Resume Action Verbs in 2026: 18 to Retire, 12 to Keep

A recruiter at a Series B fintech opened her queue on a Tuesday in May and read three resumes in a row that started every bullet the same way. Spearheaded the rollout. Orchestrated the team. Leveraged cross-functional alignment. She moved all three into a separate pile and kept reading. That is a real story she told on LinkedIn last week, and it is what the action-verb section of your resume is now scored against in 2026.

The verb list everyone has been working from since 2018 was written before ChatGPT existed. Half of it now does the opposite of what it was meant to do. This post is the updated taxonomy: 18 verbs to delete tonight, 12 verbs that still land interviews, and the rule that makes any verb carry weight regardless of which list it lives on.

Do resume action verbs still matter in 2026?

Yes, and more than they used to, because recruiters are reading more resumes per role than ever. Greenhouse said applications per recruiter jumped 412 percent since 2023. When a recruiter has 60 resumes to clear before lunch, the first word of every bullet is the only word they are guaranteed to read. That word is your action verb.

The 2018 Ladders eye-tracking study put the first-pass resume scan at 7.4 seconds. Most of those seconds land on the F-pattern: name, current title, recent dates, then a vertical skim down the bullet stems. The recruiter reads the first one to three words of each line and decides whether the rest of the line is worth a second look. If the verb at the front does not make her eyes stop, the rest of the bullet does not happen.

Multiple career sites cite a finding that resumes leading with strong action verbs draw a roughly 140 percent interview lift versus resumes leading with passive openers (responsible for, helped with, tasked with). The original source for that figure is harder to nail down than the repetition suggests, so treat it as directional, not gospel. What is not directional is the inverse: every recruiter survey since 2023 finds responsible for and helped with as the two phrases that get bullets skipped fastest. Same study, every year.

The first word of every bullet is the only word a recruiter is guaranteed to read. Pick it like that is true, because it is.

Why the 2018 power-verb list is half-dead in 2026

Two years ago, the standard advice was the Resume Worded and Harvard lists: spearheaded, orchestrated, leveraged, delved, optimized, streamlined. Those words signaled seniority. By June 2026 they signal something different. They signal that ChatGPT wrote your resume.

The mechanism behind the shift is documented. In December 2024, two Florida State University researchers (Tom Juzek and Zina Ward) published a paper titled Why Does ChatGPT "Delve" So Much?They identified 21 focal words whose use in scientific writing spiked after ChatGPT launched, including delve, intricate, underscore, surpass, boast, meticulous, strategically, and garner. The cause they pinned down was Reinforcement Learning from Human Feedback (RLHF): when human raters keep giving thumbs-up to responses containing these words, the model learns to use them more often.

Then in August 2025, a follow-up FSU study by Bryce Anderson and Riley Galpin analyzed 22.1 million words of unscripted spoken podcast and YouTube transcripts and found the same AI vocabulary now spreading into conversational speech. The buzzword underscore is up; accentuate (its synonym) is not. Some target words more than doubled in spoken frequency. The model is not just writing your resume. It is reshaping the vocabulary recruiters are now trained to flag.

The detection numbers track. Resume Genius' 2026 Hiring Insights Report (1,000 US hiring managers) put detection at 80 percent, with only 4 percent saying they never notice signs of AI use. Two years earlier, the same survey put detection at 53 percent. The line is steep. And the cost is concrete: Resume Now's AI and the Applicant Report found 62 percent of employers say a resume generated by AI without personalization often leads to rejection.

On the volume side, a ResumeCoach analysis of 110 common resume words across 111 million Indeed Resume Search profiles found managedon 9.1 million resumes (one in twelve), organized on 7.1 million, accurate on 5.3 million. None of those numbers prove a word should never be used. They prove that a recruiter has seen the word already today, twice, and is not going to read it as a signal. The verbs at the top of the 2018 lists are now the verbs at the top of the recruiter's skip-list.

The 2026 verb taxonomy: 4 buckets your verbs fall into

Every verb on your resume now lives in one of four buckets. The bucket decides how the recruiter reads the bullet, before she reads the rest of it. The framework:

1. Dead verbs. Originally power verbs, now AI signatures. Delete unless you are quoting yourself in a quoted client testimonial.
2. Bleached verbs. Technically fine, but recruiters see them on every fifth resume. Only safe if paired with a specific noun and a number.
3. Working verbs. The boring concrete ones. Two syllables, one meaning, what humans say at a bar. The list to lead with.
4. Underused verbs. Verbs that everyone quietly avoids because they sound too plain. They are the highest-signal verbs in 2026, because nobody else is using them.

