Resume keywords in 2026: what ATS matching really means
A keyword only counts where the software stores it. Here is what Greenhouse, Workday and Ashby do with the words on your resume, and which ones to change.
Jordan Reyes · · 10 min read
A keyword only counts where the software keeps it. Your resume is turned into a short list of fields — skills, job titles, employment dates, company names — and a term that lands inside one of those fields can be searched, compared and scored. The same term in the fourth bullet of a job you left in 2021 usually cannot.
That one distinction settles most of the arguments people have about keywords. It explains why mirroring a job posting's phrasing does close to nothing, why spelling a tool name correctly does a great deal, and why two candidates with identical experience can sit in different halves of a recruiter's screen.
What "matching" means in the systems you are actually applying through
Three products cover most of the forms you will fill in this month, and they do three different things with your words.
Greenhouse's Talent Matching asks the recruiter to set criteria first — "required skills, relevant experience, job titles, and (optionally) industry," along with "the importance assigned to each criteria." Then it compares your resume against those. The comparison is not a string match. Greenhouse says it "uses embedding representation of skills and job titles so that we can run a semantic search on them," and gives the worked example: a search for "software developer" returns people who wrote "software engineer" and people who wrote "web developer." For skills specifically, it "compares related terms on each resume to the calibrated skills you set, and multiple terms can map to the same calibrated skill." The output is a bucket: Strong, Good, Partial, Limited, or Needs manual review.
Knowing exactly what goes into that comparison matters, because the list is short. Greenhouse's data processing documentation says Talent Matching extracts "skills extracted from resumes, years of experience, job titles, start/end dates of employment, company names from employment history, and where available, a derived industry classification." Six things. A Digital Marketing Specialist whose resume runs to two pages of campaign results is being matched on six extracted facts, and the campaign results are not among them — they are in the attached file, waiting for a person to open it.
Workday does something different again. It suggests skills for your application by reading the file, and its documentation is unusually specific that this is not confined to one section — the machine learning "uses logic to consider matches from the entire uploaded resume, not just the Skills section," naming twelve places it looks: Certificates, Company, Degree, Description, Education, Experience, Honors, Major and minor, Objective, Publication, Skills, Summary. Those suggestions land on your application as editable chips. You can keep them or take them off before you submit, and most applicants never notice the step.
Ashby is the literal one. Its candidate search includes a full text resume search filter, with boolean modifiers: quotation marks for an exact phrase, prefix matching where "script" matches javascript and typescript, "not" or "!" to exclude a term, and brackets to group sub-queries. No embeddings, no synonym cloud. If a recruiter searches "HubSpot" and your resume says "marketing automation platform," you are not in that result set.
Three ways an applicant tracking system reads a keyword
Same word, three mechanisms. Which one you are facing depends entirely on which form you filled in.
| Greenhouse Talent Matching | Workday Skills Cloud | Ashby candidate search | |
|---|---|---|---|
| How your words are compared | Semantic — embeddings of skills and job titles | Suggested skills mapped to canonical skills and synonyms | Literal full text search of the resume |
| Synonyms handled for you | Yes — "web developer" matches "software developer" | Yes — the skill string you typed is retained alongside the canonical one | No — prefix matching only, e.g. "script" finds "javascript" |
| Where it reads from | Skills, titles, dates, companies, derived industry | The entire uploaded resume, not just the Skills section | The resume text as filed |
| What comes out | Strong, Good, Partial, Limited, or Needs manual review | Skills on your application you can keep or remove | A list of candidates, ranked by the recruiter's query |
| Can it reject you | No — "does not auto-reject or auto-advance any candidate" | No — it populates data, a human reviews | No — you are simply absent from the results |
Where the word has to sit
Take a Digital Marketing Specialist with eight years of paid search and analytics work. Everything below is on her resume already. The column that matters is the second one.
| Where the term appears | What the software does with it | Example |
|---|---|---|
| Job title line | Extracted as a title; matched semantically in Greenhouse | "Digital Marketing Specialist" beats "Growth Wizard" |
| Skills line | Extracted as a skill; the highest-value real estate on the page | "Google Analytics, Tableau, HubSpot" |
| Company name | Extracted, and used to derive an industry classification | "Ridgeline Retail" reads as retail |
| Employment dates | Extracted, and turned into years of experience | "Mar 2018 – Present", not "'18–now" |
| Bullet text | Stored in the file; read by a person, not indexed as a field | "Cut cost per acquisition 34%" |
| Section heading | Tells the parser where a section starts | "Skills", not "What I Bring" |
The bullets are not wasted — they are the reason a recruiter says yes once they are looking at you. They are simply not what puts you in front of the recruiter. That is the whole trade, and most resume advice collapses the two.
