A clarification first, because this topic attracts a lot of moralizing and moralizing is not useful.
Nobody serious objects to you using AI. Recruiters use it. Hiring teams draft job ads with it. The tool is fine.
What triggers a negative reaction is a CV in which the candidate cannot be found. And there are very consistent, very learnable signals for that — signals that around 74% of hiring managers say they can spot, with more than half reporting it makes them significantly less likely to hire.
Here is the list, in roughly the order they get noticed.
1. The spearheaded-leveraged verb pile
Every bullet begins with a verb from the same small set: spearheaded, leveraged, orchestrated, drove, championed, pioneered, cultivated, delivered.
"Spearheaded" has been described as the single favourite verb of every major language model. "Leveraged" is filler that almost always replaces a specific technical detail the model did not have.
None of these words is wrong in isolation. The flag triggers when six consecutive bullets open this way with nothing concrete attached. Real people write unevenly. They say built, fixed, took over, argued for, inherited. A CV where every line has the same rhetorical temperature was written by something that does not get tired.
The fix: vary your verbs to match reality, and let some of them be modest. "Inherited a broken reporting process and rebuilt it" says more than "spearheaded reporting transformation initiatives."
2. Vocabulary that does not exist in offices
Linguistic research has identified a specific cluster of statistically AI-overused words: realm, intricate, showcasing, pivotal. Add to that robust, synergy, delve, seamless, holistic, multifaceted.
Nobody has ever said out loud that they operated within the realm of supply chain optimization. That is not how a person describes their job to another person.
The fix: read your CV aloud. Any sentence you would be embarrassed to say to a colleague in a kitchen gets rewritten.
3. Robotic bullet symmetry
Every bullet is between eighteen and twenty-two words. Every section has exactly four. Every line follows an identical verb-object-metric structure.
Humans do not write with that regularity. Some accomplishments need a clause of context; some are a punchy six words. Perfect symmetry is a formatting tell and one of the easiest to notice at a glance.
The fix: break the pattern deliberately. Let your strongest bullet run long. Let a simple one stay short.
4. No specific tools, no specific obstacles
This is the one that actually decides outcomes.
An AI-written bullet says: *optimised data pipeline performance, improving processing efficiency by 40%.*
A human-written bullet says: *rewrote our nightly batch jobs after the warehouse migration — cut the window from six hours to ninety minutes, which meant the Ops team finally had numbers before their 8am standup.*
The second has a named cause, a named beneficiary, and a reason anyone cared. A model cannot produce that, because it does not know your stack, your deadlines, or who was annoyed about the standup.
The fix: for every claim, add one detail only you could know. One is usually enough.
5. Placeholder residue and prompt leakage
Bracketed placeholders left in place. Meta-commentary such as "here is a resume tailored to the role of." Tense that switches mid-bullet because the model forgot it was writing in past tense. Occasionally a hallucinated employer that does not exist.
And the serious version: hidden white-text instructions telling the screening software to rate the candidate highly. Industry data from 2026 attributed 22% of detected deception cases to these prompt injections. Recruiters routinely paste CV text into a plain editor to strip formatting and catch them. It results in immediate rejection and a growing risk of being flagged across platforms.
The fix: proofread in plain text, always. Ninety seconds, and it catches all of this.
6. The em dash tell
Language models use em dashes at roughly two to three times the rate of human writers. Nobody rejects a candidate over punctuation, but combined with the other signals it contributes to the overall impression.
Worth knowing, not worth panicking about.
7. Inflated metrics with no context
AI loves metrics because it has learned that humans love metrics. But it does not know which metrics were available to you, so it invents plausible-sounding ones.
The result is a CV full of suspiciously round, suspiciously large numbers. Increased revenue by 200%. Improved team productivity by 50%. Reduced costs by 45%.
In the current market, a metric without context is a negative signal. It actively costs credibility, which is genuinely new — five years ago any number helped.
The fix: baseline, scope, timeframe. "Grew the SMB book from €800k to €1.1M over four quarters across 60 accounts" is unglamorous and completely credible. It also survives the follow-up question, which the 200% claim will not.
The pattern underneath all seven
Every signal on this list is a symptom of one thing: the document describes a role rather than a person.
That is what is really being detected. Not AI. Absence — the generic version of your job, rendered fluently.
And this matters more now precisely because everyone's CV is polished. Polish is worthless as a differentiator. The scarce thing is specificity. A CV with three genuinely specific, slightly awkward, verifiably human details will beat a flawless generated one in almost any process.
How to use AI without triggering any of this
The right division of labor is not "use less AI." It is "point it at the right job."
Never let AI: write your bullets from scratch, invent your achievements, generate metrics, or write your cover letter end to end. It does not know your tools, your obstacles or your wins, so it will fabricate reasonable-sounding substitutes that you then have to defend in a room.
Always let AI: tighten your phrasing, kill passive voice, catch inconsistencies, restructure a rambling paragraph, and tell you what a recruiter would think.
That last one is the highest-value use and most people never try it. The LeveliU CV Analyzer scores your CV 0 to 100 from an HR recruiter's perspective and itemises which claims read as unproven, which skills are missing, and what to fix. It is effectively a pre-screening: you get the critical feedback while you can still act on it, rather than learning it through silence.
For the raw material, the Achievement Vault does the work AI cannot. You log real wins as they happen, with the messy specifics intact, and the AI polishes your description rather than inventing one. That is the whole distinction — AI editing your truth, not generating a substitute for it.
And when you want your phrasing to sound professional without sounding synthetic, the Corporate Translator turns a plain description of what you actually did into polished business language while showing you the reasoning, so you learn the register rather than outsourcing it permanently.
One test before you send
Take your three strongest bullets. For each, ask: *could a stranger who has never met me have written this from the job ad alone?*
If yes, it is doing nothing for you. Rewrite it with something only you could know.
That single exercise will do more for your callback rate than any amount of keyword optimization.
---
FAQ
What words should I avoid on my CV in 2026?
The most-flagged terms are spearheaded, leveraged, orchestrated, championed, pivotal, intricate, showcasing, realm, delve, robust and synergy. They are not banned words — the flag triggers when they dominate a CV without specific detail attached.
Can recruiters really tell if a CV was written by AI?
Around 74% of hiring managers say they can. What they are actually detecting is the absence of specifics — no named tools, no obstacles, no defensible metrics — rather than the AI itself. A CV with genuine detail does not read as generated even if AI helped write it.
Is it bad to use AI on my resume at all?
No. Using AI to edit, tighten and pressure-test your real material is increasingly standard. Using it to generate achievements you cannot defend in an interview is what causes rejections.
Why are metrics on my CV hurting me?
Because unverified round numbers have become a known AI tell. Add baseline, scope and timeframe to every figure. A smaller number with context beats a bigger one without it.
How do I know if my CV reads as AI-generated?
Get it scored from a recruiter's perspective before you send it, so unproven claims and weak sections can be fixed while it still matters.
---
Find out how your CV reads to a recruiter — before one reads it. Score it free in 60 seconds in the LeveliU app.