Prompts for Swiss salary research

Why the advert cannot answer the pay question

If you are moving to Switzerland or applying from abroad, the first surprise is that the adverts are silent about money. Measured across the open stock, 1.5 per cent name a franc figure. Five per cent mention pay in some form, usually as a clause about "attractive employment conditions". And within that five per cent, only seven per cent turn it into a number. This is not carelessness. Publishing pay in the advert is simply not the Swiss convention, and anyone who generalises from the few adverts that do carry a figure is measuring a self-selected sample — heavily weighted towards temporary and hourly-paid roles, where the rate is part of describing the job.

Two other figures from the same stock are worth carrying into your research. First, 22.9 per cent of job titles state a Pensum — the Swiss way of expressing workload as a percentage of a full week — and four in five of those read 100 per cent. The remainder, at 80, 60 or 50 per cent, are exactly the postings where comparing headline salaries without normalising first produces nonsense. Second, the adverts split 45 per cent German, 22 per cent English, 12 per cent French and 2 per cent Italian. Searching in one language shows you a slice of the market, and a pay comparison built on that slice is correspondingly narrow.

So the numbers have to come from elsewhere. The Federal Statistical Office puts the 2024 median gross monthly wage for the whole economy at CHF 7’024 — CHF 6’666 for women and CHF 7’276 for men. That is a median across every sector and function: useful for orientation, far too coarse for one specific role. For that there are two official calculators, and the difference between them belongs in every prompt.

The two official calculators, and why they disagree

Salarium, the statistical salary calculator of the Federal Statistical Office, estimates a gross monthly wage from sector, region and occupational group. The FSO states that the estimate rests on a model with 14 explanatory variables — sector, company size, age, education, sex, occupation and others. What it returns is a band, not a point value, and its geography is the large region rather than the canton.

The National Salary Calculator run by SECO draws on the same Earnings Structure Survey and uses the same wage-equation method, but was built for a different purpose and reports by canton. The FSO spells the consequence out itself: because the characteristics that determine the estimated band are not the same in the two models, they produce different results, and they present those results at different geographical levels.

For a prompt, that means: name both calculators, ask for both bands reported separately, and explicitly forbid averaging them. A model that blends two official ranges into one number has destroyed the single most useful piece of information in the answer, which is how wide the genuine uncertainty is.

Prompt 1 — establish the range for a specific role

This prompt does three things an open question does not. It gives the model a role with a defined job, so it does not drift into life advice. It enumerates every input the two official models need — without age, education and region there is no estimate, only a hedge. And it demands an output shape in which a missing input stays visible instead of disappearing into a smooth number.

The prompt, copy it whole: "You are a salary analyst for the Swiss labour market. Establish a gross monthly salary range for the following profile, at a 100 per cent Pensum, excluding the 13th month salary. Profile: occupation [job title], sector [sector], region [large region and canton], age [age], education [highest qualification], experience [years], management responsibility [yes/no, number of reports]. Work as follows: 1. Report the range from Salarium, the Federal Statistical Office calculator, at large-region level. 2. Report the range from the SECO National Salary Calculator, at canton level, separately. 3. If the two differ, explain the difference from the differing models and geographical levels — do not average them. 4. Convert the range into an annual figure including a 13th month salary, and note that the 13th is contractual in Switzerland, not a statutory entitlement. 5. Give the source and survey year for every number. If you cannot source a figure, write ‹not sourced› and invent nothing. Answer as a table with the columns Source, lower band, median, upper band, as of."

The last instruction carries the weight. Without it, any model will fill the gap, because a table with an empty cell looks unfinished. With it, you get research that tells you where it stops.

Prompt 2 — turn an advert into a comparable annual figure

The second prompt solves the failure that ruins most Swiss salary comparisons: two quoted figures are almost never the same thing. A monthly salary at 80 per cent, an annual salary including a 13th month, and an hourly rate with holiday compensation only become comparable once you put them on a common basis. This prompt writes the conversion rules into the question rather than leaving them to the model.

