Greenhouse Real Talent: The 5 Buckets (and the One to Avoid)

You hit Submit on a Greenhouse job. The page says thanks and shows you a status of New. Then nothing for two weeks. What actually happened on the recruiter's side in 2026 is more specific than the "black hole" story you have heard. Your application landed in a tiered inbox, was scored against a recruiter-built calibration, and was placed into one of five buckets that decide which folder the recruiter opens first.

This post walks through the recruiter UI as it works today, names the five buckets and what gets you into each, and gives you the seven specific moves that lift a typical resume from Partial or Limited into Strong. Everything below is based on Greenhouse's own product documentation and support center as of May 2026.

What changed about Greenhouse in 2025 and 2026

Greenhouse runs the application pipeline for about 11,500 verified companies in 2026, and it has been the highest- ranked applicant tracking system on G2's grid reports through the year. For most of its history, Greenhouse was the "human reviewer" ATS. Recruiters scored you against a scorecard they wrote. There was no parser ranking you behind the scenes. That changed in June 2025 with the launch of Greenhouse Real Talent.

Real Talent is a bundle of three things, not one:

• AI-assisted Talent Matching: an AI ranks every candidate against a recruiter-defined calibration (the weighted criteria) and assigns a match score plus a bucket label.
• Fraud and spam detection: a system layer that flags bot submissions, mass-application patterns, and known abusive IP addresses or domains.
• Identity verification through CLEAR: an optional flow where the candidate confirms identity in MyGreenhouse with a selfie and a government ID, processed by CLEAR rather than the employer.

Real Talent does not auto-reject and does not auto-advance. Greenhouse is explicit about this in the support docs: all disposition and hiring decisions are made by the hiring team. What the AI does is order the inbox. A recruiter looking at 800 applicants opens the Strong folder first and might never open Limited. That ordering is what decides whether a human sees your resume in week one or week six.

Real Talent does not auto-reject. It orders the queue. That ordering is what decides if a human reads your resume in week one or week six.

The five buckets, in order

Every application that enters a Greenhouse pipeline with Talent Matching enabled gets sorted into one of five buckets. The buckets are Strong, Good, Partial, Limited, and Needs manual review. Inside each bucket, candidates are ordered by application date so newer applications appear at the top.

Strong

The top tier. Your resume matched most of the calibration's weighted skills and the job-description language. A recruiter screening a busy role might spend 80 percent of their first session here and never scroll past it. If the role has a screening call queue, Strong candidates fill it first. In practice, this is the bucket that gets the first-round phone screen within the first two weeks of applying.

Good

The second tier. You hit some of the weighted criteria clearly but missed others, or you hit all of them at slightly lower confidence. Recruiters work through Good after Strong is exhausted. On a role with 80 to 200 applicants, Good often gets the second batch of phone screens. On a role with 2,000 applicants, Good may never get opened.

Partial

The middle. The AI found some signal but not enough. This is the most common landing spot for a generic resume that was not tailored. Recruiters dip into Partial when the Strong and Good folders dry up, which on competitive roles never happens. Most Partial candidates get the automated rejection email two weeks after the role closes.

Limited

The bucket to actually fear. Your resume did not match the calibration well, and on competitive roles the recruiter never opens this folder. Being in Limited is not the same as being unqualified for the job. It is the AI's read based on the parsed text it had access to and the skills the recruiter weighted. A resume in a heavy two-column template, a resume that uses long-form prose instead of a Skills section, and a resume that uses different terms than the job description all land here often.

Needs manual review

The label that sounds the most alarming and is usually the least bad. A candidate ends up here for two reasons:

• The candidate opted out of AI processing on the application (Greenhouse exposes an opt-out where the customer has enabled it, often in jurisdictions with AI disclosure rules).
• The resume could not be parsed cleanly (a corrupted file, an unsupported format, a heavily styled document the parser could not extract text from).

Either way, the AI is removed from the loop and a human is required to review the application before it moves forward. On large pipelines this can mean a longer wait, but it also means a human reads every word. On smaller pipelines, candidates in Needs manual review get treated the same as the rest of the queue.

How the AI actually scores you

Before any candidate is scored, the recruiter or hiring manager builds a calibration. A calibration is a list of skills, qualifications, and weights specific to the role. The recruiter starts with the job description, often pastes in the internal scorecard attributes, and clicks Generate calibration. Greenhouse suggests a list of skills. The recruiter accepts some, edits others, adds skills the AI missed, and rates each one's importance. The importance rating is the weight.

Once the calibration is applied, every existing application is scored, and every new application is scored on submit. Each candidate gets a match score and a bucket label.

You do not need to keyword-stuff because the system already considers synonyms. You do need to mention each calibrated skill at least once, somewhere parseable. If the job description names "Snowflake" and your resume only says "cloud data warehouse," the embedding is close but not as close as a literal match would be on a less generous parser. Use the exact tool name and the generic phrase, once each.