Dead verbs: the 18 to retire tonight

The signature of an AI-written resume. Each of these used to land. In 2026 each one tells the recruiter you did not edit what the model gave you.

• Spearheaded. The single most-cited AI tell. Replace with Led, Built, Started, Owned, or just Ran.
• Orchestrated. Co-occurs with spearheaded in nearly every AI-flagged cluster. Replace with Coordinated, Ran, or Organized.
• Leveraged (as a verb). The classic tell. Used, Applied, Built on, or Drew on are all shorter and clearer. The noun (leverage) is fine.
• Streamlined. Process-improvement buzzword that now reads as filler. Replace with the specific change you made (Cut, Combined, Rewrote, Replaced).
• Optimized. Either name what you changed and by how much, or cut the word. Optimization without a metric is a tell.
• Showcased. Marketing-deck padding. Replace with Presented, Shipped, or Demonstrated.
• Utilized. A six-letter way of writing Used.
• Harnessed. AI-favored verb. Used or Applied carries the same meaning in half the syllables.
• Empowered. Corporate cliche before ChatGPT, AI signature after. Replace with Trained, Coached, or Hired.
• Elevated. Same family as empowered. Cut.
• Synergized. Was never a real verb. Delete entirely.
• Pivoted. Tech-startup jargon that AI now overproduces. Replace with Changed, Switched, or Moved.
• Catalyzed. Pretentious. Started or Triggered carries the same meaning.
• Architected. Acceptable only when you literally designed software architecture. Otherwise Designed or Built.
• Engineered. Same rule. Use only for engineering work; otherwise Built or Made.
• Crafted. Etsy-tier. Wrote, Built, or Made.
• Delved. The flagship AI verb. Never appears in normal writing. Replace with Studied, Reviewed, or Researched.
• Drove (without a number). Drove $4.2M ARR is fine. Drove growth, drove engagement, drove alignment is filler. Cut or quantify.

Bleached verbs: 8 that need a noun rescue

These are technically strong verbs, but they show up on millions of resumes (managed alone is on 9.1 million per the ResumeCoach analysis). They are safe to use, but only when paired with a specific noun and a number in the same bullet.

• Managed. Followed by what (a 6-person team, $1.2M budget, the EMEA pipeline) and a result.
• Led. Same rule. Led what, for whom, with what outcome.
• Improved. Improved what metric from what to what. No abstract improvements.
• Increased. Same. Always followed by a number.
• Developed. Developed what specific thing, used by whom.
• Implemented. Implemented what tool, replacing what, in how long.
• Coordinated. Coordinated whom, across how many teams or sites.
• Created. Created what, that did what. Created alone is empty.

Working verbs: the 12 still landing interviews

The boring two-syllable verbs that survived the AI wave because they always meant exactly one thing. They sound plain on the page. They are the highest-signal verbs in 2026.

• Built. The strongest single verb in 2026 for any maker role. Built means a thing now exists because of you.
• Shipped. Engineering and product gold. Shipped means the thing left the building, not just the design doc.
• Wrote. Underused for any role that produced a real artifact (docs, contracts, copy, code, RFPs).
• Cut. The cleanest verb for any reduction. Cut churn from 7.2 percent to 4.8 percent.
• Hired. Specific, verifiable, hard to fake. Hired the first 4 engineers on the data team.
• Closed. Sales gold. Closed $2.3M in pipeline across 11 SMB accounts.
• Sold. Same family. Plain and undeniable when paired with a number.
• Reduced. Bleached cousin, but Reduced still works when paired with a real metric and a baseline.
• Grew. Grew the newsletter from 800 to 14,400 in 9 months. Specific scope, specific math.
• Designed. For any work where you produced a plan or a system that others executed.
• Negotiated. Sales, legal, ops, procurement. Negotiated a 22 percent price cut on the AWS Reserved Instance contract.
• Migrated. Specific to data, systems, customer base. Hard to fake, easy to verify on a back-channel call.

Underused verbs: 6 nobody picks but should

These are the verbs everyone leaves out because they sound too plain. In 2026 plain is the signal. Each of these names a specific human action no model produces by default.

• Called. Called 40 customers in week 1, found the cancellation reason.
• Mailed. Sent the cold outreach by name, not via tool. Mailed 120 hand-typed cold notes.
• Bought. Specific to buyer roles (sourcing, procurement, M&A, growth-via-acquisition).
• Replaced. Replaced the legacy CRM with HubSpot in 11 weeks. Specific, dated, scoped.
• Talked. Talked 230 prospective students through admissions in one cycle. Plain. Verifiable. Recruiter-trusted.
• Fired. Underused but powerful for management roles. Fired the underperforming vendor; brought it in-house; cut spend 38 percent.