Which keywords employers are actually typing
O*NET tracks the software named in employer job postings and publishes the counts. Its Hot Technologies list is drawn from "requirements frequently included across all employer job postings," and the pull on O*NET's Hot Technologies page as of September 2026 covers unique US postings between January 1 and December 31, 2025 — 46,885,153 postings in total, from Lightcast job postings data.
Read the gap. Excel appears in more than three million postings; Google Analytics, which a Digital Marketing Specialist would call the center of her job, appears in 62,279. That is not a ranking of importance. It is a map of how many different searches a term can put you into — Excel is a passport, Google Analytics is a key to one building.
What this data does not tell you is how often a recruiter typed any of these terms into a candidate search, or whether the people who listed them were interviewed. It counts what employers wrote in postings. Postings and searches are close cousins, not the same thing, and no public source publishes the second one.
Should you rewrite your resume for each posting?
Mostly no, and this is where I will take a position that a lot of resume advice would argue with: rewriting your bullets to echo a job description's phrasing is close to wasted effort in 2026 — copying its tool names is not.
The reason is in the mechanisms above. Greenhouse resolves phrasing for you through semantic search; Workday retains the skill string you typed while mapping it to a canonical skill; both are built precisely so that a candidate who writes "paid search" and a recruiter who calibrated "SEM" end up connected. Phrasing is the problem those systems solve. Proper nouns are the problem they cannot solve, because "HubSpot" has no synonym, and in Ashby's literal search it has no near-miss either. So the tailoring worth doing per application is narrow: check the posting for named tools, certifications and the exact job title, and make sure each one you genuinely have appears in a field, spelled the vendor's way.
Which spelling of a tool name should you use?
The one in the posting. If the posting does not name it, write the long form and the short form together, once. That is the whole answer, and the rest of this section is why it is not obvious.
The three systems disagree about how forgiving they are. Ashby compares strings, so "SQL" and "Structured Query Language" are two different terms and a recruiter who typed one will not see the other unless prefix matching happens to catch it. Greenhouse's embeddings will usually bridge the two. Workday keeps whatever you typed alongside the canonical skill it maps you to. Write for the strictest of the three and you have satisfied all of them, which is why the long-plus-short habit is worth four extra words on the page.
O*NET's own list is a useful hint here, because the names on it come out of the postings themselves rather than out of a style guide. It files the language as "Structured query language SQL" — abbreviation and expansion jammed into one entry, 758,250 postings — and files Amazon's cloud as "Amazon Web Services AWS software." That is not O*NET being clumsy. It is what employers write when they want both spellings found.
The Digital Marketing Specialist has this problem in miniature. She writes "GA4," because that is what it is called inside her team. The posting says "Google Analytics." She is returned for neither in a literal search unless she writes both, and "GA4" on its own is a term that only reads as a credential to somebody already sitting in the job. So: "Google Analytics (GA4)." Also "Search Engine Marketing (SEM)," also "Structured Query Language (SQL)." A human reads the parenthesis and skips over it, which costs nothing. A string comparison has no idea the two are the same thing, which costs you the search.
The version that gets found
Here is what the recruiter sees. A Digital Marketing Specialist, before:
Title: Growth Marketing Wizard Skills: (none extracted) Experience: 2018 – now, Ridgeline Retail
And after, with the same eight years and the same work behind it:
Title: Digital Marketing Specialist (Paid Search & Analytics) Skills: Google Analytics, Google Ads, HubSpot, Tableau, SQL, WordPress Experience: Mar 2018 – Present, Ridgeline Retail
Nothing was invented. One line moved out of a bullet and into a skills line, one title was translated from the internal name to the market's name, and one date format was made unambiguous so the system could compute years of experience from it.
The second version takes about six minutes to produce and it is the only edit on this page that has to be made once rather than per application. Note what did not change: no bullet was rewritten, no phrase was borrowed from a posting, and the two pages of campaign results underneath are word for word the ones she had before. The candidate is identical. The record is not.
Two words about job titles
Your employer's title for you is a fact about your employer. The title the market posts is what gets searched — and O*NET's entry for Search Marketing Strategists, 13-1161.01, lists twenty detailed work activities that would each be recognizable to a Digital Marketing Specialist, from "implement advertising or marketing initiatives" to "analyze website or related online data to track trends or usage." Use both: the market title as the heading, the internal one in parentheses if the company is known for it.
The skills no keyword line can carry
Look at what O*NET actually lists as the skills for Search Marketing Strategists, the occupation code a Digital Marketing Specialist sits closest to. Not the software — the skills. Seven of them O*NET files as Essential Skills: Reading Comprehension, Active Learning, Active Listening, Critical Thinking, Speaking, Writing and Monitoring. Eleven more sit underneath as Transferable Skills, among them Complex Problem Solving, Judgment and Decision Making, Negotiation, Time Management and Social Perceptiveness. Eighteen skills, every one of them real, every one of them what the job is actually made of — and not one of them belongs on a keyword line.
