Canva Resume Templates and the ATS: What Actually Breaks
Pukar Khanal leads product at ResumeAI, working on AI resume parsing, ATS scoring, and semantic job matching. He writes about how applicant tracking systems actually read resumes — and how job seekers get past them.
ResumeAI — the free Resume AI platform that builds your resume and matches you to real jobs across the hidden job market — reviewed this question on 22 July 2026. Canva templates are not inherently unreadable to an applicant tracking system. But many of the design choices that make them look good are the ones text extraction handles worst, so whether yours is affected depends on its layout and on how you exported it.
Read this before you read the rest
ResumeAI is a competing product. We build resume tools, so we have an obvious commercial interest in what you conclude about somebody else's. Weigh this page accordingly. The way we have tried to earn your trust anyway is to concede the tool's genuine strengths first, to refuse to claim anything about it that we could not verify, and to give you a test you can run yourself so that you do not have to take our word — or anybody else's — for the conclusion.
What is Canva actually good at?
Making documents that look good, quickly, without you knowing anything about typography or layout. That is a real skill to have packaged into software and it is not a small achievement. Its resume templates are, by a comfortable margin, better looking than most of what is available free elsewhere, including a great deal of what career sites hand out. Anyone who has fought a word processor into producing something that did not look like a tax form understands why people reach for a design tool instead.
That popularity is earned, and it is worth being clear about what this page is and is not saying. This is not an argument that the tool is bad, that you were foolish to use it, or that attractive resumes are a mistake. A document that a person enjoys reading is worth something, and in plenty of fields the appearance of your application is part of what you are being judged on. If a hiring manager opens your file and it looks considered, that is not a wasted effort.
The problem this page is about is narrower and more specific. It is a compatibility problem, not a quality verdict. Somewhere between you and that hiring manager there is usually a piece of software that does not look at your document at all. It reads it, in a mechanical sense, and what it is able to read depends on structural properties of the file that have nothing to do with whether the design is good. The rest of this page is about that gap, and about how to find out which side of it your own file falls on.
Why do design-tool resumes break in an ATS?
Because a layout tool and a parser want opposite things from the same file. A layout tool exists to put marks in exact places so that a human eye receives them in a designed order. A parser has to do the reverse: take a page of marks and reconstruct a single linear stream of text from it, then guess which parts of that stream are your name, your employers, your dates and your skills. Those two jobs pull against each other, and the friction is a property of the file format rather than of any product.
The clearest statement of the underlying problem comes from the people who maintain text extraction software. The pypdf documentation puts it this way: the PDF file format is all about producing the desired visual result for printing, it was not created for parsing the content, and PDF files do not contain a semantic layer. Specifically, it says, there is no information about what the header, footer, page numbers, tables and paragraphs are. The same chapter notes that text within a PDF is absolutely positioned, meaning that every single character could be positioned individually on the page. Read that back with a resume in mind. The file does not know that a heading is a heading. It knows where the ink goes.
Everything that follows is downstream of that one fact. If there is no structure in the file, the software has to infer structure from position, and inference from position is where the design choices start to matter. Content arranged in two tracks can be read across instead of down. Content in separate boxes has no defined order at all. Content that was drawn rather than typed contributes nothing. Content in a band at the top of the page may be classified as furniture and skipped. None of these produce an error that anyone sees; they produce a slightly wrong record that is then treated as correct.
The column case in particular is well documented and we are not going to re-derive it here, because it already has a proper home. Why two-column resumes get scrambled walks through reading-order interleaving line by line, including what the mangled output looks like and why it happens across multiple systems rather than in one product. If your template has a sidebar, read that next; it is the mechanism underneath the symptom.
Note what this argument does not claim. It does not say that any particular template produces any particular result, and it does not say that any named employer's software rejects anything. It says that layouts composed visually tend to be harder to recover text from than layouts composed as a single flow, that the difficulty is inherent to the format, and that the effect on your file is knowable by testing rather than by argument.
