6 in 10 Now Write to Dodge AI Suspicion. Fix Your Claims, Not Dashes

You're probably editing your applications so they don't sound like AI wrote them. Six in 10 job seekers are doing the same thing, according to Greenhouse, and most of that effort goes into the wrong things. Most of those people use AI anyway. Recruiters already assume AI was involved. What they can't do is verify what you claim, and that part is in your hands.

What did Greenhouse's job seeker survey actually find?

Most job seekers use AI, most recruiters already assume it, and 6 in 10 candidates are still rewriting their applications so they won't sound like it. Greenhouse surveyed more than 2,800 job seekers in the U.S., U.K., Ireland, Germany and Australia.

The headline numbers:

• 77% of U.S. job seekers used AI in their search in the past year.
• 60% have changed or considered changing how they write to avoid sounding AI-generated. In the U.K. and Ireland it's 63%.
• 71% took an extra step to get noticed. Nearly half went after a referral, and 27% paid for a resume service.

In a separate Oct. 15 release, Greenhouse published the recruiter side.

Put those together. If the reader assumes AI from the start, sounding a little more human doesn't move you out of the "probably AI" group. Everyone is already in it. Greenhouse CPO Sharawn Tipton describes employers sifting through "AI-polished resumes that all look the same." The complaint there is sameness. Nobody can fix sameness with punctuation.

Why does a generic resume get punished so hard right now?

The pile is far bigger than it was in 2022, and your resume only has to look different from the other resumes in it. As reported by Forbes, recruiting teams now handle 412% more applications per recruiter than in 2022, and 111% more applications per job.

Greenhouse CEO Daniel Chait points at one cause: auto-apply tools cost "like 20 bucks, and now they can just shoot out job applications willy-nilly." Applications produced that way lean on the same verbs and the same templates. When a recruiter is skimming, "spearheaded cross-functional initiatives to drive growth" reads exactly like the files before it. That isn't because it sounds robotic. It's because dozens of other people wrote the same line.

The cost is easy to see. In the Greenhouse report, 81% of U.S. candidates were rejected from a job in the past year without speaking to anyone at the company, human or AI. You put real effort into an application and it disappears without a reply. In a pile that size, the generic version is the one that disappears.

Do recruiters check for AI resumes?

Most of them don't need to: 86% of U.S. recruiters already assume at least half their applications are AI-written. The people doing the reading are also mostly not AI specialists, going by Refolk's index of professional profiles.

Some caveats. Every title was matched the same way, but the U.K. count is probably low because "Recruitment Consultant," a common U.K. title, wasn't included. In small 25-profile samples, "Talent Acquisition Partner" was the most common U.K. title (14 of 25) and "Technical Recruiter" the most common U.S. title (8 of 25). Treat those as rough indications. The index includes recruiters at employers such as Experis and K2 Partnering Solutions in the U.S., and Hays, BT Group and Wayfair in the U.K.

Here's my read, and it's an interpretation the data doesn't prove: most people screening you are generalists, not AI specialists. Generalists who are short on time use shortcuts like "it has em dashes" or "it says spearheaded." If you write to get past a shortcut, you're betting it won't change. If you write so your claims can be checked, it doesn't matter which shortcut is popular this month.

Do em dashes and words like "delve" give you away?

They're a weak signal. Taking them out hides the signal but doesn't make the resume any more believable. The punctuation theory spread because it's easy to spot, not because it says much about who wrote the resume.

The evidence:

• A vendor test of 500 ChatGPT-written resumes (vendor content, not peer-reviewed) found "delve" zero times and "tapestry" zero times. The famous AI words didn't show up.
• The same test found that 92% of the 500 resumes had at least one em dash.
• In a small newsletter experiment, the author removed the em dashes from AI-generated resumes and readers then judged them human. As the author put it, when readers have nothing to compare against, "the em dash becomes the whole theory."

The trap is in that last point. Take out the dash and the resume passes the shortcut. It's still the same resume, though, with the same claims anyone could make, and nothing in it is easier to believe. If dashes bother you, swapping them for commas takes seconds. Just don't count it as editing.

Should you run your resume through an AI detector?

No. An AI detector guesses whether text was machine-written by measuring how predictable the wording is. A good resume is supposed to be plain and predictable, so detectors flag the very qualities you want.

The research is not kind to these tools:

Detectors tend to flag word choices that are predictable and sentences that are simple. To score as "human," you have to make your sentences more elaborate and your vocabulary less common. That's the opposite of what a skimming recruiter wants. If English is your second language, it's worse: Weixin Liang and James Zou's study suggests you'll be flagged most often even when you wrote every word yourself. A higher "human" score can leave you with a weaker resume.

