Does Your Resume Sound Like AI? The Tells, and How to Fix Them

The short answer: the things that make a resume read as generated are on the page, where you can find them yourself: a big verb opening every bullet, a summary made of adjectives, results with no numbers, claims nobody could check, and bullets that all come out the same length. Those are also the marks of a generic resume, whoever typed it, and generic is the real cost. Below are nine tells, a rewrite for each, and a five-minute check to run before you send anything.

Plenty of people now start a resume in ChatGPT and then wonder whether it shows. We looked at the detector side of that in can recruiters tell a resume was written by AI. This guide is the hands-on half, and none of it depends on how the resume was written. A person can produce every one of these tells by hand, and an AI draft can lose all of them in one sitting.

A roast can't tell whether a resume was AI-written, so we don't publish a share. What our research page does show is that the phrases our detector quotes most often ("cross-functional", "strategic", "spearheaded" among them) are the ones language models reach for when asked to make a resume sound impressive.

1. The inflated-verb pile-up

What it looks like: spearheaded, orchestrated, leveraged, championed, pioneered. One per bullet, all the way down the page.

Why it reads as generated: ask a model to make a bullet stronger and the verb is the first thing it upgrades. One "spearheaded" is fine. Seven in a row tell the reader the verbs were chosen for sound, and they start discounting all of them.

Before: Spearheaded the implementation of a new inventory system, leveraging cross-functional partnerships to orchestrate a seamless transition.

After: Led the move from spreadsheets to a new inventory system at two warehouses; worked with IT and purchasing to transfer 4,200 SKUs over one weekend with no missed shipments.

Use a plain verb that fits the size of the job (led, ran, built, fixed, set up), then add the detail the fancy verb was covering for. Resume buzzwords to avoid has replacements for the usual offenders.

2. The three-adjective summary

What it looks like: a string of adjectives, a job noun, and "proven track record."

Why it reads as generated: it's the default shape of a generated summary. Each adjective is a claim the page never proves, and the sentence fits every applicant in your field.

Before: Dynamic, results-driven and detail-oriented operations professional with a proven track record of driving efficiency.

After: Operations manager, eight years in third-party logistics. Run a 30-person night shift; raised on-time dispatch from 88% to 97% last year.

Keep one claim and back it with one fact. Let the reader decide you're detail-oriented.

3. Every bullet the same length and rhythm

What it looks like: verb, object, a "resulting in" or "driving" clause, a vague outcome, repeated. Every bullet runs to about two lines, so the right edge is a straight wall.

Why it reads as generated: people write unevenly, because some of what they did takes six words to say and some takes thirty. A model fills one template over and over, and the sameness is visible before anyone reads a word.

Before:

• Managed customer inquiries across multiple channels, resulting in improved satisfaction scores.
• Collaborated with internal teams to resolve escalations, driving faster resolution times.
• Developed knowledge base articles to empower customers, reducing overall ticket volume.

After:

• Answered 60 to 80 tickets a day across chat, email and phone.
• Wrote the 25 help articles that now handle most password and billing questions; monthly tickets on those topics fell by about a third.
• Took over escalations for our two largest accounts.

Don't roughen them on purpose. Write each bullet as long as what you did needs, and the lengths will vary on their own.

4. Outcomes with no numbers

What it looks like: "significantly improved", "greatly enhanced", "drove substantial growth", "streamlined processes".

Why it reads as generated: a model asked to improve a bullet adds an outcome without knowing what the outcome was, so you get the shape of a result with nothing inside. In our own data, claims with no number in them are the largest single category the detector flags, though not by much, and only about half of experience bullets carry a number at all.

Before: Significantly improved the month-end close process, enhancing accuracy and efficiency across the finance team.

After: Cut month-end close from nine working days to five by moving reconciliations onto a shared checklist; audit adjustments dropped from 14 to 3.

If you don't have the exact figure, estimate it and say "roughly", or count something you do know: volume, people, before and after. How to quantify your resume covers where to find them.

5. Suspiciously perfect parallel structure

What it looks like: every list inside a sentence has three items, sometimes two such lists in one line.

Why it reads as generated: real lists are as long as the number of things that happened, sometimes two, sometimes five. When every list has exactly three items, the count is coming from a habit, and models have this one badly.

Before: Partnered with sales, marketing and product to define, launch and scale a new pricing tier.

After: Set pricing for a new mid-market tier with the sales team and ran the launch; 41 accounts upgraded in the first quarter.