There is no version of "Active Listening" that helps a recruiter's search return you, because nobody searches for it and no calibration is built on it. Those skills are what a bullet is for: the sentence about the quarter you spent talking the merchandising team out of a campaign that would not have worked demonstrates critical thinking, and demonstrating is the only way that particular claim ever lands. Keep the extracted fields for tools and titles. Keep the prose for the things a person has to believe.
What the match category is not
It is worth being blunt here, because "the ATS filtered me out" has become the standard explanation for silence and it is usually wrong. Greenhouse's own FAQ says Talent Matching "does not auto-reject or auto-advance any candidate, including candidates in lower match categories," and that results are "presented to recruiters and hiring managers in their dashboard for further review, so humans decide how to advance, reject, or follow up."
So a Limited Match is not a door closing. It is a worse place in a queue, on a screen belonging to somebody who has a lot of these to get through — and the field is not small. The Bureau of Labor Statistics counts 952,700 market research analysts and marketing specialists in 2025, with a median wage of $78,760 a year as of May 2025 and 7% growth projected from 2025 to 2035. A Digital Marketing Specialist is competing inside a category that size. Sorting order matters more when the pile is deep.
What to do this week
Do three things, in this order, and stop there.
- Move your tools into a skills line. Open your resume and find every product name currently living inside a bullet — Google Analytics, HubSpot, Tableau, SQL. Put each one on a plain, comma-separated skills line near the top, spelled the way its vendor spells it. That single edit changes what is extractable more than a full rewrite would.
- Fix your title and your dates. Replace an internal title with the one employers post, and write every date as month and year, with "Present" for the current role. Greenhouse derives years of experience from those start and end dates; an unreadable format is experience you do not get counted for.
- Search your own resume the way Ashby would. Take the three tool names from a posting you want, and check that each appears literally in your file. If one is only implied, add it or drop the ambition — do not write a skill you cannot discuss for ten minutes.
Then send the file. The keyword work is finished once those fields are clean, and re-editing the same document for the eleventh time is not the reason anyone gets hired.
If the retyping is what is slowing you down, HeroApply fills each application from that resume — the same title, skills and date fields this post is about, entered the same way every time.
Sources
- Greenhouse Support — Talent Matching FAQ
- Greenhouse Support — Talent Matching data processing FAQ
- Workday Administrator Guide — Setup considerations for skills in recruiting
- Ashby Knowledge Base — Candidate Search
- O*NET OnLine — Employer-Based Hot Technologies, job postings January–December 2025
- O*NET OnLine Help — Hot Technologies
- O*NET OnLine — Search Marketing Strategists, 13-1161.01
- BLS Occupational Outlook Handbook — Market Research Analysts and Marketing Specialists
Frequently asked questions
Do I need to match the job description word for word?
No, and for most phrases it does nothing. Greenhouse says it "uses embedding representation of skills and job titles so that we can run a semantic search on them," which is why "software engineer" and "web developer" both come back for a search on "software developer." The exception is product names — a tool is spelled one way, and no synonym model will guess that you meant it.
Where on my resume should keywords go?
In the fields the software extracts as data — your job titles, your employment dates, your company names, and a plain skills line. Greenhouse's Talent Matching pulls "skills extracted from resumes, years of experience, job titles, start/end dates of employment, company names from employment history." A tool named only in the fourth bullet of a role from 2021 is in the file, not in the index.
Will an ATS reject my resume for missing a keyword?
Not in Greenhouse, which states plainly that "Talent Matching does not auto-reject or auto-advance any candidate, including candidates in lower match categories." What a missing keyword changes is your position in the list a recruiter is scrolling. Being sorted into Limited Match is not a rejection, but it is a worse queue.
Does Workday only read my skills section?
No. Workday's documentation says its machine learning "uses logic to consider matches from the entire uploaded resume, not just the Skills section," and names twelve places it looks, including Summary, Experience, Description, Publication and Honors. It then suggests skills on your application, which you can keep or remove before submitting.
Is a skills list still worth putting on a resume?
Yes, because it is the cheapest way to get a term into an extracted field. Keep it to tools and methods you could be questioned on in an interview, spelled the way the vendor spells them. Padding it with soft skills adds nothing a recruiter searches for.
Let HeroApply do the applying
It finds the jobs that match your resume, answers the screening questions in your voice, and submits the application — desktop, phone or cloud.
Jordan Reyes · Careers Editor
Jordan writes the resume, cover letter and interview guides — the parts of the search that happen before anyone presses Apply. Every example comes from a real job title.
The best of the blog, once a week
One email with the week’s guides. No spam, unsubscribe any time.
