The short answer
A small number of document constructions genuinely damage text extraction: multiple columns, floating text boxes, headers and footers, text saved as an image and unconventional section headings. Almost everything else on the usual scare list is fine, and a plain single-column document handles all of it.
What matters most
- The risk is not decoration. It is anything that changes the reading order of the text or removes the text layer entirely.
- Multiple columns, floating text boxes and content in headers or footers cause most real extraction damage.
- Standard fonts, bold text, horizontal rules, colour and a simple table are all normally fine.
- If you can select and copy the text out of your file in the correct order, a parser can read it too.
The one test that replaces the whole list
Open your finished file. Select all of the text, copy it, and paste it into a plain notepad application. Read what appears. If the words arrive in the order a human would read them, with your name at the top and each job attached to its own dates, your document is extractable. If your skills sidebar is interleaved through your job descriptions, or your phone number never appears, you have found a real problem in under a minute.
That test works because it is doing roughly what a parser does first. Everything below is an explanation of why particular constructions fail it, so you can predict problems instead of discovering them.
Reading order is the whole game
A PDF is not a page. It is a set of instructions for drawing marks at coordinates, plus, usually, a text layer recording which characters were drawn. Extraction walks that layer. Word files are structured differently but the principle survives: there is an underlying linear sequence, and your visual layout may or may not match it.
When the two disagree, a parser produces text that is individually correct and collectively nonsense. This is why two-column resumes are the single most cited problem. A design that shows skills on the left and experience on the right may store those as alternating fragments, and the extracted result reads like two documents shuffled together. Some extractors handle columns well; whether yours will is not a bet worth taking on an application you care about.
| Construction | Risk | What actually happens |
|---|---|---|
| Two or three columns | High | Text may be extracted row by row across the page, interleaving unrelated content |
| Floating text boxes and shapes | High | Content in a drawing layer is frequently skipped entirely |
| Contact details in the page header | High | Header content is often excluded from the extracted body, so the record has no email address |
| Text saved as an image, or outlined in a design tool | Fatal | There is no text layer at all; nothing is extracted |
| Nested tables used as page scaffolding | Medium | Cell order can be read in an unexpected sequence |
| A single simple table for a skills grid | Low | Usually extracts cleanly, though a plain list is still safer |
| Bold, italics, colour, horizontal rules | None | Formatting instructions do not affect the characters themselves |
| Standard bullet characters | None | Extract as ordinary list markers |
| Decorative dingbats and emoji as bullets | Low to medium | May extract as replacement characters that clutter the text |
Headings the parser recognises
Interpretation depends on landmarks. A parser looks for conventional section names to decide where experience stops and education begins. Creative headings do not break extraction, but they can break the mapping, which leaves your jobs unattached to any section.
My Journey So Far β Where I Made Things Happen β What I Know
Professional Experience β Education β Skills
Senior Analyst | Northline Group | 2019-2023
Senior Analyst
Northline Group β Chicago, IL
March 2019 β August 2023
Consistency matters as much as the words. If your first job puts the title above the employer and your third reverses them, you have taught the parser one pattern and then broken it. Pick an order and repeat it exactly for every entry, including the older ones you edited least recently.
Dates, the quiet troublemaker
Date handling causes more damage than any font choice ever has. A missing end date can make a role look current; a misread start date can make eighteen months of experience look like six. Write months in words or standard three-letter abbreviations, keep the range on one line beside its role, and use the same format from top to bottom.
- Good. March 2019 β August 2023. Unambiguous in every region and trivially parsed.
- Acceptable. Mar 2019 β Aug 2023. Shorter, still unambiguous.
- Risky. 03/2019 β 08/2023. Numeric formats invite the day-month versus month-day confusion, and some extractors normalise them wrongly.
- Avoid. Spring 2019 β Summer 2023, or a bare "2019-2023" for a role that lasted four months of that span.
- Avoid. Dates in a separate right-hand column aligned by tab stops into their own visual track. Keep them on the same line as the role they belong to.
File type, settled
Submit a text-based PDF unless the posting asks for something else, in which case do what the posting asks. Modern parsers read PDFs without difficulty; the advice to always send Word documents is a survival from an era of much weaker extraction. The genuine PDF failure case is an image masquerading as a document, which the select-and-copy test catches immediately.
Name the file like a professional document rather than a draft. Your name, the role, and nothing else β the file name appears in the recruiter's attachment list and in their downloads folder, and "resume_final_v7_USE_THIS.pdf" is a small avoidable signal.
What you are allowed to keep
The anxiety around this topic has quietly deleted a lot of legitimate design. You can use colour. You can use a second typeface for headings. You can use a horizontal rule between sections, generous white space, a subtle accent on your name, small caps for section labels, and a two-column skills list underneath a single-column layout as long as the reading order remains sane. None of that touches the characters in the text layer.
What you should skip is the photograph, the skill rating bars, the logo grid of technologies, the personal branding statement set inside a coloured shape, and anything that puts words into a drawing object. Those cost you either extractable content or, in the case of photographs on United States applications, introduce information into a hiring decision that has nothing to do with your ability and that many employers prefer not to receive. The EEOC summary of prohibited employment practices explains why employers are careful about that category of information.
A short pre-flight check
- Text can be selected and copied out of the exported file.
- Pasted into a plain text editor, the content reads in the right order.
- Name, email and phone number are in the body of the page, not the header.
- Section headings use conventional words a parser will recognise.
- Every role uses the same line order for title, employer, location and dates.
- Months are written in words or standard abbreviations, consistently.
- No text sits inside a text box, shape, chart or image.
- Bullets are ordinary bullet characters.
- The file name is your name and the role.
Every template in our template gallery that is marked ATS-safe is built to pass that list without you thinking about it, and the ATS checker will show you the extracted text from any file you already have. If the formatting is sound and you want a view on the writing instead, the length guide and the resume format guide deal with the structural decisions underneath the styling.
Questions people actually ask
Are two-column resumes always rejected?
They are not rejected, but they are the most common cause of scrambled extraction, because the visual order and the underlying text order can disagree. Some extractors handle columns correctly and some do not, and you cannot tell which one an employer is running. On an application that matters, a single-column layout removes the question entirely.
Do I have to use Arial or Times New Roman?
No. Any ordinary text font embeds and extracts normally, including most modern sans-serif faces. The fonts worth avoiding are heavily decorative display faces and anything where ligatures or unusual glyph mapping can produce odd characters on extraction. Legibility for the human reader is the stronger reason to keep the choice conservative.
Is a table on a resume a problem?
One simple table with a clear grid usually extracts fine. Trouble comes from using tables as invisible page scaffolding, where several nested tables control the entire layout, because the cell reading order can be unexpected. If a table is doing design work rather than presenting tabular data, replace it.
Should I remove all colour and formatting to be safe?
No. Colour, bold, italics and rules are drawing instructions and do not affect the extracted characters at all. A resume stripped to plain black text gains nothing mechanically and loses readability for the person who eventually opens it.
Does a photo on my resume break anything?
Technically it is just an image and does not damage extraction, unless text is embedded in it. The reason to leave it off a United States application is different: it introduces personal characteristics into a hiring decision, and many employers would rather not receive it at all. Conventions differ in some other countries.