How ATS parsing actually breaks a resume

Summary

An ATS does not look at your resume. It reads a text layer underneath it, in the order the file was written — not the order your eye follows. Almost every parsing failure comes from that gap.

What the parser actually receives

A PDF stores text as positioned fragments; a DOCX stores it as paragraphs inside a document tree. When a tool extracts text, it walks that structure and emits a string. Your two columns, your sidebar, your icon row — none of that survives as layout. Only the order in which the fragments were written survives.

So the failure mode is never "the ATS could not read my file". It is "the ATS read the file in a different order than I intended", or "the ATS never saw some of it at all".

Tables

A table is a grid of cells. Text extracted from a table arrives cell by cell, and the reading order depends on how the generator wrote it. A two-column table used as a layout device — dates in one cell, job title in the other — often comes out as all the dates, then all the titles, or interleaved.

Under the surface the same thing happens in Word: a table that visually looks like a header block is still a table, and the text still comes out by cell.

Columns and sidebars

A two-column resume has two blocks of text on the same baseline. Any extractor has to decide whether to read across the page or down one column first. Different tools decide differently, and none of them is wrong — which is exactly why you cannot rely on it.

The practical case: a Canva or Word sidebar template puts your name in the sidebar and your summary in the main column. If the extractor reads across the page, your name and the first line of your summary can end up on the same line. If it reads down the sidebar first, your skills list lands above your summary.

Headers, footers, icons and text boxes

• —Contact details placed in the Word header or footer are frequently dropped before the body is read. Keep a plain contact line as ordinary body text on page one.
• —An icon carries no text. A phone icon next to a number is an image; the number may or may not be extractable depending on how the file was generated.
• —Text boxes and floating shapes are positioned objects, not paragraphs. Their text can be written to the file after everything else, which puts it at the end of the extracted string.
• —Skill bars, photos, and decorative rules do not add text, and they can push real content later in the stream.

The three fixes that cover almost everything

• —One column. No sidebar, no table used as a layout grid. Headings as real paragraphs.
• —A plain-text contact line at the top of page one: name, email, phone, city. No icons, no text box.
• —Standard headings — Experience, Education, Skills — as plain paragraphs, so a parser can find the boundaries of each block.

How to check your own file

Open the PDF, select everything with the keyboard, copy, and paste into a plain text editor. What you get is close to what a parser gets. If the order looks scrambled, or a line reads "Operations Manager 2019 – 2021 Chicago, IL" with the city stapled to the title, a parser will see the same mess.

For a DOCX, the same test works, but also check the header and footer separately: if your email only exists there, assume it is lost.