What "Needs manual review" actually means for you

This is the most misunderstood bucket on the SERP. Articles treat it as the worst outcome. It is not. Three things to know about Needs manual review:

1. It bypasses the AI entirely. A human must read your application before it advances. There is no algorithmic decision in your path.
2. It is the destination if you opt out. Where the employer has enabled candidate AI opt-out, you can choose manual review during the application. If you choose that, Greenhouse deletes any existing match score and reasoning for your application.
3. The bucket to actually fear is Limited. Limited means the AI scored you and put you near the bottom. Needs manual review means the AI did not score you at all. On most pipelines a human reviewer is more forgiving than an algorithm tuned to a calibration you could not see.

That said, do not aim for Needs manual review as a strategy. It usually slows down your application by a few days because the bucket is processed separately. The play is to write a resume the AI can read clearly and to land in Strong.

Identity verification through CLEAR

Identity verification through CLEAR is the third leg of Real Talent and the most visible to candidates because it requires action from you. Here is how it works in practice.

If the employer has enabled identity verification on the role, you will see a prompt in MyGreenhouse during or shortly after the application. The first time, you confirm your identity by uploading a government ID and taking a selfie inside CLEAR. CLEAR handles the verification using biometric signals, document checks, and cross-referenced trusted data sources. After the first verification, any future application that asks for it can be completed with just a selfie.

Two privacy details that matter to candidates:

• Recruiters and hiring managers never see your ID or your selfie. CLEAR keeps those. Only a pass-or-fail result, your name, email, phone, and IP address are shared with Greenhouse.
• You can decline. Declining verification is not treated as a failed check. It is recorded as declined, and the employer chooses how to handle that (most ask for verification later in the process rather than blocking the application).

The reason this feature exists at all is volume. Greenhouse cites recruiter surveys that show 90 percent of teams now report a spike in spam applications, and on some roles up to 23 percent of applications show evidence of being machine-generated. A 2025 Resume.org survey found 44 percent of applicants would use AI to misrepresent their resumes if it landed them an interview. CLEAR plus the fraud detection layer is the response.

How to land in Strong instead of Limited

Seven moves that consistently move a resume up at least one bucket on Greenhouse pipelines. None of these are keyword-stuffing. They are about giving the parser unambiguous signal and giving the calibration a clean read.

1. Use a real, named Skills section. Greenhouse leans on the Skills section heavily for the calibration match. Put your tools, languages, and named frameworks in a labeled Skills block, comma-separated, in plain text. Group by category if you want (Languages, Cloud, Data, ML) but keep the section findable.
2. Mirror the nouns from the job description. The calibration was built from the job description and the internal scorecard. The exact nouns from the JD are almost certainly the weighted skills. Use the literal tool names ("Snowflake" not "cloud data warehouse") at least once in your Skills section and once in a bullet where you actually used the tool.
3. Match the job title language in your most recent role.Embedding-based matching helps but does not solve everything. A resume that says "Software Engineer III" matches a calibration weighted on "Senior Software Engineer" better when the title on your resume actually contains the right level word.
4. Single column, standard section headings. Sidebars and two-column templates break parsing on most ATSs, Greenhouse included. Use plain section headings: Experience, Education, Skills, Projects, Certifications. Anything more decorative is a coin flip on the parser.
5. Save as a text-selectable PDF. Open the file after saving and try to select a paragraph with your cursor. If it selects as a single shape (image), the file is not parseable. Re-export from your source document.
6. Lead each bullet with a verb that owned the work. The calibration also reads experience bullets, not just the Skills section. Bullets that start with concrete verbs (Built, Shipped, Closed, Reduced) carry more signal than bullets that start with vague ones (Responsible for, Helped with, Worked on). For the full list, our bullet-point examples post has 80 plus role-specific bullets to model on.
7. Mention each calibrated skill once, not five times.Because of the embedding-based dedupe, listing "Python, Python 3, Py, Python programming" on the same line wastes characters. One literal mention plus one instance in a bullet where you used the skill is the cleanest signal.

Five mistakes that drop you a bucket

Common patterns we see drop otherwise qualified candidates from Good or Strong into Partial or Limited.

• Resume in a two-column template with the Skills section in the sidebar. The parser often reads the sidebar as a separate document or merges it with the wrong block. Move skills inline.
• Skills written as full sentences instead of a list. "Experienced in building data pipelines using Snowflake and dbt" is harder for the parser to map to a calibrated skill than a clean line that says "Snowflake, dbt, Airflow, Python."
• A custom header that hides your contact info in an image or a graphic. Greenhouse parses structured fields including name, email, phone, and links. If those are inside an image, they are invisible.
• A role title rewritten to be aspirational. If your title was "Customer Support Lead" but your resume says "Customer Success Architect," the calibration may not match cleanly. Use your real title; describe the elevated work in the bullets.
• A summary block that buries the role being targeted in marketing language.The summary is parsed and weighted on Greenhouse. A direct line like "Senior Data Engineer with eight years across Snowflake, dbt, and Airflow" is read better than a four-sentence narrative paragraph.

