Skip to content
GUUL
← Back to the edit

Parsing / 8 MIN READ + INTERACTIVE EXAMPLES

We Tested Our Own Résumé Designs. Here Is What Broke.

What our own checks found when they read Guul’s résumé designs: sidebars that lost words when sections moved, text readers that disagreed, and the limits of what a simulation can tell you.

Skip to the interactive example ↓
Two printed resume sheets on an architect light table: a single-column page with straight emerald alignment rules beside a two-column sidebar page under a magnifying glass showing orange scan lines crossing the column gutter. AI editorial illustration.
Fictional sample page on an AI editorial illustration
THE TAKEAWAY, BEFORE THE SCROLL

Single-column designs came through every check we ran. Sidebar designs needed repairs, and open-source text readers still disagree about their reading order. None of this is a test of a real employer’s system.

Note: every result on this page comes from simulations and open-source text-extraction tools we run ourselves. None of it is a test of a live Workday, Greenhouse, Lever, iCIMS, Taleo or SAP SuccessFactors system. None of it tells you what an employer decided. Employers set these systems up in different ways.

We sell résumé designs. Some of them have a sidebar, a photo or skill graphics. Those are the features people most often warn you about. So we pointed our own checks at our own designs and wrote down what failed.

Some of what broke was ours. We say so below.

What we ran, and what we did not

We publish two downloadable sets of evidence, and each answers a narrower question than “will this résumé get me an interview?”

  1. A release check on all 43 designs. Before a design ships, an automated check prints one fictional résumé in it as a PDF, reads the text back with one PDF reader (pdf-parse 1.1.1), and compares the extracted words with what was typed. Where a design allows moving sections between a sidebar and the main column, it tests those moves too. The certificates are re-issued every time a design or the code that draws it changes, always on US Letter paper with the same sample résumé.
  2. A published extraction study. 40 fictional résumés — 10 profiles exported in both a single-column layout and a sidebar layout, as PDF and Word .docx — read across 80 runs by four open-source text-extraction libraries. Every input file, extracted .txt file, results.json summary and reproduction script is public on our parse tests page.

What we did not do: run each of our 43 designs through live vendor accounts, or publish a 43-design by 6-system matrix. We have not done that run, and we will not guess at its results here.

Our designs, by layout

Of our 43 live designs, 22 are single-column, 18 have a sidebar and 3 have two columns. Those counts come directly from each design’s declared layout contract.

All 43 pass our release check today. That is less impressive than it sounds: a design that fails the release check cannot ship, so 43 of 43 passing is a gate, not a research finding.

The useful detail is how each layout class passes:

Layout class Live Guul designs Declared layout score What the release check certifies
Single-column 22 designs 90 / 100 Every section and entry reads back in top-to-bottom order
Two-column 3 designs 85 / 100 Words come back intact; side regions certified as unordered lists
Sidebar 18 designs 80 / 100 Words come back intact; sidebar certified as an unordered list

In the 22 single-column designs, the check confirms the text reads back in sequence. In the 21 designs with a sidebar or a second column, the check certifies the side column as a list: the words come back, but different PDF readers can return the sidebar before, after or interleaved with the main column.

What broke in our own designs

The first time we ran our arrangement release check, on 26 July 2026, it tested 14 designs. Only 5 passed. 9 failed.

All 9 failures were designs with a sidebar or two columns. In 8 of them, moving the skills list from the sidebar into the main column changed what the PDF reader got back:

  • In three designs, the extracted skills text shrank from 160 characters to 108.
  • In three others, it shrank from 160 characters to 92, and jobs or schools also disappeared from what the reader found.
  • In one design, two of five schools went missing and the skills text shrank from 160 characters to 154.
  • In one design, one of seven jobs stopped being found.

In the ninth design, the check itself refused to write its record, so it gave no answer and no certificate at all.

All 4 single-column designs in that July run passed on the first try. So did one sidebar design, Onyx.

We repaired those bugs over the following rounds and re-ran the check. In one later round, we switched off section moving on three sidebar designs — Onyx, Ember and Summit — rather than offer a move we could not prove was safe. Today’s certificates prove section moves on all three again, after later rounds fixed what blocked them. Inside the builder, the download menu on every design also offers PDF — ATS Optimized, which prints your résumé in a single column with standard headings.

