How the CV check works: what it inspects and what it does not

All nineteen graded items

As of 6 September 2026, counted in the code.

Two modes, and the check decides which one runs

Given an advert, the check compares your CV against that specific text: the skills it asks for, its vocabulary, the job title it names. Given none, it compares against a fixed Swiss benchmark written into the tool — two to three pages, a professional photo as the local convention, permit and eligibility stated near the top, whether you are already in Switzerland, a three to four line profile with a visible differentiator, genuine motivation for this role, a signal about the environment you fit (small generalist team versus large specialised organisation), quantified results rather than duty lists, specific soft skills, every language with its level, and a section on interests.

Which mode runs is decided by the text that arrives, not by the button you pressed. Pick a posting from our own index for which no full text is held, and the check falls back to the Swiss benchmark and says so in the report, rather than scoring a fit against an empty field and printing a number for it. For a minority of postings we hold only a short public excerpt; those run in Swiss mode for the same reason.

A pasted advert must be at least forty characters and a CV at least eighty, below which the run stops with a message. Very long inputs are truncated, though only well beyond the length of any normal CV; if you paste a whole application dossier with references attached, put the CV first anyway.

What you upload, and what actually gets graded

PDF, DOCX and the old Word .doc format are accepted, or you paste the text directly. What a file actually is gets decided by its leading bytes rather than by its extension or the type the browser claims. That is why a file downloaded out of a cloud drive works here when it is refused as an unsupported type elsewhere.

Plain text is extracted from the file, and that text is what is graded. Fonts, colours, margins, columns, logos and embedded images are invisible to the grading. One exception is built in: if the first page of a PDF carries embedded raster images above a size threshold, a note is appended to the text saying a headshot is probably there. The photo is therefore inferred rather than seen — and the grader is explicitly told never to penalise a photo just because the text does not mention one.

If you pick an advert from our index instead of pasting it, the stored text is resolved on our side and used for scoring only. It never travels back to the browser; what stays visible is the short public excerpt. Edit the field yourself and your text wins — editing it is you saying to use that and nothing else.

The seven scored dimensions

The scored dimensions are ATS and keyword match, hard skills, soft skills, motivation, personality and environment fit, unique selling point, and the key outcomes you offer. Each gets an integer from 0 to 100 and at most two short sentences of reasoning. The headline percentage at the top of the report is the unweighted mean of those seven — no weighting, no bonus points, no post-processing.

Two rules make those numbers harsher than they would otherwise be. The first is the absence rule: where a dimension is simply not in the CV — no trace of motivation for this particular role, no quantified outcome anywhere, no stated differentiator — the grader must score it under thirty rather than give it something middling, state plainly that the CV contains none of it, and say which section the missing thing belongs in. The second is a separation rule: the keyword dimension may raise wording findings only. Letter-spaced headings, columns and date formats belong in the readability list and are not counted twice.

Against an advert, two tables come with it: matched and missing hard skills with how often each appears in the CV and in the posting, and the same for soft skills. Without an advert the missing column stays empty, because there is nothing for a skill to be missing from.

The fixed nine-item readability list

This pass is deliberately a fixed checklist rather than a free opinion: parseable contact details, standard section headings, single-column layout, job-title match, date-format consistency, file type and text layer, length, keyword coverage, and photo. Each gets have, medium or missing, one sentence on what was found, and a concrete fix wherever it is not a clean have.

The list is fixed so it cannot turn into a victory lap. The pass is instructed to grade honestly and not to hand back an all-green result. One example of how strict that is: headings set with spaced-out individual letters grade medium at best, even when the heading names themselves are perfectly conventional, because the spacing breaks the words apart when a machine reads them.

Length is measured against the Swiss norm, and in this tool that norm is two to three pages — not the single page of an American résumé. Someone arriving from Germany, France or the UK is therefore graded against a different ruler here than an international tool would use, and that difference is the point.

