Most resume keyword lists are somebody's opinion. This one is a count. Between 2026-08-29 and 2026-09-09 we read every open posting on 81 companies' public job boards for seven roles — 3,518 postings — and counted how many of them name each term.
Each row below links to that role's full table, where every term carries its own count and interval. The method and the limits are at the bottom of this page, and they apply to all seven.
| Role | Postings | Companies | Terms | Collected |
|---|---|---|---|---|
| Software engineer → Practice words outrank the languages | 1,360 | 72 | 45 | 2026-08-29 |
| Account executive → A methodology is named more often than the CRM | 921 | 52 | 44 | 2026-09-09 |
| Product manager → The vocabulary PM resumes lead with is largely absent | 421 | 62 | 44 | 2026-09-09 |
| Marketing → The ad platforms barely appear; pipeline does | 347 | 62 | 44 | 2026-08-30 |
| Cybersecurity → Python is asked for more often than any certification | 212 | 51 | 40 | 2026-08-29 |
| DevOps and SRE → Named tools rank higher here than in any other role | 134 | 41 | 42 | 2026-09-09 |
| Accountant → The credential and the close, ahead of any system | 123 | 49 | 30 | 2026-08-29 |
3,518 postings, 81 distinct companies. The role totals sum to more company entries than that, because many companies hire for several of these roles.
Each bar is the share of that role's postings naming the term. Read them down a column, not across: every role has its own sample and its own term list, so the shapes are comparable and the numbers are not.
The generic word beats the branded one. In every role where both were counted, the practice outranks the product that implements it: observability over Grafana, SIEM over Splunk, experimentation over A/B testing. A resume that names only the tool misses the word the posting actually uses.
Tools are not the top of any list except one. Software engineer and the rest lead with practices; DevOps is the single role where named products reach the top five, and even there automation at 81.3% leads them.
The same work is described in different words by role. That is the reason these are seven studies and not one list: the terms that matter for accountant postings barely occur in software engineer postings, and a single combined ranking would hide both.
Where the postings came from. The Greenhouse job board API, which companies publish so their own careers pages can render their listings. Nothing scraped, nothing private. The collector and the analyzer are in the repository, so the numbers can be reproduced against the same endpoint.
Which postings counted. Title match on the role, individual contributors only. Managers, directors and program managers are excluded, because a posting for the person who runs the team asks for different things than a posting for someone on it.
Duplicates. One requisition advertised in several cities returns several records. Company, normalized title and content length together decide whether two records are the same posting. The rule is shared across all seven studies, so the sample sizes stay comparable.
What a percentage means. Document frequency: the share of postings that mention the term at least once, not how often it is repeated. Each role has its own term list, and terms outside that list were not counted.
Margins of error. They follow the sample. At 1,360 postings (software engineer) no figure moves more than about ±0.3 points; at 123 (accountant) the worst case is about ±0.9. Read close rows as ties rather than as a ranking.
It is one ATS vendor's public boards. Mostly US technology companies, mostly venture-funded. Banks, hospitals, government and agencies post elsewhere, and they ask for different things.
Absence is not evidence. A term missing from a role's list was never counted, so nothing here says it does not matter.
It is a snapshot, not a trend. Every study is one day of open roles, between 2026-08-29 and 2026-09-09. Comparing two of these studies across roles is fair; reading them as change over time is not.
And an aggregate is not your application. One posting weights its own terms differently from all 3,518. The study tells you which words the market uses; only the posting in front of you says which ones you need.
Every term in all seven studies, one row each — 289 rows. Role, term, category, share of postings, how many postings named it, the sample size, the margin of error and the collection date. The same numbers the tables above render.
One column has gaps. The margin of error is empty for the software engineer rows, which were collected before the analyzer recorded a margin per term. The worst-case figure for that sample size is in the method above, and it still applies to them.
Download the CSV →Free to use, including commercially, with attribution to ZoeVera and a link to this page.
There is one external source. Everything else on this page is computed from the postings it returned.
Each study is a single day's snapshot of open roles. When a role is re-collected the figures here, on its own page and in the CSV change together, and the new date is added to this list.
Paste your resume and one job posting. The score is that posting's terms measured against your resume, not this aggregate.
Check My Resume →Or read the keyword guides by role and keywords by industry.