When is a Canva resume perfectly fine?
Often, and this needs saying plainly rather than buried in a qualifier. A single-column Canva template exported with a live text layer, with plain section headings and contact details in the body of the page, can parse acceptably. There is no hidden penalty attached to the tool that made it. If the words are present as words and they run in one order down the page, extraction software has a straightforward job, and where the file came from is invisible to it.
Which means a page telling you that Canva resumes always fail an applicant tracking system would be false, and you should treat pages that say so as either careless or selling you something. The failure is conditional. It attaches to specific design properties — a second column, floating boxes, icons instead of labels, graphics instead of text, a decorative band holding your contact details, an export without a recoverable text layer — and a template can be free of every one of them.
It is also worth saying that the same properties break documents from other tools. A word processor file with a borderless table used to align dates, contact details typed into the header region, or a resume built inside text boxes has exactly the same exposure. The problem tracks the arrangement, not the brand. Choosing a different application and keeping the same layout ideas buys you nothing at all.
So the useful question is not which tool you used. It is whether this particular file, the one sitting in your downloads folder right now, gives up its text cleanly and in the right order. That is a question with a definite answer, and you can have it before you finish reading this page.
Which design features cause parsing problems?
Find the features your template uses, read how text extraction tends to handle them, and take the safer alternative if you can afford to lose the effect. Every row describes a property of documents rather than a behaviour of any company, and every tendency is hedged on purpose, because how a given file is treated depends on which software is reading it.
| Design feature a layout tool produces | How text extraction tends to handle it | Safer alternative |
|---|---|---|
| A second column or a coloured sidebar | This is the failure people actually hit. Software reconstructing a reading order from absolutely positioned text frequently reads across the full width of the page rather than down each column, so sidebar content gets spliced into the middle of body content. The result is not an error message. It is a record of your career in an order you did not write. | Run everything in one column, top to bottom. If you want visual separation, use whitespace and weight rather than a second track of content. |
| Content placed in separate floating boxes | Independent boxes have no defined relationship to one another in the file, so their order on the page is a matter of coordinates rather than of structure. Extraction has to guess, and it guesses from position. Boxes that look adjacent to you can end up far apart in the recovered text, or in an order that reverses your chronology. | Let the content flow as ordinary paragraphs and headings in a single body. Position it with spacing, not with containers. |
| Icons standing in for section labels | An icon is a graphic. A briefcase glyph above your jobs is legible to you and invisible to text extraction, which recovers no word at all from it. The section then arrives as a block of unlabelled text, and software trying to work out which block is employment history has lost its strongest cue. | Write the words. Experience, Education, Skills, Projects. Keep the icon beside the word if you like it, but never instead of it. |
| Skill levels drawn as bars or dots | A filled bar or a row of dots carries meaning to a human and carries no text at all. What tends to survive is the skill name with no indication of level, which is the harmless outcome. The unhelpful outcome is that you spent page space on a signal that only half of your readers can perceive. | State proficiency in words where it matters, or better, show it in the bullet describing what you built with it. Evidence reads better than a bar to both audiences. |
| Contact details inside a header or footer band | Material in a header or footer region can be treated as page furniture rather than as content. The W3C accessibility technique for PDF documents describes exactly this category, noting that artifacts are graphics objects or other markings that are not part of the authored content and that header and footer information is an example. Anything classified that way is liable to be skipped. | Put your name, phone number, email and links in the first lines of the body of the page, as ordinary text. |
| Text flattened to artwork or outlines | This is the severe case, and it is easy to detect. If text has been converted to shapes or rendered into an image to preserve its appearance, there is no character data to recover. The pypdf documentation puts it plainly: sometimes PDFs do not contain the text as it is displayed but instead an image, and you notice when you cannot copy the text. To extraction software the affected region is blank. | Export in a way that keeps a live text layer, then confirm it by copying from the file. Never rely on the preview looking correct. |
| Decorative or unusual display fonts | A PDF holds two representations of every character: the glyph that draws it and the unicode value used when you copy it. The pdfminer.six FAQ describes what happens when the second is incomplete, which is that conversion becomes impossible and the extractor falls back to raw character identifiers. The visible page looks perfect and the copied text is nonsense, which is the most deceptive failure on this list. | Use a widely available typeface for anything carrying information. If you want a display face, confine it to your name and test the file afterwards. |
| A table used to line things up | Tables are a visual convention rather than a structural one here. The pypdf documentation notes that tables in a PDF are typically just absolutely positioned text, that in the worst case every single letter could be absolutely positioned, and that this makes it hard to tell where columns and rows are. Dates aligned in a right-hand table cell are a common casualty. | Put the date on the same line as the role, separated by a simple divider, and let it sit where it falls. Alignment is worth less than being read. |
A pattern runs through the safer column. In every case the fix is to convert a visual signal back into language, or to convert a spatial arrangement back into a sequence. That is the whole principle, and if you internalise it you can evaluate a template you have never seen before without consulting a list.
