The AI Match Input Reference, One Row per Resume Field the Screener Reads

This reference is for anyone sending tailored resumes to dozens of postings that carry a notice saying an AI tool will compare the application to the job requirements. It lists each resume input that matching layers say in their own documentation they read, what that input proves to the matcher, the specific way it gets misread, and the one edit that fixes it before you submit. Keep it open next to the posting and fix the two or three rows that turn a requirement from missing to met.

What does an AI screening notice on a job application mean?

An AI screening notice means a matching layer inside the employer's applicant tracking system (the software that stores and routes applications) will compare your resume against criteria a recruiter set for the role. It then attaches a grade or match label before a human looks. Two such layers publish enough documentation to work from: Greenhouse Talent Matching and Workday HiredScore.

What each one says about itself:

• Greenhouse Talent Matching. A live opt-out page states that "Talent Matching collects and processes resume text" and that it "matches applications against skills and experience we have decided are important for the role." Greenhouse's product page says "Recruiters define and weigh the job criteria." The candidate FAQ says that when you apply to a job using it, "your application is automatically eligible for AI-assisted review." Greenhouse says the tool is subject to monthly third-party bias audits by Warden AI.
• Workday HiredScore. Grades are based on "how well their CV matches the skills, education, and experience specified in the Job Description of the req." The grading view is called Spotlight. A later LLM-driven release called Fit & Gap scores each qualification separately against the parsed resume. An LLM is a large language model, the kind of AI that reads and writes text.

How to tell which matcher you face

If the notice says "Greenhouse's AI-powered Talent Matching," it is Greenhouse. Workday postings rarely name HiredScore, so a Workday-hosted application with a generic AI notice is a likely HiredScore case, but you cannot be sure. Write "unknown" and work the rows that apply to both.

Market and sector shift the odds. In Refolk's index of professional profiles, I counted recruiters and talent acquisition staff who list each system as a skill:

The samples lean toward different sectors. The Greenhouse-skilled group includes people at Meta, Palo Alto Networks and Paraform. The Workday-skilled group includes people at Boeing, Lifepoint Health and C&S Wholesale Grocers. My reading is that volume applicants to tech companies are more likely to meet Greenhouse's resume-only matching, and applicants to healthcare and industrial employers are more likely to meet Workday. That is an inference, not a measured fact. The skill tag is self-reported, and many people list Workday because they use its HR system rather than its recruiting module.

How does AI resume screening score your resume?

On HiredScore, the score is a letter grade from A to D, and the top two grades depend on meeting every basic (required) qualification, not on a percentage. On Greenhouse, it is a match against skills, experience, job titles and an optional industry that recruiters choose and weigh during calibration (the setup step where a recruiter tells the tool what matters for the role).

HiredScore's Candidate Grades reference defines the grades this way:

Three things follow from that table.

1. The basic qualifications are a gate. Meet all of them and you are at B or better. Miss one and you are capped at C, whatever your preferred coverage.
2. The A versus B split is partly hidden. A requires "most preferred qualifications" and "an average qualification match above a defined threshold." The threshold value is not published.
3. The LLM release keeps the same shape. Under Fit & Gap, "final grade assignment further differentiates A from B and C from D based on preferred qualification match and basic qualification scores." The release notes say an organization can keep using "the existing Spotlight basic qualifications matching," which suggests the LLM version is not switched on by default. You cannot tell from outside which version a given employer runs.

The advice you will find most often, "aim for a 65 to 80 percent match," comes from resume scanner tools that publish no method. Nothing in HiredScore's own documentation uses a percentage target. That is the core of required vs preferred qualifications in a resume match: required qualifications decide your band, and preferred ones only matter once every required one is covered.

On Greenhouse, the criteria list is shorter and the weighting belongs to the recruiter. The Talent Matching FAQ says recruiters "can select skills, experience, job titles, and (optionally) industry that are relevant to the role," and that "the system blocks attempts to add a biased or protected attribute, like gender." Education is not on that documented list.

The input reference: one row per resume field the matcher reads

Each row below covers one input that a matching layer documents that it reads, or one that applicants assume it reads. Each row gives what the input proves, how it gets misread, and the one edit to make. Jump to the row you need.

