AI Words to Avoid in Your Resume (and Plain Replacements)
Language models have favourite words, and once you know them you see them everywhere: in emails, LinkedIn posts, and resumes that came out of ChatGPT with one click. This page lists the words that published research has measured as LLM-overused, then the resume clichés that travel with them, each with a plain replacement. It ends with a find-and-replace pass you can run in ten minutes.
The evidence: two studies of LLM vocabulary
Two research groups measured how word frequencies shifted after ChatGPT arrived, using very large text collections.
Kobak, González-Márquez, Horvát and Lause, Delving into LLM-assisted writing in biomedical publications through excess vocabulary. They analysed more than 15 million PubMed abstracts from 2010 to 2024 and identified 379 "excess" style words in 2024. Their headline estimate is that "at least 13.5% of 2024 abstracts were processed with LLMs." The rarer words had the largest frequency ratios: "delves" (28.0), "underscores" (13.8) and "showcasing" (10.7). Common words such as "potential," "findings" and "crucial" showed the largest absolute increases. Other flagged words include "intricate," "notably," "comprehensive," "pivotal," "meticulous" and "meticulously."
Liang and colleagues, Monitoring AI-Modified Content at Scale. They studied peer reviews at major AI conferences and estimated that "between 6.5% and 16.9% of text submitted as peer reviews" had been substantially modified by LLMs. In ICLR 2024 reviews, they report that "commendable," "meticulous" and "intricate" showed "9.8, 34.7, and 11.2-fold increases in probability of occurring in a sentence."
Neither study looked at resumes. They show which words today's models reach for when they write formal English, and a resume is formal English.
The research-backed list, with replacements
| Word | Measured in | Why it hurts on a resume | Write instead |
|---|---|---|---|
| delve / delves / delving | Kobak et al. (r = 28.0 for 'delves') | Says you looked into something, not what you found | analysed, investigated, audited, plus what you found |
| meticulous / meticulously | Liang et al. (34.7x); Kobak et al. | Self-praise with no evidence | Show the care: 'zero audit findings across 4 quarters' |
| intricate / intricacies | Liang et al. (11.2x); Kobak et al. | Vague about what was complex | Name it: '14-entity consolidation', '3 legacy APIs' |
| commendable | Liang et al. (9.8x) | A reviewer's word, not a candidate's | Cut it; state the result |
| underscores | Kobak et al. (r = 13.8) | Essay vocabulary in a bullet | shows, or drop the clause |
| showcasing | Kobak et al. (r = 10.7) | Presentation word, not an outcome | built, shipped, presented to [audience] |
| pivotal | Kobak et al. | Asserts importance instead of proving it | Say what depended on it |
| comprehensive | Kobak et al. | Unbounded claim | Give the scope: 'covering 120 SKUs' |
| crucial | Kobak et al. | Filler intensity | Cut, or name the stakes |
| notably | Kobak et al. | Essay connector | Cut |
The resume clichés that travel with them
These next words are not from the studies above. They are resume clichés that tend to appear in the same AI-generated drafts. Treat this as editorial advice, not measured data. The problem is the same: each one claims a quality instead of showing a fact.
| Cliché | Plain replacement |
|---|---|
| spearheaded | led, started, set up |
| leveraged | used |
| results-driven / dynamic / passionate | cut; let the bullets show it |
| robust / seamless | name the property: '99.9% uptime', 'no manual steps' |
| cross-functional synergy | name the teams: 'with finance and legal' |
| fostered a culture of | ran, introduced, trained |
| in today's fast-paced environment | cut |
| a proven track record of | cut; show the track record |
One exception matters. Some phrases look like buzzwords but are real job-posting requirements. In the ATS Resume AI analysis of all 503 S&P 500 careers sites and 9,260 live job postings, June 2026, "cross-functional collaboration" appeared in 63.9% of full-text product manager postings and "stakeholder management" in 55.3% of finance postings (role keyword data). If the posting uses the phrase, keep it once, attached to evidence: "Ran stakeholder management for a 6-team ERP rollout."