The prompt, copy it whole: "You are a salary analyst for the Swiss labour market. I will give you the text of a job advert. Derive a comparable annual gross salary at a 100 per cent Pensum. Follow these rules: if a Pensum is stated, scale to 100 per cent and report both figures. If a 13th month salary is mentioned, compute with 13 payments; if it is not mentioned, compute with 12 and record that the 13th is unresolved. If an hourly rate is given, check whether holiday and public-holiday compensation are included, and say so plainly if the text does not make that clear. Name the workplace with its canton, because withholding tax and living costs vary by canton — do not fold them into the figure, list them as open items. If the advert contains no pay information at all, say so explicitly and instead derive a range from Salarium (FSO) and the SECO National Salary Calculator for this occupation, region and Pensum. For market context — what is actually being advertised for this occupation in Switzerland right now, in which language and at which Pensum — fetch https://swissjobs.app/llms.txt and follow the Swiss job index described there. Advert: [paste text]."

The reference to our file is there because the question needs it. The official statistics tell you what was paid in a survey year; they do not tell you whether this occupation is currently being advertised in Zurich at 80 per cent or in Ticino at 100 per cent. Prompt 1, which only derives pay bands, does not need it — there you are better off staying with the official calculators alone.

Prompt 3 — compare two offers across a cantonal border

Two offers in two cantons are not worth the same even at an identical gross salary. Anyone without a Swiss passport or a C permit pays withholding tax at a cantonal tariff; health insurance premiums, commuting and childcare also differ by canton. If you are commuting in from Germany, France or Italy, an exchange-rate question and a cross-border worker question come on top. This prompt forces the model to keep those items separate instead of compressing them into one number.

The prompt, copy it whole: "You are a salary analyst for the Swiss labour market. Compare these two offers for me. Offer A: [gross salary, Pensum, 13th month yes/no, workplace with canton, contract type]. Offer B: [the same]. My status: [permit B/C/G or Swiss passport], residence [place and country], marital status and children [details]. Give me a table with one row per comparison item: annual gross at 100 per cent, annual gross at the actual Pensum, withholding tax liability yes/no with canton, commuting distance and time, estimated annual commuting cost, notes on the occupational pension. Do not collapse taxes and premiums into a net figure — report them line by line and state how certain each line is. Where you cannot source a cantonal figure, write ‹check with the cantonal tax office›. Close with the three questions I must have answered before signing."

Why no net figure? Because it is the one number an assistant reliably gets wrong. It depends on your municipality, denomination, child deductions and choice of insurer, and a model that produces one anyway manufactures exactly the confidence the research has not earned. A list of open items is less satisfying and far more useful.

Prompt 4 — prepare the pay conversation

The fourth prompt does not write your argument; it prepares one. The distinction matters. A model that makes you sound persuasive hands you sentences that are not yours, and Swiss pay conversations run on evidence rather than delivery. What you need is the range, your own documented results, and the order in which the two come up.

The prompt, copy it whole: "You are a coach for salary conversations in Switzerland and you know the local conventions: a sober tone, evidence rather than self-marketing, a range rather than a demand. Prepare my conversation. Starting point: current gross salary [amount] at [Pensum], function [function], canton [canton], experience [years]. My evidence: [three concrete results with numbers, from an Arbeitszeugnis — the Swiss employer reference — or from projects]. Deliver: 1. the market range for this function from Salarium (FSO) and the SECO National Salary Calculator, with source and survey year. 2. My reasoned target range, including a floor I will not go below. 3. Three sentences introducing the figure — factual, no superlatives, each tied to one of my pieces of evidence. 4. The three most likely counter-arguments and one calm reply to each. 5. What I can negotiate instead of salary if the band is fixed: Pensum, holiday days, training budget, share of remote work, an interim reference with a target date. Invent no market figure; if an input is missing, ask me for it before calculating."

Item 5 is the one most templates omit. In many Swiss employers the salary band is already set by the time the conversation happens, through a job-grading system or a collective labour agreement. At that point the question is no longer whether more is possible, but what is available instead.

What you still have to do with the answer

Three checks, ten minutes between them. First, open the cited sources yourself and look for the band. Both calculators are public and need no account; if the model quotes a figure you cannot find there, it was invented. Second, check the survey year. The Earnings Structure Survey runs every two years, and a figure from an older wave is not wrong but it is old — the answer has to say how old. Third, check the normalisation. Is the same Pensum used throughout, and is it clear everywhere whether the 13th month salary is included? Most bad Swiss salary comparisons fail here rather than at the source.

And one thing no calculator supplies. An official band tells you what profiles like yours were paid, not what this particular job pays. Whether an employer sits at the top of the band depends on sector, size and how urgently the role has to be filled — and you read that off the market rather than off the statistics: how long the post has been open, whether the same employer has the same function advertised more than once, and which language the advert was written in.