Do tables and columns break CV parsing?

Layout elements and their actual risk

Much of what circulates as an ATS rule comes from large Anglo-Saxon markets. Swiss recruiters often expect a designed document. Two columns are not a mistake — a CV that says nothing once the layout is stripped is.

How this was measured

The content was identical in every variant — the same person, the same four roles, the same CEFR language levels, the same nine keywords. Only the layout changed: one column; side column on the right; side column on the left; the classic Swiss tabellarischer Lebenslauf (the date-left, content-right table CV) built as a real table; a two-column flowing text set; free-floating absolutely positioned text boxes, the shape a design tool exports; skills in a three-column table; the same side column but placed before the main text in the file; the whole page as a single table row; the same page as a multi-row layout table; and a very narrow justified side column with automatic hyphenation, the way a word processor sets one.

Each variant was printed to A4 PDF through a browser's print function, then read back twice: once with the library our own CV check uses, and once with an independent second library from the Python world. Both are standard tools, not something we built for the occasion. The part that makes the result solid: for all eleven variants, both libraries reported the same section order. So where something goes wrong, it goes wrong in the file, not in one quirky reader.

First result: nothing goes missing

For every variant we checked whether each individual item reappeared: name, address, the work-permit line, the profile, every role, every employer, every bullet, every language level, every keyword. Ten of the eleven came back complete, and the eleventh was complete apart from one hyphenated word. No layout swallowed a section, skipped a table or dropped a column.

That disposes of the most common fear. A CV with a side column does not vanish inside a system. When an application goes unanswered, the reason is almost certainly something else — the profile, the timing, the permit question, the language the advert is written in — and not that software could not read the table.

Second result: order flips, and it follows the file, not the eye

A PDF has no sections. It has characters at coordinates, in the order the producing program wrote them. Extraction libraries hand back that order, not the order an eye scans the page in. So everything depends on the sequence of blocks inside the file — not on where they appear on the page.

The test shows this precisely. The right-hand side column and the left-hand side column both came out in perfect reading order, because in both the side column follows the main text in the file. The same page, visually indistinguishable, but with the side column written first, produced education, languages, skills and interests before the profile and the entire work history. Two pages that look identical to a person therefore produce two completely different text files.

Furthest from reading order were the free-floating text boxes: languages, skills and interests all came before the profile and before every job. If you use a template from a design tool, that is the shape you are exporting.

The genuinely awkward case: the multi-row layout table

The worst variant was not the one with the most columns. It was the one where a table is used as scaffolding and runs over several rows: a slice of side column on the left, a slice of work history on the right, row after row. Extraction walks the table row by row, and the result cuts the side column straight into the career history.

Concretely, in our test the languages block landed between the first and second job, skills between the second and third, and interests between the third and fourth. To software segmenting the text into sections, the second job no longer belongs to work experience at all — it belongs to languages. This is the only configuration in the whole test where the career history stops being contiguous, and the finished PDF gives no hint of it.

Real tables, by contrast, are unremarkable

The classic table CV — date in the left cell, content in the right, row by row — came out in perfect order, with all five date ranges intact on one line each and nothing lost. The same held for the variant that puts only the skills into a three-column table: flawless.

That is the exact opposite of the usual rule. The risk is not "table", it is "table used as page scaffolding". A table that actually holds data, whose rows correspond to the rows of the page, is read out in the order it is written — and that order is the right one. For anyone using the Swiss table-style CV, this is an all-clear.

What narrow columns do to long words

The variant with a very narrow, justified side column and automatic hyphenation was the only one in which a word actually broke. "Mehrwertsteuerabrechnung" (VAT accounting) came back as "Mehrwertsteuerabrech-" plus "nung" on the next line, and a qualification title split the same way. The inserted character is not the ordinary keyboard hyphen but a separate hyphenation character, so a plain search for the whole word no longer finds it.