6. Buzzword clusters

What it looks like: "fast-paced environment", "key stakeholders", "strategic initiatives", several packed into one line.

Why it reads as generated: any one of these can survive on its own. Four in one sentence means the sentence has the sound of work and no content. "Cross-functional" is the phrase our detector quotes more than any other, on about one resume in ten, and that's a floor, since each roast quotes only five phrases.

Before: Drove strategic cross-functional initiatives in a fast-paced, dynamic environment to deliver value to key stakeholders.

After: Ran quarterly planning for a 120-person product group: gathered requests from five teams, ranked them with the VP of Product and published the roadmap.

7. Claims nobody could verify

What it looks like: "Instrumental in the company's growth." "Recognized as a top performer."

Why it reads as generated: these sound like achievements but name no award, figure or date, so a model can write them about anyone. Humans write plenty of them too: close to half of the resumes we score make at least one self-description with nothing on the page to back it up.

Before: Recognized as a top performer and trusted advisor to senior leadership.

After: Ranked first of 22 account managers on renewal rate in 2025; presented the renewals forecast to the CFO and board twice.

The test is whether someone could check it, and whether you could tell the story in an interview. The interview-test prompt in seven ChatGPT prompts for an honest resume review is built to catch exactly these lines.

8. Em dashes everywhere

What it looks like: a long dash in nearly every bullet, splicing a result onto a duty where a comma or a period would have done.

Why it reads as generated: language models use a lot of em dashes. So do plenty of good human writers, which is why this is the weakest tell on the list. One or two on a page say nothing. A dash in almost every line suggests a draft nobody reread.

Before: a bullet that reads as three fragments joined by two long dashes: rebuilt the onboarding program, dash, ramp-up cut from six weeks to three, dash, 14 new hires trained.

After: Rebuilt the onboarding program and trained 14 new hires on it; ramp-up time fell from six weeks to three.

Fix the other eight first. This one takes a minute with find and replace.

9. A skills list that mirrors the posting

What it looks like: the skills section repeats the posting's requirements in the same order and the same words.

Why it reads as generated: paste a posting into a chat, ask for a tailored resume, and this is what tends to come back. Matching keywords is good practice. Copying the phrasing verbatim, including qualities that aren't skills, reads as a copy. "Ability to thrive in ambiguity" is something an employer asks for, and an odd thing to list about yourself.

Before: Skills: Stakeholder Management, Data-Driven Decision Making, Excellent Communication, Ability to Thrive in Ambiguity

After: Skills: SQL, Looker, Excel (pivot tables, Power Query), A/B test design, Jira. Plus, in the experience section: "Ran the weekly metrics review with the heads of sales, support and finance."

Keep the posting's hard skills that you really have, and move the soft ones into bullets where you can show them.

A five-minute check

Print it or open it full screen, and give each step about a minute.

1. Circle the first word of every bullet. If more than two or three are big verbs (spearheaded, orchestrated, leveraged, championed), swap the rest for plain ones.
2. Search for the empty outcomes. Find "significantly", "improved", "enhanced", "streamlined", "various" and "multiple". Every hit needs a number beside it or a rewrite. While the search box is open, count your long dashes.
3. Cover your name and read the summary. If it could belong to anyone in your field, trade two of the adjectives for one fact.
4. Zoom out until you can't read the words. If every bullet is the same length, cut the padded ones down. Then hold your skills list next to the posting and look for phrases copied whole.
5. Read each claim the way an interviewer would. Who, how many, compared to what. Anything you couldn't answer in a sentence gets the missing detail or comes out.

For a second opinion on step 2, the BS Detector scores buzzword density from 0 to 100 and quotes your worst phrases by name. It measures fluff, not authorship, so it can't say whether AI wrote a resume.

Using AI well

None of this means you should stop drafting with AI. It's good at getting you past a blank page and at offering a few versions of a clumsy sentence. Treat its draft as scaffolding.

Give it raw material. Paste your notes: what you did, the numbers you have, the job you want. "Make this sound more impressive" is how you get tell number one.

Treat every claim in the draft as a placeholder. Where it says "significantly improved", put in the real figure. Where it says "key stakeholders", name the teams. Where you have nothing to put in, cut the line.

Never keep a number you didn't supply. Models will fill a gap with a tidy percentage. If you couldn't explain it in an interview, it comes out.

Read it aloud once. Lines you'd never say out loud are usually the ones that came from the draft.

The goal was never to hide the tool. A resume where every line holds a detail only you know reads as yours, however the first draft got made.