Is the AI biased against me?

Reasonable question, especially if you are reading this after a long stretch of silence on applications.

Greenhouse publishes ongoing bias audit results in partnership with Warden AI, an independent firm that evaluates AI hiring tools. The audits run monthly on Talent Matching and measure outcomes across 10 demographic categories. The methodology aligns with three regulations specifically:

• NYC Local Law 144: requires bias audits on automated employment decision tools, public summary of results, and notice to candidates that an AEDT is in use.
• Colorado SB 205: the state's AI consumer protection law, with employer-specific obligations on automated decision systems used in employment.
• California FEHA: California's Fair Employment and Housing Act, which has been read to apply to automated tools used in employment decisions.

Greenhouse also commits to retraining and retesting any model that fails a fairness check before release. That is a stricter operational floor than most of the ATSs we have documented. It does not mean bias is gone. It means the team that ships Talent Matching is on the hook for it, publicly, every month. You can read the live dashboard before you apply if you want to see the current numbers.

FAQ

Does Greenhouse auto-reject my application?

No. Greenhouse Talent Matching, the AI layer inside Real Talent, sorts and ranks candidates but does not auto-reject or auto-advance them. Every rejection is a manual call by a recruiter or hiring manager. Auto-rejection on Greenhouse is only possible through application rules tied to custom screening questions (for example, "Are you legally authorized to work in the country of the role"), and even those are configured manually by the customer.

Can I see what bucket I am in?

No. The bucket label is recruiter-side. Candidates see only the application status (New, In Review, Phone Screen, and so on). You can infer your bucket loosely from how fast you hear back: Strong applicants tend to hear within two weeks, Partial and Limited tend to hear at the very end of the role or not at all.

Should I opt out of AI processing?

Usually no. Opting out sends your application into Needs manual review, which often means a slightly slower review and removes you from the ranked queue the recruiter scans first. The exception is if you have a strong reason to believe the calibration is poorly built for the role (for example, you are an obvious fit but you know the JD is narrow) and you would rather rely on a human reading the resume in full.

Does identity verification with CLEAR cost me anything?

No, not for the candidate. The cost is on the employer. The verification is a one-time setup with your government ID and selfie, then a selfie for any future checks. You can decline if you prefer and the application is not automatically failed.

Is keyword stuffing still a good idea on Greenhouse?

No. Talent Matching uses semantic embedding-based scoring, not raw keyword counts. Stuffing the same skill in five synonyms gives you the same credit as one clean mention. The play is naming each calibrated skill once in a clear Skills section and once in a bullet where you used it, then spending the rest of your characters on quantified outcomes.

Read next

For the broader parser-by-parser context on how ATSs actually read resumes, the ATS resume guide has the structural rules that work on every modern system. For the score numbers you should target on tools like ours, our good ATS score guide covers the realistic benchmarks. For the role-specific keywords that show up most often in calibrations, the keywords-by-industry post has the lists sorted by sector. And for the Workday- specific version of this same question, the Workday resume format guide breaks down what changes when the ATS is Workday and not Greenhouse.

Key takeaways

• Real Talent (June 2025, phased Q3 2025) added an AI ranking layer to Greenhouse on top of the original human-scorecard system. The buckets are Strong, Good, Partial, Limited, Needs manual review.
• The AI ranks against a recruiter-built calibration of weighted skills. It does not auto-reject. It orders the inbox a recruiter scans top to bottom.
• Needs manual review is not a rejection. It is the bucket for opted-out or unparseable applications, and a human reads every word. Limited is the bucket to fear.
• Embedding-based scoring means synonyms count, but multiple synonyms for the same skill do not stack. Mention each calibrated skill once in a clean Skills section and once in a bullet.
• Single column, standard section headings, real Skills section, mirrored JD nouns, text-selectable PDF, ownership verbs in bullets. That is most of the lift.
• CLEAR identity verification is optional, processed outside the hiring team, and declining does not auto-fail the application.
• Warden AI publishes a monthly bias audit dashboard on Talent Matching, scoped to NYC Local Law 144, Colorado SB 205, and California FEHA. The numbers are public.

For what landing in Strong actually feels like for a real candidate (Liana R, comeback after four years out, made it through a Greenhouse pipeline on the third tailored version of her CV), the CVHive stories page collects five long-form interviews from people who worked the structure end to end.

Before you apply to another Greenhouse role, run your CV through the free CV score. It returns the parsed-field readout, the keyword gap against any pasted job description, and the specific lines that read as filler. It takes about 90 seconds and is the same check we run on the first pass of every Glow Up rewrite.