Where the text readers disagreed

Our published layout study read the same 40 fictional files across four open-source libraries. Here is how many expected fields each library recovered:

Simulation chart comparing PyMuPDF, Mammoth, Poppler pdftotext, and PDF.js across 40 fictional resumes, labeled as a simulation rather than a live employer ATS test.

Reader (simulation of text extraction, not an ATS) Fields recovered Section order kept
PyMuPDF (PDF) 220 of 220 20 of 20 files
Mammoth (Word .docx) 220 of 220 20 of 20 files
Poppler pdftotext (PDF) 210 of 220 10 of 20 files
PDF.js (PDF) 178 of 220 20 of 20 files

Poppler (run with default reading-order settings) kept section order in all 10 single-column PDFs and in 0 of the 10 sidebar PDFs. All 10 fields it lost were in the sidebar files; 8 of them were degree names split across the summary text. For example, “Certificate in Business Administration” came back in two pieces with a sentence from the summary in between.

PDF.js dropped the punctuation in every phone number and the dash in every date range, in all 20 study PDFs — single-column and sidebar alike. +1 (202) 555-0101 came back as four separate groups of digits. It also returned a different apostrophe character in one name, Nora O’Neill, so that name did not match in either layout. The column layout did not cause these losses; the reader did.

So the honest answer to “do sidebars break parsing?” is: it depends on which reader touches your file. Two of the four libraries recovered every field from our sidebar files, in order. One scrambled every sidebar’s section order.

Hiring-system vendors also differ in what their own help pages warn about:

  • Greenhouse explicitly lists multi-column layouts, tables, headers and footers, text boxes, graphics and scanned images in its Unsuccessful resume parse support guide.
  • Lever notes in Understanding resume parsing that autofill extracts name, contact details, location and current company, advises applicants to check the filled fields, and mentions scanned files as a problem.
  • SAP SuccessFactors notes in Knowledge Base Article 2081576 that résumé parsing cannot read scanned résumés saved as images.
  • Oracle Taleo documents a default 1 MB candidate attachment limit in its Taleo Enterprise documentation, and we found no layout warnings in what it publishes.
DON’T JUST READ IT. PLAY WITH IT.

Same story. A little more breathing room.

Switch the layout. Watch how spacing and hierarchy change the way this sample reads.

Same words. Different presentation.
ILLUSTRATIVE RESUME EXCERPT

Alex Morgan

Product designer

Profile

Product designer focused on making complex workflows easier to understand.

Experience

Product Designer · Example Studio

  • Mapped the onboarding journey with the support team.
  • Designed and tested a clearer account setup flow.
  • Documented reusable components for the engineering team.

Skills

Interaction design · Prototyping · Design systems

Aa

Everything is asking for attention.

Closely packed headings and text leave fewer visual signposts. Try the other treatment.

Layout demonstration. Fictional sample, not a hiring-outcome test.Find a layout with room to breathe ↗

Photos and skill graphics

We publish 49 template guide pages. 25 of them carry an explicit written layout note: 21 mention a photo, 21 mention graphics such as skill bars, 18 mention a sidebar and 3 mention two columns.

We have not run an isolated experiment that toggles only a photo or a skill bar on the same résumé to measure whether extracted text changes. Today those warnings come from what vendor help pages document, not from an isolated measurement of our own. Measuring that is still on our list.

What this means if you are choosing a design

  • A single-column design came through every check we ran. When you do not know how an employer’s portal reads PDFs, a single-column layout — or exporting with PDF — ATS Optimized — avoids column-reading ambiguity altogether.
  • A sidebar design can work, provided you inspect the exported file. Use the file inspector at the end of this page, or open your exported PDF, select all text, paste it into a plain note, and check that your contact details, job titles, employers and dates stay paired.
  • A passing check is not an interview guarantee. Our checks use fictional English-language résumés, US Letter paper and named open-source tools. Your wording, your employer’s portal settings and human review decide what happens next.
NOW PICTURE YOUR NAME HERE.

Your experience. A fresh canvas.

Try a different look. These are actual Guul template previews, with illustrative sample content.

GUUL / Crisp

A clean slate. Literally.