The bar test: would you say it out loud?

The single fastest filter for whether a verb survives 2026: would you use it in a sentence at a bar? If you would tell a friend you spearheaded an initiative, keep it. If you would never say that out loud to another person, the verb is the AI talking, not you. Built, wrote, cut, sold, hired, called - all of these pass the bar test. Spearheaded, orchestrated, harnessed, elevated, none of them do.

The bar test catches the verbs the FSU researchers documented in spoken speech. The model is now leaking into the way people talk in private, but it leaks the words that sound impressive on a page, not the words people use in conversation. A bullet that survives the bar test reads as written by a human at the keyboard. A bullet that fails it reads as ChatGPT, even when ChatGPT had nothing to do with it.

How to use the working verbs: pair them with one specific noun

The single highest-impact edit on a 2026 resume is not picking a stronger verb. It is pairing any verb with one specific noun in the same bullet.

Modern semantic ATS engines (Greenhouse, Lever, Workday) score verbs based on what surrounds them. The internal rule the parsers use is sometimes called the semantic cluster: a verb like Built expects a real noun nearby (a pipeline, a dashboard, a team, an API). A verb with no noun, or with a generic noun like initiative, system, or solution, ranks lower because it cannot be matched to the required skill in the job description. Worse, the same pattern is what recruiters scan for when flagging AI copy.

The structural rule: Working verb + specific noun + number. All three or none. Examples:

• Weak: Spearheaded process improvements across the team.
• Strong: Cut onboarding from 28 days to 12 days by rewriting the role-specific runbooks.
• Weak: Orchestrated cross-functional initiatives.
• Strong: Ran the weekly sync between sales, finance, and product that closed the 2024 budget gap of $410k.
• Weak: Leveraged data to optimize results.
• Strong: Wrote the SQL queries that flagged the 240 dormant accounts; 71 reactivated in the next 30 days.

A worked example: one bullet, three rewrites

Original bullet from a real CV submitted to our free CV score tool last month (anonymized; the user gave permission to use the line). Marketing manager applying to a senior brand role:

Four dead verbs in one sentence (spearheaded, leveraged, elevate, optimize). Zero numbers. Zero specific channels. Zero specific partners. The recruiter reads the first word (spearheaded) and her eye keeps moving down the page.

Pass one (cut the dead verbs):

Better, still empty. Working verbs without nouns and numbers still read as filler. The semantic cluster is still missing.

Pass two (add the noun and the number):

Same person, same work, three completely different signals. The third version names the launch, the channels, the timeline, and two metrics with bounded before-and-after numbers. The verb is plain (Ran). The noun is specific (the Q3 2025 product launch). The numbers are not round. The recruiter cannot skip past this bullet.

The fix is not a stronger verb. The fix is the boring verb plus the noun the recruiter cannot make up.

Working verbs by role (the short version)

Every role has its own working-verb cluster. If a verb on this list does not match the work you did, do not use it. The fast lookup:

• Software engineer: Built, Shipped, Wrote, Refactored, Migrated, Tested, Deployed, Debugged, Reduced, Cut, Designed.
• Product manager: Shipped, Wrote, Ran, Prioritized, Killed (a feature), Hired, Grew, Cut, Defined, Scoped.
• Marketing: Ran, Wrote, Grew, Sent, Tested, Cut (CAC), Hired, Closed (sponsorships), Negotiated, Designed.
• Sales: Closed, Sold, Negotiated, Called, Booked, Quoted, Renewed, Grew, Cut (cycle), Won.
• Finance and FP&A: Modeled, Forecast, Audited, Cut, Grew, Negotiated, Closed (books), Reconciled, Reported, Built.
• Designer: Designed, Shipped, Wrote, Built (the system), Researched, Tested, Reduced (a step), Replaced, Drew, Coded (in tools like Figma or Webflow).
• Operations: Ran, Hired, Fired, Bought, Replaced, Cut, Built, Wrote (the SOP), Negotiated, Mapped.
• Healthcare: Treated, Triaged, Documented, Trained, Coordinated, Reduced (length of stay), Charted, Counseled.
• Teaching: Taught, Wrote, Tested, Mentored, Counseled, Designed (the lesson), Coached, Hired (or recruited students).