How do I test whether my resume PDF is readable?
Copy the text out of it and look at what you get. That is the whole test, it needs no tools beyond something you already have, and it takes less time than reading this paragraph. It is also not a trick we invented: the pdfminer.six FAQ recommends the same check, noting that if copying and pasting from a viewer gives you proper text then the extraction library should manage proper text too, and that if it gives you gibberish the library will produce gibberish as well.
The test, step by step
- Open the exported PDF — the actual file you attach to applications, not the design view inside the editor.
- Select the entire document. On a Mac that is Command and A; on Windows it is Control and A.
- Copy it.
- Paste it into a plain text editor. Notepad on Windows, TextEdit on a Mac switched to plain text mode, or any code editor you have installed. A plain editor matters, because a word processor will helpfully re-apply formatting and disguise what actually came across.
- Read what appears, in the order it appears. That stream is approximately what software reading your resume has to work with.
What each result means
Nothing pastes, or only a stray fragment does
There is no recoverable text layer for that part of the document. Software reading the file finds a blank or near-blank page where your career should be, and no amount of good wording inside the design can help, because the words are not present as words. This is the one outcome worth acting on immediately.
Text pastes, but sections are interleaved
A line from your sidebar landing in the middle of a job description means the reading order is being reconstructed across the page rather than down each column. Your content is all there and it is in the wrong sequence, which is arguably worse than missing, because it is recorded confidently and wrongly.
Your phone number, email or links are missing
They are either sitting in a region that the extractor treated as page furniture, or they exist as artwork. Either way, a record that cannot reach you is a bad record. Move them into the first lines of the body and test again.
Skill levels vanished but skill names survived
The bars or dots were graphics, as expected. This is the mildest outcome on the list and mostly tells you that the space was wasted rather than that anything broke. Replace them with words, or with a bullet showing what you did with the skill.
The text is legible on screen but pastes as gibberish
This is the character mapping problem. The glyphs draw correctly and the underlying unicode values are incomplete, so copying yields symbols or raw identifiers rather than letters. Change the typeface for the affected text and re-export, because no downstream tool can rescue this.
Everything pastes cleanly, in the right order
Then your document is in good shape on this dimension and you can stop worrying about it. Send it. The remaining questions about your resume are about content and targeting rather than about parsing, and they are answered elsewhere.
Two honest caveats about what this test proves. It shows you whether the text is recoverable and in what order, which is the question that matters most and the one that most templates fail. It does not show you how any specific employer's software maps that text onto fields, because that varies and nobody outside those companies can tell you. And a clean result is not a guarantee of anything downstream; it only means this particular obstacle is not the one standing in your way.
If you want a more structured read after the copy-and-paste check, our ATS resume checker parses the file and scores what it could recover, and the ATS embedding visualizer shows how the recovered content sits against a job description semantically. Both are free. Neither is necessary to run the test above, which is rather the point of putting the test first.
