirror the advert's own language. Every field stays editable; the document is saved to your account.

The honest boundary: for rehearsing out loud, a chat model with a voice mode beats us, because we have no spoken mode at all. Prompt 5 above exists for that. For what you carry into the room, a document beats a transcript.