Skills

• Documented by: Greenhouse calibration (skills), HiredScore grading (skills from the job description).
• What it proves: That a named capability in the posting appears in your resume. At the parser level (the software that turns your file into structured fields), Textkernel claims taxonomies covering "300,000+ synonyms, 12,000+ unique skills," so a common synonym will often map to the posting's term.
• How it misreads: A skill that sits only in a skills list, with no work behind it. Greenhouse "warns the user if a skill can't be parsed from the candidate's resume." Whether an LLM judging a qualification marks a bare list entry as unmet is not established. Fit & Gap explanations quote evidence, though, which suggests a bare list gives it little to quote.
• The edit: Give every skill that maps to a basic qualification at least one dated bullet that shows you using it.

Job titles

• Documented by: Greenhouse calibration (job titles). HiredScore's experience matching draws on them.
• What it proves: That you have held the role, or a close neighbour of it.
• How it misreads: Promotions at one employer, or two-column layouts, can cause the parser to merge separate roles into one entry or split one role into several. In a third-party test of 36 templates run through Textkernel, "5 templates had their three jobs merged into fewer entries or split into more." A merged promotion can hide your most senior title.
• The edit: Give each title its own line with its own start and end date, even under the same employer, and confirm the count in the parse preview.

Start and end dates

• Documented by: Textkernel's parser, which computes experience from them.
• What it proves: How long you held each role, and so how much experience you have in total.
• How it misreads: Year-only dates. How parsers treat a range with no month is undocumented. Secondary sources warn that you can lose up to 11 months per boundary, or the range may not parse at all.
• The edit: Write every date as month and year (MMM YYYY), at both ends of every role.

Derived years of experience

• Documented by: Textkernel's MonthsOfWorkExperience field. It is not established whether HiredScore or Greenhouse use this computed field or let an LLM judge "5+ years" directly.
• What it proves: That your dated history meets a stated minimum.
• How it misreads: Overlapping roles are merged, so a side job running alongside a full-time job adds zero months. Totals you state in your summary are ignored.
• The edit: Merge your own date ranges and make sure the dated roles alone add up to the claim. See the next section for the method.

Employer names and inferred industry

• Documented by: Greenhouse calibration (optional industry). HiredScore's Favorite/Restricted Organizations list "allows HiredScore to use these companies as insights about a candidate," which it describes as "useful for identifying candidates with relevant industry experience."
• What it proves: That you have worked in the industry, or at the kind of company, the role cares about.
• How it misreads: How either vendor infers industry from an employer name is not publicly established. A holding-company name, a rebrand, or an obscure private firm may not register as the relevant industry.
• The edit: Add a short descriptor after any ambiguous employer name, for example "Northfield Holdings (regional hospital operator)."

Education

• Documented by: HiredScore grading (education from the job description). Education is not among Greenhouse's documented calibration criteria.
• What it proves: That you meet a degree or field requirement in the posting.
• How it misreads: The specific failure modes are not documented by either vendor. On Greenhouse, do not expect education to carry a match by itself.
• The edit: Where a degree is a basic qualification, list the degree, field and institution as a plain entry in an education section, then check the parse preview shows it.

Application answers

• Documented by: Not documented as read. Greenhouse says "the matching algorithm is only using data coming from the resume" and "currently, the algorithm does not access internal Greenhouse data sources." I found no equivalent candidate-facing page for Workday.
• What it proves: To the matcher, on Greenhouse, nothing. To the human reviewer, possibly a lot.
• How it misreads: You prove a requirement in a screening question, assume it counted, and leave it out of the resume.
• The edit: Copy any fact that proves a basic qualification into the resume as well, ideally as a dated bullet.

The file itself

• Documented by: HiredScore's supported types, "DOC, DOCX, PDF, RTF, and TXT." Unsupported types include .ppt, .pptx, .xps, .pages, .xlsx, .jpg, .png and archives.
• What it proves: That anything above can be read at all.
• How it misreads: An unsupported file means no grade. Secondary sites citing Greenhouse support say it "cannot parse resumes larger than 2.5 MB." I could not confirm that in a primary document. On another system, SAP SuccessFactors, a filename with an extra dot such as Test.Name.pdf triggers a "file type is not supported" error.
• The edit: Send a text-selectable file in a supported format, under 2.5 MB, with one dot in the filename.