What are employers actually worried about?

Whether your claims are true, not who typed them. Robert Half's U.S. survey from March 2026 says this directly: 65% of hiring managers report that the surge in applications, many of them enhanced or generated by AI, has made it harder to verify candidate skills.

Here's what employers say they're doing about it:

1. Spending more time reviewing applications (42%).
2. Adding interviews per candidate (38%).
3. Rewriting job descriptions to discourage generic AI-generated responses (32%).

Look at that list. None of it is about detection. It's all about checking. Robert Half also notes that many candidates use AI responsibly to improve clarity or grammar, and flags AI tools fabricating or embellishing work history and skills. A bullet AI made up is a liability, and with 38% of employers adding interviews, you'll find out when someone asks you about it.

Ophir Samson, Greenhouse's Head of Voice AI, makes the same point: "AI can polish a resume or fire off hundreds of applications, but it can't show an employer how someone thinks or learns." Specific claims you can check are how you show that on paper.

How to make your resume not sound like AI

Replace claims anyone could make with claims only you could make, and make each one easy to check. That's how you humanize an AI resume without giving up the AI.

Run every bullet through these steps:

1. Name the system. Write the actual tool, platform or codebase you worked on, not "key platform."
2. Give the scale. Team size, user count, ticket volume, budget. Scale is hard to fake and quick to check.
3. Show before and after. "Reduced onboarding time from [X days] to [Y days]" beats "streamlined onboarding" every time.
4. Stop worrying about verbs. "Led" is fine. The verb was never the problem. The empty object after it was.
5. Add an anchor someone can check. A launch date, a public link, a named client if you're allowed, a repo.
6. Run the two-minute test. If you couldn't talk about the bullet for two minutes in an interview, rewrite it or delete it.
7. Cut whatever the AI invented. If you didn't do it, it goes, however good it sounds.

A quick before and after:

• Before: "Spearheaded cross-functional initiatives to drive operational efficiency."
• After: "Rebuilt the returns workflow in [your ticketing tool] with a [N]-person support team, cutting average resolution from [before] to [after] over [period]."

The second version could have been drafted by AI. That's fine, as long as every noun in it comes from your actual job.

Tailoring is the second half. A bullet that's specific to you still has to match the posting you're applying to, and with applications per job up 111%, rewriting by hand for every role is where people quit and go back to the generic version. This is where I fit in. Refolk builds your resume from your own history, so the systems, team sizes and numbers are real to begin with. Paste a posting and you get your own resume back, rewritten around what that employer asked for.

AI-written cover letter detection: does the same fix work?

Yes, and it matters even more there, because the cover letter is usually the most generic thing in an application. Employers are already pushing back: 32% told Robert Half they're rewriting job descriptions specifically to discourage generic AI responses. That's a hint. Respond to the specific thing they wrote.

A cover letter that holds up has three parts:

• One line that quotes or names something specific from the posting. A product, a problem, a requirement no other company would phrase the same way.
• One proof story. A real situation from your history that matches it, with the system, the scale and the result.
• A short close. What you'd want to dig into first in the role. No "I am confident I would be a great fit."

The hard part is knowing whether you even have a real story for a given posting. Refolk drafts the cover letter from your history and scores how well you actually fit the role. A low score tells you there's no honest proof story for that job, and your application is better spent on one where there is.

FAQ

Do recruiters check for AI resumes?

Mostly they don't have to, because they already assume it: 86% of U.S. recruiters believe at least half their applications are AI-written. In Refolk's index, only 492 of 122,933 U.S. recruiters mention AI in their headline, which suggests most screeners are generalists working quickly and leaning on rough rules of thumb. Being spotted isn't really the risk. Being generic and hard to verify is.

Is it bad to use AI to write my resume?

No. Greenhouse found 77% of U.S. job seekers used AI in their search, and Robert Half notes many candidates use it responsibly to improve clarity and grammar. The danger is AI inventing or inflating your experience. Use it to draft and tighten, but make sure every claim comes from something you actually did and can talk about in an interview.

How do I humanize an AI resume quickly?

Go bullet by bullet. Replace vague nouns with the named system, add the scale (team size, users, volume), and add a before-and-after result. Delete anything you couldn't defend for two minutes in an interview. That will help you more than swapping out em dashes or chasing synonyms.

Are AI detectors accurate enough to trust?

No. OpenAI withdrew its own classifier on July 20, 2023, after it caught only 26% of AI text and flagged human text 9% of the time. A Stanford study found seven detectors flagged non-native English writing as AI 61% of the time. Detectors reward elaborate, unpredictable wording, which makes for a worse resume.