If your resume came from a Canva template, a parser may be reading text you never wrote, and nobody will tell you. According to a staffing professional whose post resurfaced recently, an AI parser read a candidate's resume as a "circus assistant based in Los Angeles" because of leftover template text. Separately, a Greenhouse survey found that a third of job seekers who try to game screeners have added hidden text, and fraud tools now tag exactly that pattern. Leftover template text in a designed resume can make you look like the wrong person, or like a cheater. A 30-second paste test shows you whether yours has the problem.
What actually happened with the "circus assistant" resume
A staffing professional says an AI tool failed to pull a candidate's name, location, and contact details from her Canva resume and labeled her a "circus assistant based in Los Angeles" instead. The poster blamed placeholder text that Canva had embedded in hidden or layered elements from the original template.
Hold the details loosely. Keepcanva, which wrote up the post, says the failure "allegedly" happened and notes that the exact "ghost data" mechanism is "still being reported anecdotally." The story isn't new, either. Sulat.com ran a version about seven months earlier in which the system read "circus assistant in Los Angeles" instead of "registered nurse in Portland." That version blamed something else: text exported as vector shapes, not placeholder layers.
So the same story comes with two explanations:
- Placeholder layers. Template text you never saw is still in the file, and the parser reads it as yours.
- Vector text. Your real words were exported as shapes, so the parser finds no usable text from you and grabs whatever text is left.
You don't need to know which one is in your file, because one test catches both. The broader problem is well documented. In Enhancv's parse test, Canva resumes averaged 80.07% parse accuracy against 95.77% for Google Docs, and Canva skills sections parsed at 0%. Multi-column layouts and text inside graphics are established parser problems. Ghost text sits on top of them.
The worst part is that you never find out. Nobody emails a nurse to say her profile now reads "circus assistant." She just stops hearing back, decides the market is brutal, and sends the same broken PDF to every posting after that.
Why ghost text now looks like the hidden-text trick
Screeners now scan for hidden text on purpose, and an innocent template leftover can produce the same signals as a deliberate cheat. Being misread is one problem. Being flagged is another.
Hidden text is any text in a resume file that a parser can read but a human can't see: white font, 1px type, or words stacked under another layer. Greenhouse's survey of 2,800+ job seekers in five countries found that 62% have tried tactics to get past automated screening. Those tactics ranged from using AI to work out what the system is looking for (54%) to adding skills or keywords they don't have (40%). A third of those who tried a tactic added hidden text meant for an AI reviewer.
Derived from Greenhouse's survey of 2,800+ job seekers: 62% tried gaming screeners, and a third of them hid text.
Screening vendors now check for it. Nova by Dweet, which integrates with Greenhouse, Lever, Ashby, Teamtailor, Workable, and Pinpoint, documents checks for:
- Invisible text via tiny font sizes
- Keyword stuffing
- Overlapping duplicate text layers
Results go back to Greenhouse as a note and a risk tag on the candidate. Look at the third item. A Canva template that stacks a placeholder text box under your real one produces overlapping duplicate text whether you meant it or not.
For scale, earlier detection figures: Greenhouse found white-text messages in around 1% of the roughly 300 million resumes it processed in the first half of 2025, and ManpowerGroup says it finds hidden text in about 10% of the resumes it scans with AI. Greenhouse CPO Sharawn Tipton summed up the mood: "For the first time I can remember, neither side thinks hiring is working." Employers who think candidates are gaming them will look harder at files that look odd.
Nova's fraud scan only checks resumes good enough to shortlist
Nova's fraud check runs after scoring, and only on candidates above the employer's configured score threshold, which defaults to 7 out of 10. A weak resume with ghost text never gets scanned. A qualified one does.
Ghost text in a strong resume gets inspected. The same text in a weak resume never does.
From the employer's side this makes sense, since you check the people you're about to call. For you, it means your strengths decide whether your file gets inspected. A qualified nurse with a ghost layer in her PDF is exactly the profile that gets the second look.
Don't panic over one stray box, though. Nova's documentation says fraud probability rises "when concerns from different dimensions converge," and "when only surface-level issues are present, risk stays low." A single leftover "Lorem ipsum" is unlikely to trip a flag on its own. Trouble comes when signals stack up, like this:
- A duplicate text layer under your job title
- Tiny labels on decorative skill bars
- A two-column layout that scrambles the reading order
Any one of those looks like sloppy design. All three together start to look like a pattern.
Who reads the parsed version of you
The parsed profile, not your designed PDF, can be the first version of you a recruiter sees, and the healthcare staffing world where this story started is large. In Refolk's index of professional profiles, 8,443 US recruiters work in healthcare, nursing, clinical, or medical hiring. Top employers in that group include RCM Healthcare Services and Green Key Resources.
| Segment (Refolk's index) | Count |
|---|---|
| US healthcare/nurse/clinical/medical recruiters | 8,443 |
| US recruiters/TA listing "Greenhouse" as a skill | 298 |
| UK recruiters/TA listing "Greenhouse" as a skill | 14 |
| US:UK ratio, Greenhouse-listing recruiters | ~21x |
| US healthcare recruiters per Greenhouse-listing recruiter | ~28x |
One caveat: listing "Greenhouse" as a skill badly undercounts the people who actually use it, so read these rows as a picture of who's in the market, not as ATS market share. In the index, the US recruiters who list Greenhouse cluster at companies like Meta, Snowflake, Warby Parker, Datavant, and Rent The Runway. In the UK they work at Monzo, the Financial Times, Atlassian Williams F1 Team, and Hudl.