How are years of experience calculated from resume dates?

At the parser level, years of experience are computed from the start and end dates of your roles, with overlapping ranges merged, and never from what you state in your summary. Textkernel's definition: "the number of months of work experience as indicated by the range of start and end date values in the various jobs/positions in the resume. Overlapping date ranges are not double-counted. This value is NOT derived from text like 'I have 15 years of experience'."

Practical consequences:

• Concurrent roles inflate nothing. If you count 7 years by adding a full-time job and a side role that ran alongside it, the parser may credit 5. Merge the overlapping intervals yourself before you claim "X+ years."
• The summary line is decoration to the parser. "15 years of experience" in your summary does not reach the computed figure. The dated roles alone must add up to it.
• Management months are counted the same way, but only for positions the parser has classed as management.
• Part-time roles and year-only dates are not covered in any primary document I found. Treat both as risks: use month and year everywhere, and do not rely on a part-time role to reach a minimum unless its dates stand on their own.

Textkernel's release notes list a fix to "report correct MonthsOfWorkExperience when CV has overlapping date ranges for jobs." That tells you overlap handling has had bugs before. If an older parser version sits behind a portal, your computed figure may differ from the documented behaviour. The parse preview is the only local check.

What happens to an application that gets no grade?

An application the matcher cannot process gets no grade, not a low grade, and both documented vendors route it out of the sorted view rather than ranking it last. That is a different outcome from a D, and you should plan for it differently.

HiredScore states: "If a candidate submits their CV in an unsupported format, they will not be assigned a grade in Spotlight." It also says it "omits entry level jobs from Spotlight Grading. This is because there are typically fewer requirements." Its product summary extends that to campus and student requisitions. Greenhouse lists three cases, then "the candidate is labeled 'Needs manual review.'"

No grade is not a low grade. It is a missing row in the recruiter's sorted view.

Whether a manual-review label helps or hurts you is not publicly established. What is clear is that on a HiredScore entry-level or campus requisition, time spent tailoring for the grade is wasted, because no grade exists. Spend that time on the human reader instead.

The opt-out works the same way on Greenhouse. If you opt out, it "deletes any existing AI-generated match score and reasoning for that application and flags it for manual review." Opting out puts you in the same queue as a resume it could not read.

How do you tailor a resume for AI screening without keyword stuffing?

Tailor by mapping each basic qualification in the posting to one real piece of evidence in your resume, then fixing the dates, employer lines and file that decide whether that evidence is read. Do not add terms you cannot back with a dated role. The procedure below is the per-posting run. My desk estimate is roughly half an hour to forty minutes per posting, but that timing is unverified.

Step 4 is the one that changes outcomes, and the one that hurts most at volume, because each posting words its requirements differently. Refolk tailors your resume to each posting from your own history and scores how well you actually fit, which gets you the requirement-by-requirement map without rebuilding it by hand every time. Check its output against the posting's basic list the same way you would check your own.

On the order of the final checks, guides disagree. Some put the parse-preview check before any knockout questions, others after upload. Either works, as long as you look at the preview before you press submit.

Use this sheet for step 4. One row per basic qualification; a row with an empty evidence cell is a C-grade risk on HiredScore.

POSTING: [company] / [role] / matcher: [Greenhouse | HiredScore | unknown]

BASIC QUALIFICATION      | EVIDENCE TYPE (title / dates / bullet) | WHERE IN RESUME          | MET?
1. [requirement text]    | [type]                                 | [employer, role, line]   | [Y / N]
2. [requirement text]    | [type]                                 | [employer, role, line]   | [Y / N]
3. [requirement text]    | [type]                                 | [employer, role, line]   | [Y / N]

YEARS CHECK
Stated minimum: [N] years
Merged total from dated roles: [N] years [N] months
Overlaps removed: [list role pairs]

EMPLOYER LINES NEEDING A DESCRIPTOR
[Employer name] ([one-phrase industry descriptor])

PREFERRED QUALIFICATIONS COVERED: [x] of [y]  (A needs most)

How the match goes wrong: failure modes and false positives

The match goes wrong mostly where you think a requirement is met and the parser or matcher disagrees. The fix is almost always a check against the dated record, not more keywords. Each failure mode below gives the false positive you will believe and the check that catches it.

