The short version
- Ranking is now real. Greenhouse Talent Matching sorts applicants into Strong, Good, Partial, Limited or Needs manual review. Lever markets a percentage match with fit tiers.
- Auto-rejection by ranking is not. Greenhouse states its matching does not auto-reject or auto-advance any candidate. The ranking itself dispositions nobody.
- A Greenhouse scorecard is post-interview feedback typed by a human. It has nothing to do with your resume, despite what most articles on this subject claim.
- Greenhouse's recruiter filters have no skills field and no years-of-experience field. Skills are reachable only through keyword search over your resume text.
- Stuffing still fails: Greenhouse says multiple terms map to the same skill, so a longer list of matched terms does not always raise the score.
Greenhouse and Lever are the applicant tracking systems most technology and startup roles run on, which makes them worth understanding specifically rather than as a generic ATS. Both changed materially in the last year, and most advice about them has not caught up.
Two claims dominate the internet on this subject. The first is that these systems score your resume against a scorecard. The second, usually written by people correcting the first, is that they do not score anything and merely file your details for humans to search. As of 2026 both are wrong, and the accurate picture is more useful than either.
What a Greenhouse scorecard actually is
A scorecard in Greenhouse is a feedback form an interviewer fills in. It lives inside the interview kit, alongside the interview instructions, the questions and the candidate's resume, and Greenhouse describes it as the place to record feedback during and after the interview.
The rating scale is a set of red, yellow and green icons that a person clicks against each attribute the company decided to assess. There is no automation in it and no resume input to it.
The logic is simple once stated: a scorecard cannot exist before an interview happens. If you are optimising a resume against "Greenhouse scorecards", you are optimising against a form that only gets created after someone has already decided to talk to you.
“A scorecard is what an interviewer writes about you afterwards. It is not a gate your resume passes through.”
How Greenhouse actually finds you
This is the part worth internalising, because it is specific and almost nobody covers it. Greenhouse's candidate and prospect filters cover application type, job and stage, prospect pool and last touchpoint, custom fields, tags, status and rejection reason, source, recruiter and coordinator, pipeline tasks, and education fields such as school, degree and discipline.
Read that list again for what is missing. There is no skills filter. There is no years-of-experience filter. Those are the two things candidates assume they are being filtered on, and in Greenhouse they are not filterable fields at all.
Skills become reachable a different way: through free-text search over your resume. Talent Filtering lets a recruiter search keywords in applications and then filter the results, and Talent Rediscovery does the same across past applicants and prospects for current roles. Both return a list, not a ranking.
That gives keyword presence a concrete, mechanical reason to matter, and a narrow one. Your skills are findable only as text a recruiter's query can hit, which is also why they have to survive parsing in the first place. Layout decides what gets extracted, and Greenhouse lists columned layouts, complex tables and unclear section headings among its causes of a failed parse.
Ranking is real, the decision is still human
Greenhouse now runs Talent Matching, which sorts applicants into Strong, Good, Partial, Limited or Needs manual review against calibration criteria covering required skills, relevant experience, job titles and optionally industry. Lever markets a comparable feature, showing a percentage match with Strong Fit, Good Fit and Review tiers on its AI-powered screening page.
So the flat "no ATS ranks you" correction is now out of date, and not only for these two vendors: Workday grades applicants to prioritise them as well. What has not changed is the decision. Greenhouse's FAQ states that Talent Matching is assistive AI rather than automated decision-making, that it does not auto-reject or auto-advance any candidate, that it dispositions nobody, and that results are presented to recruiters and hiring managers who decide how to advance, reject or follow up. The automated rejection Greenhouse does document is a separate feature, knockout rules on application questions, which act on your form answers rather than your resume.
Lever's own data model says the same thing structurally. In its developer documentation, a candidate opportunity carries a stage, tags, a source and archive reasons. There is no candidate-level rank field. Score and scorecard appear only as field types inside interview feedback forms, exactly as in Greenhouse.
There is also a reason both vendors word this so carefully. New York City's Local Law 144 regulates automated employment decision tools, defined as computational processes used to substantially assist or replace discretionary decision making in employment, and triggers an independent bias audit requirement. Keeping a human as the decision maker is partly a compliance posture, not only a design choice.
| What people assume | What the documentation shows | What it means for you |
|---|---|---|
| Scorecards rank your resume | Scorecards are post-interview feedback typed by a human | Nothing to optimise there |
| Recruiters filter by skills and years | Neither is a filter field in Greenhouse | Skills are found by keyword search over resume text |
| The system rejects low matches | Greenhouse states Talent Matching never auto-rejects or auto-advances | A weak match costs you attention, not existence |
| More keyword repetition raises the match | Multiple terms map to one calibrated skill | Say it once, next to the evidence |
| Ranking is a myth | Both vendors now ship match tiers | Rank affects reading order, not survival |
Check how your resume parses before you apply
Optimize your application materials in under 2 minutes.
Tools that simulate extraction show which fields came through cleanly and which requirements are evidenced, which is the part these systems actually consume.
What this changes about your resume
- Make the text extractable first, and confirm it with an ATS resume checker. A skill the parser never pulled out cannot be found by any search, ranked by any matcher, or read by any recruiter.
- Use the posting's own phrasing for genuinely standard skills, once each. Repetition is not the input to the match.
- Put skills inside bullets that show the work, since keyword placement beats keyword count.
- Keep the file clean and conventional, because file problems cause parse failures more often than format choice does.
- Write bullets a human can scan in seconds, using the action, context, result shape. Rank decides reading order, and a human still does the reading.
What nobody can tell you yet
Two honest gaps are worth stating, because most writing on this subject implies more certainty than exists.
First, no independent party has audited how well either matcher works. Everything published about Talent Matching traces to one Greenhouse FAQ, and everything about Lever's fit scores traces to Lever's own marketing, where the percentages shown are illustrative interface mockups rather than measured output. The vendor claims have not been tested in public.
Second, adoption is not universal, and it may be falling. The 2025 Recruiter Nation Report from Employ, which surveyed more than 1,200 US recruiters and HR managers through Zogby Analytics, found AI used for candidate matching dropped to 40% from 55% the year before, while human oversight ranked as the second most important criterion when buying AI tooling. Note that Employ owns Lever, so this is a vendor reporting on its own category.
The picture that leaves you with is less dramatic than the folklore and easier to act on. Something may sort you into a bucket. A person still decides, under more volume than they can comfortably read, which is why clean extraction and legible evidence do more for you than any attempt to game a score.