Pair any of these with a noun the recruiter cannot invent. The verb is the doorway. The noun and the number are the room.

Common mistakes (with the specific fix)

• Starting every bullet with the same verb. Even working verbs lose signal when stacked. If three bullets in a row start with Built, change two to Wrote, Shipped, or Designed. Variety reads as the writer paying attention.
• Using an inflated verb to claim authority you did not have. Spearheaded for work you contributed to as one of six gets caught on the back-channel reference call. About 5 to 10 percent of offers are pulled after the reference catches a claim that does not hold up. Use Contributed to or Co-led when honest.
• Pairing a working verb with an empty noun. Built initiatives. Shipped solutions. Designed frameworks. Initiative, solution, and framework are empty calories. If the noun is generic, the bullet is still filler.
• Round numbers next to a working verb. Cut costs by 50 percent reads as fabricated because real measurements rarely land on a multiple of five. Cut costs from $148k to $89k (a 40 percent drop) reads as measured.
• Writing for the ATS, not the recruiter. Even semantic-matching ATS engines (Greenhouse, Lever, Workday) now penalize unnatural repetition in their ranking layer. A bullet stuffed with the job-description phrase three times scores lower than one that names the work once, in plain English, with the noun and the number.
• Cutting all the bleached verbs. Managed and Led are bleached, not banned. A bullet like Managed a 6-person customer-success team that cut churn from 7.2 to 4.1 percent is fine. The verb does not have to be rare; the noun and the metric have to be specific.

If you want the verb-and-noun audit done on your current CV automatically, our Glow Up rewrite runs the 2026 verb taxonomy on every bullet (the dead ones get swapped, the bleached ones get noun-paired, the working ones get a number) and gives you a free preview before you decide whether to apply the rewrites. If you want the diagnostic first without the rewrite, the free CV score flags every dead verb and every empty noun pairing in about 90 seconds.

FAQ

What is the strongest single action verb for a resume in 2026?

Built. Two syllables, one meaning, hard to fake on a back-channel reference call, applies cleanly to engineering, product, design, ops, sales, and creative work. The next three are Shipped, Wrote, and Cut. None of them sound impressive on the page, and that is exactly why they survived the AI wave.

Is spearheaded still acceptable on a resume?

Almost never in 2026. Multiple recruiter surveys this year put spearheaded on the top of the AI-tell list. If you actually did spearhead something, Led or Started or Owned carries the same meaning in fewer syllables, without triggering the AI flag. Reserve spearheaded for situations where no other verb fits (you were the named founder of an initiative that did not exist before you), and pair it with a specific noun and a number, or do not use it.

How many action verbs should appear on one resume?

One per bullet, no repeats within any given role. If you have 6 bullets under a role, that is 6 distinct verbs. If the same verb fits twice, the bullet is probably saying the same thing twice and one of them should be cut. Across an entire resume of 4 roles and roughly 18 bullets, aim for 12 to 15 unique verbs. Anything tighter than that reads as repetitive; anything wider reads as a thesaurus sprint.

Will the AI signature change again in 2027?

Yes. The retired list in this post is accurate for June 2026 because the FSU studies caught the words ChatGPT was overproducing through 2024 and 2025. The newer models (GPT-5, Claude 4.6, Gemini 2.5) are already shifting which words they overproduce. The principle holds: long Latinate verbs with corporate connotation will keep ending up on the retired list, and short verbs that name what humans do (built, sold, hired, called) will keep being safe. Run a fresh check on your resume every 6 months.

Do recruiters actually reject resumes for using one wrong verb?

Not for one verb. They reject for the pattern of three or four wrong verbs in the first six bullets, which is how an AI-written first draft typically reads. Resume Genius found 80 percent of hiring managers say they spot AI; Resume Now found 62 percent say AI-generated resumes without personalization often lead to rejection. The cost is real. The fix is the same fix as everything else in this post: working verb, specific noun, real number.

Read next

If this post helped, three other CVHive posts go deeper on the surrounding work. The companion piece is 80+ resume bullet point examples (the XYZ formula, full bullets by role, the plausibility test). For the AI side specifically, how to humanize an AI resume names the 3-Edit Pass that ties this verb work to sentence shape and summary. The deeper detection-side view lives in can recruiters detect an AI resume (the 7 tells that pair with the verb signature). For the cold-application version of the same problem, the cold-email hiring-manager template applies the same verb taxonomy to outreach. And five real users who rebuilt their CVs around the working-verb approach (Maria K., Daniel O., Priya S., Liana R., James T.) are on the CVHive stories page.