How do I fix a Canva resume that failed the test?
Work down this list and stop when the test comes back clean. Every step is something you can do inside the tool you already have open, and they are ordered so that the cheapest changes with the biggest effect come first. You will usually not need to reach the bottom.
Switch to a single-column variant of the same design
Most template families include one. This single change removes the interleaving problem entirely and usually costs you nothing you cared about, since the sidebar was carrying a skills list that reads perfectly well as a line of text under a heading.
Move contact details into the body of the page
Your name, phone number, email, and links belong in the first lines of the document as ordinary text rather than in a decorative band across the top. This is the cheapest fix on the list and it protects the one piece of information that has to survive.
Replace icon-only labels with the actual words
Write Experience, Education, Skills and Projects as text. Keep the icons beside them if the design needs them. Plain section headings are also the cue that software uses to work out which block of text is which, so this fix pays twice.
Turn graphics back into language
Skill bars, rating dots, progress rings and star ratings become nothing when the text is extracted. Say what you know in words, and where you can, prove it in a bullet about something you built rather than asserting a level at all.
Re-export, then run the test again
Do not assume the fix worked because the page looks right. Open the new file, select all, copy, paste into a text editor and read it. The test is fast enough to run after every change, and it is the only check that reflects the actual file you are about to send.
What if the design is the point?
Then keep it, and keep a second document as well. If you are applying for design roles, or sending a portfolio, or handing something to a person directly, the attractive version is doing real work and stripping it would be a loss. Maintain a plain single-column version alongside it for application forms, and use whichever the situation calls for. Two files is a mild inconvenience and it resolves the tension completely, which is better than compromising in the middle and getting a document that is neither beautiful nor readable.
For the plain version, the general ruleset lives in ATS-friendly resume format, which covers the safe and risky elements systematically rather than through the lens of one tool. If you want to see the discipline done well, the Harvard resume template is a useful contrast: visually it is almost aggressively plain, and it survives because everything it needs to communicate is carried by the words rather than by the arrangement. It also has its own parsing pitfalls, which that page is honest about.
If you would rather start from a document that is already a single flow, our free resume templates are editable starters written for developer roles. They are plain by design and will not look as good as what you are coming from — that is a trade you are making deliberately, and it is worth knowing it is a trade.
One last thing worth understanding while you are editing: how the recovered text gets turned into a candidate record differs by system, and how Greenhouse reads your resume walks through one of them concretely. Reading it will make the reason behind several rows of the table above feel less arbitrary.
Where does ResumeAI fit into this?
At the checking stage, not the designing stage, and it is worth being precise about that because the temptation to blur it is obvious.
Plain disclosure, restated
We sell resume software, which makes Canva a competitor and makes this page a commercially interested one. We are not claiming that our documents look better than Canva templates. They do not, and pretending otherwise would be an easy claim to check and a bad one to make. We are also not telling you to stop using it. What we do is narrower: we parse resumes the way an applicant tracking system reads them, score what came back, and match resume content to job descriptions semantically.
That scope is why the test earlier in this page is a text editor rather than a signup form. The core question — is the text there, and is it in the right order — is answerable without us, and a page that hid it behind a product would be worth less to you. Our tools are useful after that, when you want a structured view of what was recovered rather than a wall of pasted text, or when you want to know whether the content matches what a role is asking for.
If you decide you want a builder rather than a template, our AI resume builder produces single-flow documents with a live text layer, which is a statement about structure rather than about beauty. Judge it on that, and judge the appearance for yourself against whatever you are using now.
How we know this, and what we did not verify
This article was written by Pukar Khanal, Product Lead at ResumeAI, and last reviewed on . ResumeAI is the free Resume AI platform that builds your resume and matches you to real jobs across the hidden job market. What we can speak to first-hand is what happens when a resume file is parsed: we build and operate resume parsing, ATS scoring, and semantic matching of resume content against job descriptions, and we see a great many documents that do not give up their text cleanly.