1. Overlapping roles. Parsers merge overlapping ranges, so a side job alongside a full-time job adds zero months.
   • False positive: you count 7 years by hand; the parser credits 5.
   • Check: merge the intervals yourself before claiming "X+ years."
2. Year-only dates. Handling of ranges with no month is undocumented. Secondary sources warn of up to 11 months lost per boundary, or a failed parse.
   • False positive: "2019 to 2022" looks like three years to you.
   • Check: MMM YYYY on every role, both ends.
3. Job segmentation. Promotions at one employer, or columns, can merge or split jobs. That happened to 5 of 36 templates in one third-party test.
   • False positive: you see three roles; the parser sees one, or five.
   • Check: compare the employer and date count in the parse preview against your resume.
4. Experience claimed only in the summary. Textkernel ignores text like "15 years of experience."
   • False positive: the claim is right there at the top.
   • Check: the dated roles alone must add up to the claim.
5. Skill only in a list. Greenhouse warns recruiters when a skill cannot be parsed. An LLM judging a qualification with no supporting bullet may mark it unmet. That is unverified, but Fit & Gap explanations quote evidence.
   • False positive: the keyword is on the page, so you assume it counts.
   • Check: one dated bullet for each basic-qualification skill.
6. Oversized file. A file can upload without error and never be parsed. Secondary sources put the Greenhouse limit at 2.5 MB.
   • False positive: the portal said "uploaded."
   • Check: file size under 2.5 MB, text selectable.
7. Entry-level and campus requisitions. On HiredScore no grade exists for these.
   • False positive: you spent the time tailoring for a grade.
   • Check: if the requisition is campus, student or entry level, write for the human reader instead.
8. Employer-to-industry mapping. It is undocumented. A holding-company name or a rebrand may not register as relevant industry.
   • False positive: everyone in your field knows that company.
   • Check: a short industry descriptor after the employer name.

That finding cuts both ways. It means the common scare claim is unsupported. It also means a stuffed resume might grade higher and then fail in front of a recruiter, which is a worse outcome at volume than a lower grade, because you burn the application. Do not hide text, and do not add skills you cannot point to in a dated role.

A ninth, quieter failure: treating an application answer as evidence. On Greenhouse, the resume is the only scored surface. A perfect answer to "Do you have 5 years of Python?" proves nothing to the matcher if the resume's dated roles do not show it.

Before you submit, and keeping this reference current

Before submitting, confirm the file, the basic-qualification map, the merged years and the employer lines, in that order. The checklist below is the gate. After it, re-check the vendor documents whenever your results stop matching what this reference predicts.

How to keep this current

Vendor behaviour changes through release notes, not announcements. Re-check these documents by name when you notice a shift:

• Workday HiredScore Candidate Grades reference. Watch for any published threshold for the A grade, or a change to the basic-qualification gate.
• HiredScore release notes for Fit & Gap. Watch for whether LLM scoring becomes the default, and whether it starts reading more than the parsed resume.
• Greenhouse Talent Matching FAQ and Data Processing FAQ. Watch the line that says the algorithm uses resume data only. If application answers are ever added, the application-answers row above flips.
• MyGreenhouse FAQ for Candidates. Watch the opt-out consequence.
• Textkernel Tx Parse API docs. Watch the MonthsOfWorkExperience definition and any note on part-time or year-only dates, both currently undocumented.
• Workday Responsible AI summary for Spotlight. Watch for any candidate-facing disclosure of inputs.

At volume, the expensive part is not knowing the rows. It is rerunning the map for every posting without letting the resume drift into claims you cannot defend. Refolk keeps every tailored version grounded in your own history and shows the fit score per posting, so you can see which basic qualifications are still bare before you send. Whatever tool you use, the test stays the same: every requirement you claim to meet points to one dated line in the resume.