Here's why it matters. A recruiter filling clinical roles across many requisitions can't open every PDF, so the parsed record is what gets read. If that record says "circus assistant," the beautiful layout never gets seen.
How to run the resume copy-paste test
The resume copy-paste test means selecting all the text in your exported PDF and pasting it into a plain-text editor. It takes about 30 seconds and shows you roughly what a parser extracts. Color is a property of how text is drawn, not part of the text itself, so white or hidden words show up in the paste just as they do for the parser.
- Open the exported PDF in a PDF viewer or browser. Not in Canva.
- Press Ctrl+A (Cmd+A on a Mac), then Ctrl+C (Cmd+C).
- Paste into a plain-text editor, one that strips all formatting.
- Read it top to bottom as if you were a stranger.
What ghost text and duplicate layers look like
Here is an illustrative paste from a resume with template leftovers:
Your Job Title
Jordan Lee
Registered Nurse
Registered Nurse
Lorem ipsum dolor sit amet
Patient Care IV Therapy Led night-shift patient handoffs
That paste has four problems: a placeholder heading, a duplicated title, filler text, and sidebar skills run into a job bullet. Here's how to read what you find:
| What you see in the paste | Likely cause | What a parser may do |
|---|---|---|
| Words you never wrote (template names, cities, "Your Job Title") | Hidden or off-canvas placeholder layers | Read them as your name, title, or location |
| The same line twice | Duplicate text layers stacked on top of each other | Trip an "overlapping text" check |
| Sidebar skills mixed into job bullets | Column reading order | Attach skills to the wrong job, or drop them |
| An empty or near-empty paste | Text exported as vectors, or a flattened PDF | Extract nothing from you at all |
If the paste reads cleanly in order, with your name first and no strangers in it, your file is probably fine. That's the honest answer to "is my resume ATS friendly" at the file level.
How to fix ghost text and re-export clean
Fix ghost text in Canva's Layers panel: delete what you don't own, put your text in reading order, and export without flattening. Then paste-test again, because every export is a new file.
- Open the Layers panel. Canva reads the bottom layer first, so the layer list is effectively your reading order.
- Delete anything you didn't write. Check every text element, including tiny or transparent boxes, anything sitting off the edge of the page, and anything tucked behind a shape.
- Remove duplicates. If two text boxes hold the same title, delete one. Don't just cover it.
- Replace skill bars with words. "IV therapy (advanced)" parses. A mostly filled bar does not.
- Order the layers so your name and contact details sit at the bottom of the list (read first), followed by your summary, experience, and so on.
- Download as PDF, tick "Match reading order to layers," and leave "Flatten PDF" unticked.
- Run the paste test again.
Step 6 goes against a lot of tutorials, which tell you to tick Flatten PDF to "clean up" the file. Flattening merges layers into a single flat image, and one printer warns it can rasterize most of your artwork. For a resume, that means the parser may get no text at all. You'd fix the ghost text by erasing yourself.
If your paste comes back empty even with Flatten unticked, your text is probably exported as vector shapes, and the fastest fix is to stop repairing the template. Enhancv's numbers point toward a plain document (95.77% parse accuracy for Google Docs versus 80.07% for Canva). Rebuilding from your own history is the work Refolk takes off you: it writes your resume as clean, parseable text from what you've actually done, so there are no template layers for anything to hide in. Keep the Canva version for handing out in person if you like the look.
Hidden keywords don't work, and recruiters can see them
Hiding keywords doesn't help you, because white text gets extracted like any other text. It shows up in the parsed profile the recruiter reads and in any plain-text copy of your resume. The trick only works if the machine reads it and the human doesn't, and those two now see the same extracted text.
Nova's own marketing shows what that looks like from the employer's side. Its example of a high-risk flag reads "Invisible text detected (1px white font, 47 hidden keywords)." The vendor claims over 93% fraud detection precision. Treat that as a vendor number, but don't bet your application against it.
If the parser can read your hidden keyword, so can the recruiter. That is the whole trick, and it backfires.
The 40% in Greenhouse's survey who added skills they don't have are solving the wrong problem. If a posting keeps asking for something you've done but your resume doesn't say it, write it plainly in a bullet with the result attached. If you haven't done it, no font color will change that. Refolk handles the honest version: paste the posting, get your own resume back rewritten to surface the experience that matches, plus a fit score that tells you whether the role is worth applying to before you hit submit.
FAQ
Is my Canva resume ATS friendly?
It might be, but don't assume it. Enhancv found Canva resumes averaged 80.07% parse accuracy against 95.77% for Google Docs, and Canva skills sections parsed at 0%. Export your PDF, select all, paste into a plain-text editor, and read the result. If your name comes first, nothing you didn't write appears, and the sections come out in order, the file is in good shape. If not, fix the layers or rebuild in a single-column document.
Will white text get my resume flagged?
It can. Tools like Nova by Dweet check for invisible text via tiny font sizes, keyword stuffing, and overlapping duplicate text layers, and write a risk tag back to the candidate's record in Greenhouse. Nova only runs that check on candidates who score above the threshold (7/10 by default), and risk rises mostly when several signals converge. So white text hurts strong candidates most, which is exactly who shouldn't be using it.
Should I tick "Flatten PDF" when exporting from Canva?
No, not for a resume. Flattening merges layers into a single flat image and can rasterize your content, which can leave a parser with no text to read. Leave Flatten unticked, tick "Match reading order to layers," and run the paste test on the downloaded file, every time you export.