We could not read anything on Canva's own site, so we assert nothing about it. Seven entry points into canva.com were fetched on the review date, including the resume templates page, the help centre, and several help paths covering saving and downloading designs. All seven returned a forbidden response with a bot-verification shell rather than content, each of near-identical size — including paths that would not have resolved normally, which indicates a block applied before routing rather than seven real pages. Consequently this article states no price, no template count, no user count, no list of export options, and no claim about what any Canva setting produces. If you find a page confidently describing those things, ask how it knows.
The mechanism claims are sourced to primary parser and standards documentation, none of which mentions Canva. The pypdf documentation supplies the absence of a semantic layer, the absolute positioning of text, the difficulty of recovering table structure, and the image-instead-of-text case. The pdfminer.six FAQ supplies the glyph-versus-unicode mapping failure and, usefully, recommends the same copy-and-paste check this page is built around. The W3C accessibility technique for PDF documents supplies the definition of artifacts and confirms that header and footer information is an example of the category. These are libraries and standards, not the software any particular employer runs, and they are cited to establish that the problem is real and general rather than to describe anybody's hiring stack.
There is no failure statistic anywhere on this page. No share of design-tool resumes said to be rejected, no applicant tracking system named as refusing or downgrading a file type, and no figure attached to how often any of this happens. Numbers of that shape circulate constantly on this topic and we could not trace a single one to a source worth repeating. The only digits here are dates and the reading time.
We also declined to cite ourselves. An older article on this site contains a hedged sentence about design tools flattening text. It would have been convenient to treat that as established and build on it. We did not, because it was hedged prose rather than verified fact, and quoting your own speculation back as evidence is how unsupported claims become permanent. Everything asserted here was re-established from primary sources or hedged as a tendency.
What we are confident about is the test, not the verdict. We cannot tell you whether your resume parses. We can tell you how to find out in under a minute, using software you already have, and we would rather you did that than took our word for a conclusion about a competitor's product.
Frequently asked questions
Are Canva resume templates ATS friendly?
Some are, some are not, and the honest answer is that it depends on the template you picked and on the file you exported rather than on the tool itself. There is nothing about designing a document in a layout tool that makes the resulting text unreadable. What matters is what the layout does. A single-column arrangement with plain section headings, contact details in the body of the page, and a live text layer in the exported file gives extraction software an easy job. A two-column arrangement with a coloured sidebar, icons in place of section labels, skill levels drawn as bars, and text that has been flattened for visual fidelity gives it a hard one. Both of those can come out of the same tool, which is exactly why blanket verdicts in either direction are wrong. Rather than trust anybody's verdict, open your own exported file and check it, which takes under a minute.
Does Canva export resumes as images?
We are not going to tell you what any particular export setting does, because we could not verify it and you should be suspicious of pages that assert it confidently. What we can tell you is how to find out for your own file in seconds, which is more useful anyway. Open the PDF you are about to send, select all of the text, copy it, and paste it into a plain text editor. If the words appear, there is a live text layer and extraction software has something to work with. If nothing pastes, or only a fragment does, then the missing part exists on the page as artwork rather than as text, and software reading the file has nothing to recover there. The pypdf documentation describes this same situation plainly: some PDFs do not contain the text as it is displayed but instead an image, and you notice because you cannot copy the text.
Which Canva resume templates are safe to use?
Judge by layout rather than by name, because template names change and layout properties are what extraction actually responds to. The properties that tend to travel well are a single column running top to bottom, contact details typed into the body of the page rather than into a decorative band, section headings written as ordinary words such as Experience and Education, dates and job titles as text rather than as graphics, and a common typeface. The properties that tend to travel badly are a sidebar or any second column, information placed in separate floating boxes, icons doing the work of a label, competence shown as a drawn bar or a row of dots, and anything flattened to preserve its appearance. Pick on those grounds and then verify by testing the export, because the only reliable check is the one you run on the actual file you are sending.