One column, distinct sections, and room for your experience to lead. A simple place to start.

1 column · Linear streamDeclared layout score: 90/100Arrangement gate verified ✓

0 multi-column reading notes across all 6 modelled hiring systems.

See full template parse table →
Make this layout yours ↗The builder opens with your selected template.
Replace all sample details with your own experience.
Explore the resume lookbook →

Crisp / a closer look

Single-column layout

Guul template · illustrative sample content

Full Crisp resume layout with illustrative sample content
A clean slate. Literally.One column, distinct sections, and room for your experience to lead. A simple place to start.
Make this layout yours ↗

How we did this, and the limits

  • The arrangement release check uses one fictional résumé (ats-certification-senior-lists), US Letter paper and pdf-parse 1.1.1. It does not cover A4 paper. It does not cover Word files either, except for the few designs that also carry a separate Word certificate.
  • The published extraction study uses 10 fictional English-language profiles across 2 layouts and 4 open-source libraries (PyMuPDF, Mammoth, Poppler and PDF.js). Download all 40 files and raw outputs on /resume-template-parse-tests.
  • Our resume checker models six hiring-system families from public vendor documentation; it never connects to a live vendor or employer account.
  • We did not test other companies’ templates, so we make no claims about them.

Frequently asked questions

Did you test these resume designs inside live Workday or Greenhouse employer accounts?

No. Both checks described here run on our own infrastructure using open-source text-extraction libraries (pdf-parse, PyMuPDF, Mammoth, Poppler and PDF.js) paired with published vendor help documentation. Employers also set these systems up in different ways.

Why did 9 multi-column designs fail your first internal check in July 2026?

In 8 of them, moving the Skills list from a narrow sidebar into the main column made the PDF’s text lose part of the skills or whole entries, because of bugs in the code that draws those designs. In the ninth, the check refused to write its record. We fixed those bugs. Today the check runs before every release, and a design that fails it cannot ship.

How can I check whether my own PDF scrambles columns before I apply?

Drop your exported PDF into the browser-only inspector on this page or on Resume PDF Text Check. Your file stays in your browser tab. It shows the text in the order one PDF reader (PDF.js) extracts it, and flags section headings that land on the same line.

IN-ARTICLE BROWSER CHECK · PRIVATE & LOCAL

Inspect what a text extractor actually reads

Drop a .pdf or .txt resume, paste your draft, or compare two sample streams. Files stay in your browser memory and are never uploaded or saved.

Inspecting: Fictional 1-column PDF stream1 pg · 99 words · 662 chars
01 · TEXT LAYERSelectable text stream present

Read 99 words directly from the character stream.

02 · BODY CONTACT FIELDSEmail ✓ · Phone ✓

Both email and phone digits appear in the main extracted text.

03 · STANDARD HEADINGSSummary, Experience, Education, Skills

Conventional section titles appear on their own lines.

04 · COLUMN ROW ORDERNo collided section headings

No two uppercase section labels were merged onto the same horizontal line.

View raw extracted text order (99 words)
PRIYA NAIR
Senior Backend Engineer | San Francisco, CA | priya.nair@example.com | +1 (415) 555-0148

PROFESSIONAL SUMMARY
Backend engineer with 7 years building payment and ledger services in Go, Python, and PostgreSQL.

WORK EXPERIENCE
Senior Backend Engineer — Acme Cloud
May 2021 – November 2024 | San Francisco, CA
• Redesigned checkout idempotency keys in PostgreSQL, reducing duplicate settlement retries by 64%.
• Cut p95 API latency from 480 ms to 185 ms across 1.4M daily transactions.

EDUCATION
B.S. in Computer Science — University of California, Davis
Graduated June 2017

TECHNICAL SKILLS
Go, Python, PostgreSQL, Apache Kafka, Kubernetes, Terraform
SOURCES

Sources linked in this guide

The outside pages this guide refers to, in the order they appear. Each opens in a new tab.

FROM “GOOD IDEA” TO “THAT’S MY RESUME.”

Your story.
Let’s give it a page.

Bring your experience into Guul. Choose a layout, make it yours, then inspect the document you export.

Start shaping your resume ↗Already have one? Check its extracted text →

ONE GOOD READ DESERVES ANOTHER