Canva, Figma, and many online templates handle custom fonts by converting them into vector outlines instead of embedding font data. Visually identical. But underneath, the text is now a collection of shapes. This is one of the biggest reasons why dual-submission strategies are now required for technical roles.
When an ATS encounters these shapes, it runs OCR to convert them back into text. The result:
What you wrote: "5 years of experience with React and TypeScript"
What the ATS reads: "5years ofexperience wxth Reac7 and TypeScripl"
Test this yourself: open your PDF, select all text, copy it, and paste into Notepad. If it is garbled, that is exactly what the ATS sees.
Parsing engines read from left to right. When two columns exist the engine frequently mashes the dates on the left directly into the job titles on the right destroying chronological logic.
Does invisible keyword text still work in resumes?
No. Modern algorithmic screeners actively detect and aggressively penalize hidden white text tactics explicitly flagging the applicant for manipulative fraud.
Should I use tables to format my skills section?
Absolutely not. HTML and PDF tables commonly scramble semantic extraction outputs. Use strictly linear standard unordered bullet lists to guarantee flawless data extraction.
Recruiters rely on automated filters. If the ATS reads your "2023" as "2O23" (using the letter O instead of the number zero), you might be filtered out of a search for candidates with recent experience. Subtle glitches in OCR create massive gaps in your data integrity. Web profiles provide the raw text, ensuring 100% accuracy for every tool that reads them. This accuracy is important for your visual hierarchy to remain effective.
Messy Background Layers
Designed resumes use background colors and sidebars as separate layers. The parser does not understand layers. It reads characters in coordinate order regardless of which visual layer they belong to. This is another reason static PDFs are increasingly unreliable.
A sidebar heading "Experience" next to a job title "Senior Software Engineer" can become:
"ExSenior Software Engineerperience"
Characters merged based on vertical position, not visual grouping.
Semantic Tags for the Win
Web profiles use semantic HTML tags. This tells the reader (and the machine) exactly what is what. An h1 tag is always a title. A li tag is always a list item. This eliminates the "coordinate guessing game" that PDF parsers have to play. It is the difference between reading a recipe and trying to guess one from a picture of a meal.
Common Questions
Do big tech companies use OCR on resumes?
Almost all of them do. They handle thousands of applications per day, and manual data entry is impossible. If their machine cannot read your file, you are basically invisible.
Is an exported Word document better than a Canva PDF?
Usually, yes, because Word tends to preserve text layers better. However, it still lacks the screen-responsiveness and brand-authority of a custom web profile.
How can I check if my current PDF is machine-readable?
Try to copy a paragraph and paste it into a plain text editor. If the words are joined together or letters are replaced with symbols, it is failing the machine test.
Even careful exporters get tripped by print oriented settings. Exporting for print embeds fonts as vectors. Exporting with compression flattens layers. Password protection blocks parsers entirely. Some ATS vendors strip metadata and garble unicode accents in names. A candidate named José becomes Jose or a random symbol pair.
If you must ship a PDF, export from Google Docs or Word with standard fonts, single column, no text boxes, no headers and footers duplicated in body. Run the copy paste test on every new version before you upload.
Scanned PDFs Are the Worst Case
Printing your resume and scanning it back creates a pure image file. OCR quality drops sharply. Handwritten notes in margins become noise. Staple holes become black blobs. Yet candidates still do this when a portal asks for PDF and they only have a paper copy from a career fair. Digital native hiring expects digital native documents.
The recruiter side of the pipeline
After parsing, recruiters search inside the ATS for strings like "Kubernetes" or "staff engineer." If OCR mangled those strings, you fail search even when a human would understand the PDF visually. Machine readability is gate one. Human readability is gate two.
Column Layouts Destroy Chronology
Two column PDFs are the default design on template sites. They look sharp on a monitor. Parsers read left to right across the full page width. Your left column dates mash into right column job titles. The stored record might show your 2019 internship dates next to your 2024 senior role. Automated filters then think you have fifteen years of experience or duplicate entries.
Single column layouts feel boring. They parse correctly. That tradeoff matters more than aesthetics when your application competes with four hundred others in the same ATS queue.
Icons and Graphics Become Noise
Phone icons, envelope icons, and skill stars look clean in Figma. To a parser they are empty squares or random unicode. Some systems strip them. Others insert placeholder characters that break keyword matching. Plain text contact lines survive every pipeline: email, phone, city, URL.
The copy paste audit
Before you upload any PDF, copy all text and paste into Notepad. If line breaks look wrong or words merge, assume the ATS sees the same mess. Fix the source file or switch to a web profile.
Hidden Text Triggers Fraud Flags
Old advice told candidates to stuff white keywords in margins. Modern parsers detect color mismatches and font size anomalies. Fraud scores rise. Recruiters get automatic warnings. You are worse off than if you had sent an honest plain document.
Put keywords in visible bullets where they belong. If you used Kubernetes, say what you did with it. Authentic density beats invisible tricks.
Why Web Profiles Win the Parser Game
HTML gives every element a type. Headings, lists, links, dates in structured fields. No coordinate guessing. No OCR. When a recruiter shares your CVin.Bio link internally, the data stays intact. When they forward a PDF, each copy might parse differently depending on the reader software.
Upload your CV once to generate the structured profile. Keep a minimal PDF for forms that demand uploads. Lead humans to the URL every time. Read how to test ATS parsing on any file you still send.
Vendor Differences Matter Less Than You Think
Workday, Greenhouse, Lever, and Taleo all parse differently, but they share the same failure modes: columns, graphics, and OCR. A plain single column document survives most of them. A designed Canva export fails most of them. Do not tune for one vendor. Tune for plain structure.
When a portal lets you paste a URL, paste it. The human reviewer often has more authority than the parser score. Give them something readable while the parser chews on your plain upload.
Building a Parser Safe PDF Backup
Keep a boring PDF in your downloads folder for portals that demand uploads. Same text as your web profile, zero design flourishes. Update it whenever you update your live page so both stay aligned. Name the file with your name and role, not resume_final_v7.pdf.
Run the copy paste test after every edit. If the boring PDF fails, fix text in the source profile and export again. The web profile remains your primary artifact. The PDF is a compatibility shim.
When Humans Never See Your PDF
Some pipelines auto reject below a parser confidence score. You never get a human. Garbled PDFs die there. If you suspect auto rejection, simplify the file radically and reapply where allowed, or route outreach to humans with your web link through email and referrals.
Parser failure is silent. Response silence after perfect qualifications often traces back to formatting, not fit.
PDF metadata recruiters rarely see but parsers use
Some ATS tools read PDF title and author fields. Set title to Your Name - Software Engineer Resume and author to your email. Empty metadata is not fatal. Correct metadata is one more signal that you treat the application seriously. Export from the same source as your web profile so title, dates, and bullets stay aligned.
Your hiring document strategy for 2026 should be simple. Web profile first, plain PDF second, designed PDF never. Recruiters who care about design will visit your link. Bots that care about structure will parse your plain backup. Nobody needs a gradient sidebar that breaks both channels.
Spend design energy on project pages and case studies linked from your profile. Let the resume layer stay boring and legible. Boring wins parsers. Case studies win humans.
When someone asks for your resume, send the link. When a portal demands a file, send the plain export. When someone compliments your Canva layout, redirect them to the project that actually proves you can build.
Plain structure is a feature, not a failure of creativity. Recruiters reward clarity over decoration every time they open a candidate file.
Anna M.