The three Swiss checks

A third pass reads the CV as a Swiss recruiter would and grades three things separately: work permit and eligibility, meaning whether a permit or EU/EFTA status is stated clearly; languages, meaning whether each one carries a level and how that sits against what the role needs; and the profile photo. A short Switzerland-specific note comes with it.

The photo therefore appears twice, on purpose. In the readability list it is about the tension with purely machine-read application routes; in the Swiss block it is about the local convention. Two different questions about the same object — read the two lines side by side rather than as a contradiction.

This pass also raises recruiter-facing findings that have no line of their own and files each against the dimension it belongs to: missing measurable results go to outcomes, a tone pitched below or above the advertised seniority goes to personality fit, an absent portfolio or web presence goes to the differentiator.

What the report actually contains

The percentage and a one-sentence verdict at the top, then an honest two to three sentence summary of the fit. Then three to five prioritised fixes, each tagged to a dimension and marked critical or warning — critical means this is the sort of thing that gets an application filtered out. If the pass returns fewer than three, they are derived from the worst individual findings so the list is never empty.

Below that: the seven dimensions with score, reasoning and the problems found; the three Swiss lines as their own block; two to four strengths; and the nine-item list. Confirmations are explicitly unwanted — a strong dimension returns an empty problem list rather than filling up with reassurance. The score and the reasoning carry the positive signal.

Every fix names the section to change, using the section names the CV builder itself uses: Profile, What Drives Me, Competences, Key Achievements, the skills lists, Languages, Availability. That is not decoration; it is what makes a report directly actionable in the editor.

What the check does not inspect

It does not see your document. Design, typeface, colour, margins and image quality are not judged, because none of that reaches the grading at all. You will not get a verdict on how your CV looks here — and any tool that gives one without rendering the page is guessing.

It does not look inside the employer's software. No vendor is queried and no filter rule is read. The readability grade is a grade against our fixed list, not a prediction about Workday, Prospective or anything else. Nobody can promise an applicant system will be passed, and we do not.

It does not rank you. There is no comparison with other applicants, no percentile and no odds of an interview, because we do not hold anyone else's dossier. It does not verify what you claim either: dates, employers and qualifications are not checked. It is not a proofreader — spelling, grammar and style are not reviewed, and a typo lowers no score. It does not rewrite the CV; rewriting is a separate step in the editor. It does not judge salary, decode Arbeitszeugnis (Swiss reference) phrasing, or file anything with the RAV, where job-search efforts only count when they are recorded officially.

And it is not deterministic. A language model does the grading, not a rulebook, so two runs on the same CV and the same advert can land a few points apart. If you want to measure an edit, read the individual findings rather than the third digit of the headline number.

Where the account is asked for, and where it is not

The first check runs with no signup, one per browser, and the full report is visible without an account: the overall score, all seven dimensions, both checklists, the fixes. The account is asked for on a second run, when you keep the report, and on the button that hands the findings to the CV builder.

On a free account, a standalone check does not currently count against the document budget of five a day and ten per rolling week. What is metered is the score button inside the CV editor itself, which draws on the AI-edit budget — also five a day and ten per rolling week. After that comes a seven-day trial with full access, then CHF 15 a month or CHF 95 a year.

A report produced without an account is held for thirty days and then expires. Sign up within that window and it is claimed into your account and the temporary copy deleted. A report inside an account can be deleted there at any time.

The language of the report, and what your CV is measured against

The report follows your interface language, not the language of the CV. Upload a German dossier and read the assessment in English, or the other way round. The CV is graded as written; nothing is translated.

That is a market fact rather than a feature demo. In our own Swiss job index, roughly two thirds of the non-apprenticeship adverts name at least one language requirement — the remaining third name none at all. Among those that do name one, 64.9 per cent ask for German, 46.6 per cent for English, 22.3 per cent for French and 3.6 per cent for Italian, and naming more than one is common rather than exceptional.

For the check that means two things. Where the advert states a language requirement, your entry is measured against it. Where it states none — about a third of the time — the line falls back to the Swiss norm: every language named, each with a level from A1 to C2 or native. A bare 'fluent' with no level is exactly the case this line flags.