This hits German harder than English, and no English-language guide mentions it — which matters here, because German-language adverts are the largest slice of the Swiss market and the terms that count in a Swiss dossier are long compounds: Quellensteuerabrechnung, Debitorenbuchhaltung, Sachbearbeiterin Rechnungswesen. Long compounds are exactly what a narrow column breaks. The rule that follows is small and specific: turn hyphenation off in the side column, or make the column wide enough for your technical terms.

The side columns had a second, quieter effect too. In the narrow column, "KV Zürich Business School · 2019 - 2020" did not fit on one line, so the years slipped onto the next one by themselves. Four of five date ranges survived as a unit instead of five of five. Anyone extracting employment periods from that CV loses a little precision.

The trap nobody writes about: letter-spaced headings

The single largest effect in the whole test came from neither a table nor a column, but from a pure design decision: letter-spaced capitals in the section headings — the effect that makes "PROFESSIONAL EXPERIENCE" look airy.

We set the same single-column CV with increasing letter spacing and extracted each time. Up to a tenth of the font size, all six headings survived intact. From roughly twelve hundredths upward it flipped abruptly: not one of the six headings was findable as a word any more, and all six came back as separated letters with spaces between them. A page that looks like any other loses every one of its section markers — and for software that segments a dossier by its headings, that is the worst possible outcome.

The threshold sits between two values that are two clicks apart in a layout tool. This is not a reason to panic, but it is a very good reason to open your own PDF once, select the text, paste it into a plain text editor and check whether it says "EDUCATION" or "E D U C A T I O N".

Which adverts this actually matters for

Before rebuilding a layout, look at where the apply button goes. Across the live Swiss adverts we track, about half lead to the employer's own site, a good third into an applicant system, and the rest to a platform or an agency. What is striking is how little that shifts with employer size: among employers with a single open advert, roughly a third of apply links lead into an applicant system; among those with 21 to 100 open adverts it is just under 38 per cent. A large employer is no guarantee that a machine reads first.

And even when an applicant system is involved, it does not follow that the system parses the CV. On 29 August 2026 we opened the application form of a Swiss vendor: salutation, first and last name, street, postcode, town, e-mail, phone and date of birth are all typed in by hand, and beside them sit separate upload slots for the cover letter, the CV, Arbeitszeugnisse (Swiss employer references), diplomas and certificates, and further attachments — plus a field for the application photo. None of it is pre-filled from the PDF. In that configuration the layout decides nothing about data capture; it only decides how quickly a person understands the page.

This is a Swiss pattern international guides do not know. The full dossier — references, diplomas, photo — is the norm here, and the two largest systems behind Swiss adverts are Workday and a Swiss vendor, Prospective, that no English-language ATS guide mentions. Together they account for a good two fifths of the adverts that route into an applicant system at all; across the whole set we counted a good two dozen different systems.

If you are applying from abroad, one practical consequence: the same employer may run an advert in German and another in English, and the English one is far more likely to route into an applicant system. Among the adverts we track, close to 64 per cent of English-language adverts lead into one, against roughly 30 per cent of German-language and under a quarter of French-language adverts. English-language adverts are also the group aimed most often at people arriving from outside Switzerland — so the reader most likely to be parsed by software is also the one least likely to know the Swiss dossier conventions.

What follows from this, and what does not

No tool and no layout can guarantee an application "gets through"; nobody outside can see what happens inside a specific filter. What can be described is the mechanism: a CV whose sections come out in the order they were meant in, and whose technical terms survive as whole words, is easier for software to organise than one where neither holds. That is the whole claim, and it is enough.

In practice: side column yes, but placed after the main text in the file and wide enough for long words. Table yes, as long as it holds data rather than scaffolding two parallel narratives. Hyphenation off in narrow columns. Letter spacing in headings kept modest. And before you send, copy the text out of your own PDF once — the test takes half a minute and replaces every guess about your layout with an observation.