Before and after
A summary written in one click from a job title:
Meticulous, results-driven finance professional who delves into intricate data to deliver comprehensive insights, showcasing a commendable track record of pivotal contributions.
It has seven of the research-flagged words and no facts. The rewrite uses only the person's real material (the numbers here are illustrative):
Senior financial analyst, 6 years in FP&A at two SaaS companies. Built the 3-statement model behind a $40M Series C; cut month-end close from 9 days to 5 by automating reconciliations in Excel and NetSuite.
Nothing in the second version could be produced without the person's input. That is why it does not read as AI.
Why models write this way, and why it matters on a resume
Neither study claims these words are wrong. "Crucial" and "comprehensive" are ordinary English. The problem is frequency and substitution. A model asked to make a bullet "sound professional" reaches for words that signal quality ("meticulous," "pivotal"), because it does not have the facts that would show quality. You do have them.
That is also why removing the words is not enough. Swapping "meticulously" for "carefully" leaves the same empty claim. The fix that works is to replace the adjective with the evidence it stood in for:
- not "meticulous attention to detail" but "reconciled 1,200 vendor invoices a month with no payment errors in 2025"
- not "pivotal role in the launch" but "wrote the pricing page and onboarding emails for the launch; 2,300 trial signups in week one"
- not "comprehensive knowledge of AWS" but "run production on EC2, RDS and Lambda; AWS Solutions Architect – Associate".
The figures above are placeholders for your own. If you do not have a number, describe scope instead: how many people, systems, customers or markets.
Words worth keeping
Plain, specific verbs never read as AI because they force a concrete object: built, wrote, ran, cut, hired, trained, closed, shipped, negotiated, migrated, reconciled, forecast, audited, designed, tested, fixed. Pair each with a named tool or system and a result, and the bullet carries its own proof. Our resume action verbs list groups them by function.
A ten-minute find-and-replace pass
Search for each flagged word
Open your resume and search for: delve, meticulous, intricate, commendable, underscore, showcas, pivotal, comprehensive, crucial, notably. Searching word stems catches every form.
Ask what the word was hiding
For each hit, write down the fact it replaced: a tool, a number, a team, a deadline. If there is no fact, delete the word.
Rewrite as verb + object + result
"Analysed 18 months of churn data in SQL; found two onboarding steps behind 40% of cancellations." Use your own figures.
Run the cliché list
Search for spearhead, leverag, results-driven, dynamic, robust, seamless, synergy. Replace or cut.
Check the posting's own terms survived
Compare against the job description with the free keyword generator so editing out jargon did not also remove required skills.
Stop the words at the source
If you use an AI model to draft, put the ban in the prompt so you are not cleaning up afterwards:
Rewrite these resume bullets for the job posting below. Use only facts from my resume. Do not use: delve, meticulous, intricate, commendable, underscore, showcase, pivotal, comprehensive, crucial, notably, spearhead, leverage, results-driven, dynamic, robust, seamless. Start each bullet with a plain past-tense verb. Where a number is missing, write [add metric] instead of inventing one.
For more prompts, see ChatGPT prompts to tailor a resume, and for bullet structure, resume bullet points for ATS.
Sources
- Kobak, González-Márquez, Horvát and Lause, arXiv — Delving into LLM-assisted writing in biomedical publications through excess vocabulary
- Liang et al., arXiv — Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews
Frequently asked questions
- Is 'leverage' an AI word?
- It is not one of the words the two research papers cited here measured. It is a resume cliché that often appears in AI drafts. 'Used' is almost always clearer, so replacing it costs nothing.
- Will an ATS reject my resume for using these words?
- The ATS documentation reviewed for this series (Workday HiredScore, Eightfold, Phenom) describes matching your resume against the job's qualifications, not scoring your vocabulary. The cost is with the human reader, who sees a generic resume.
- Can I keep 'stakeholder management' if the job posting uses it?
- Yes. When a posting names a skill, it is a keyword the recruiter may search for. Keep it once, attached to a concrete example, rather than as free-floating jargon.
- Do these words matter in a cover letter too?
- Yes, arguably more. A cover letter is longer prose, which is where model habits like 'delve' and 'underscore' show most.