Should I use Canva or a Word template for my resume?
Use whichever gets you a document that a human likes reading and a machine can read back, and be aware that the two tools fail in different directions. A layout tool gives you visual control and a result that generally looks better, and the risk is that the control encourages arrangements which extraction handles poorly. A word processor gives you a plainer result and the risk is different, because features like text boxes, borderless tables used for alignment, and content placed in the header region can cause similar trouble while looking perfectly ordinary on screen. Neither tool is safe by virtue of being that tool. If the role is one where design is part of what you are being assessed on, a reasonable approach is to keep the attractive version for people and a plain version for application forms, since the two documents are being read by very different things.
How do I check whether my Canva resume parses?
Open the exported PDF, select the entire document, copy it, and paste it into a plain text editor such as Notepad, TextEdit in plain text mode, or any code editor. Then read what you get. If the text appears in a sensible order, extraction is likely to reconstruct your record correctly. If sections interleave, so that a line from the sidebar lands in the middle of a job description, the arrangement is being read across the page rather than down the column. If your phone number and email are absent, they are either in a region that the extractor skipped or they exist as artwork. If nothing pastes at all, there is no recoverable text layer. The pdfminer.six FAQ recommends the same check for a related reason and puts it well: if copy and paste gives you proper text, the library should be able to extract proper text too, and if it gives you gibberish, the library will produce gibberish as well.
Is the Canva resume builder good?
As a tool for producing a document that looks good, it has a strong reputation and we are not going to pretend otherwise, particularly among free options. We are also a company that sells resume tools, so treat our opinion on a competitor with appropriate scepticism and weigh it against your own experience of using it. What we can speak to with more confidence is the part that sits downstream, which is whether the file you produce survives being read by software. That question is independent of which tool made it and it is answerable by testing rather than by opinion. So the useful framing is not whether the builder is good but whether the specific document it produced for you is readable, and you can settle that yourself in under a minute with a text editor.
Will a two-column Canva resume get rejected?
Rejection is a strong word and nobody outside the company doing the reading can tell you what happens to your file, so be wary of anyone who states it as fact. The mechanism that is well understood is milder and worse in a way: text in a PDF is absolutely positioned rather than semantically structured, so software reconstructing a linear reading order from a two-column page can interleave the columns, producing a record where your skills list is spliced through your employment history. Nothing is rejected in that scenario. Your experience is simply recorded as something other than what you wrote, and any later matching works from the mangled version. Our article on why Workday scrambles two-column resumes covers that mechanism properly. If you are attached to the two-column look, run the copy-and-paste test and read the order you get back.
What to ask next
If you arrived here from a generative-search prompt, these are the natural follow-ups — each links to the page that resolves it.
Ran the test and did not like what you saw?
Upload the same file and see a structured version of what came back — which sections were recovered, what order they arrived in, and what a matching system would have to work with. Free, no signup, and it will not try to redesign anything.
Continue Reading
Google Docs Resume Templates: Which Ones Survive an ATS
Google Docs is a word processor rather than a layout canvas, which makes it a safer starting point than a design tool: its documents are text-first by default, and a plain single-column file exported with a live text layer generally parses fine. The risk sits somewhere narrower and less discussed — templates that use a table to line the page up, contact details typed into the header region where they can be treated as page furniture, and above all the export step, where a PDF, a Word file, a shared link and a scan hand a parser four completely different things. A six-row decoder maps each export path onto what a parser actually receives, plus a copy-and-paste self-test for your own file.
Read moreWhat Is an ATS-Friendly Resume Format? (2026)
An ATS-friendly resume format is single-column with standard headings, selectable text, and no tables, text boxes, or header/footer contact info. What actually matters for parsing, the PDF-vs-Word truth, and how to build one free.
Read moreAI Resume Builder: Create ATS-Optimized Resumes That Actually Get Interviews
Learn how to build professional, ATS-friendly resumes with AI assistance. 6 customizable templates and instant PDF export.
Read more