This guide supports policy alignment between Talent Acquisition (TA), Legal, and Security teams when using the Talent Matching feature in Greenhouse Recruiting’s Real Talent suite.
The goal is to define a clear approach for using assistive AI in Talent Matching while maintaining fairness, transparency, and compliance. By the end of these discussions, your organization should have clear answers to questions such as:
- What relevant AI laws apply to your roles, offices, or organization?
- How will candidates notify your team that they want to be opted out of AI processing of their resume?
- How will your internal processes ensure human oversight for all decisions made using Talent Matching?
- How are calibrations created at your organization?
- How will calibration history be reviewed and documented at your organization?
- Which users can view the results of calibrations? Can those same users create, update, or delete the calibration?
- Under what conditions may match scores be overridden, and how is this documented?
Note: This document is intended as a reference for Greenhouse customers and their usage of Real Talent features. None of this is intended to be legal advice, neither are any of the answers provided intended to override or contradict the advice of your legal counsel.
Understanding Talent Matching
Talent Matching is an AI-assisted feature designed to streamline hiring by matching candidates with a recruiter’s defined and weighted calibration.
Recruiters create a calibration, a structured list of job-related qualifications and weights, and apply it to a candidate pipeline. Each candidate receives a match score based on how closely their resume aligns with the recruiter’s calibration.
If a candidate cannot be assigned a match score (for example, if their resume cannot be read by the system or they have opted out of AI-assisted review) they are flagged as “Needs manual review” and highlighted in the user interface to be fully reviewed by a human.
For transparency, in addition to the match score, Talent Matching provides resume highlights and a breakdown of matched skills, experience, and criteria contributing to the score.
Talent Matching is designed as assistive AI, not automated decision‑making:
- It does not automatically reject or advance candidates.
- Recruiters and hiring managers remain responsible for all hiring decisions.
- Candidates who cannot be scored are marked as needing manual review.
Legal and security teams can review Greenhouse’s Talent Matching FAQ for answers to common questions.
Designing your Talent Matching process
Your policy should define how calibrations are created, how match scores inform actions, and what documentation and disclosures are required.
1. Understand which legal requirements may affect your organization’s use of AI tools
Some jurisdictions require specific rules related to disclosing usage of AI tools and the retention of data once AI tools are developed. Your legal team should decide on the exact language required in notifications for your jurisdictions, as well as whether and how long to retain the AI outputs and underlying data. This may include calibration history, match scores, and override decisions if your organization relies on them for documentation or audits.
Key Question: What relevant AI laws apply to your roles, offices, or organization?
Example discussion topics:
- Do any AI laws require the specific disclosure of the usage of AI tools in hiring?
- If disclosure is required, what specific language will your organization use to explain the use of AI tools in hiring?
- Do any laws require the specific retention of data as it's related to AI tools in hiring?
- Do any laws specify how long Talent Matching calibration history, match scores, or override decisions must be retained?
- Are there any other laws in your jurisdiction that may affect your usage of Talent Matching?
Note: Specific legal disclaimers on job posts can be updated in Greenhouse Recruiting on the Real Talent settings page. Read more about job post disclaimers..
2. Define the opt-out process for candidates
Some jurisdictions require that candidates be able to opt out of AI-based evaluation and request an alternative screening method. In these cases, employers must provide a non-AI option.
Candidate opt-outs from Talent Matching are supported through your internal processes rather than automated controls in Greenhouse. Your policy should define how opt-out requests are received, documented, and honored, including any non‑AI review steps required. Organizations should ensure these processes are in place to accommodate opt-out requests.
Key Question: How will candidates notify your team that they want to be opted out of AI processing of their resume?
Example discussion topics:
- What is the channel for opt-out requests?
- What processes are in place for documentation, auditability, and routing opted‑out candidates through your non‑AI review steps?
3. Define “human-in-the-loop” processes for your organization
Talent Matching helps assess candidate resumes against your defined and weighted calibration, but it does not make hiring decisions. Final decisions require human judgment. This human-in-the-loop approach supports compliance and fairness in hiring.
Key Question: How will your internal processes ensure human oversight for all decisions made using Talent Matching?
Example discussion topics:
- How will recruiters use Talent Matching during application review?
- Which Greenhouse features, such as the activity log or rejection reasons, will help keep recruiter actions current and accountable?
- What training will TA teams receive before using Talent Matching?
- How should recruiters handle candidates marked "Needs manual review" (for example, in jurisdictions where AI cannot be used or when a candidate has opted out), and how will those decisions be documented?
4. Define calibration standards
To ensure consistency, policies should define approved criteria and weighting practices for calibrations. While Greenhouse provides built-in safeguards to block protected attributes from being used in calibrations, your policy should outline your organization's approved job-related requirements and any prohibited criteria.
Standardizing these inputs supports transparency, consistency, and fair prioritization.
Your policy should also define how calibration history is reviewed over time, to confirm that changes remain aligned with approved criteria and do not introduce prohibited or proxy attributes.
Key Question: How are calibrations created at your organization?
| Example standard | How it might be implemented |
|---|---|
| Permitted criteria | Define how objective, job-related requirements such as skills, relevant industry experience, or job titles are used in calibration. |
| Prohibited criteria | Define how protected characteristics (or proxies, such as school prestige, “culture fit,” or personal traits unrelated to job performance) are prohibited from calibration. |
| Essential vs. preferred weighting | Assign higher weights to non-negotiable requirements. Use lower weights for preferred skills to avoid screening out otherwise qualified candidates. |
5. Define which users can view, create, edit, or delete calibrations
Another standard you can consider is the limited usage of Talent Matching to users who are trained on how to use AI-assisted tools in the hiring process.
This control can be updated by updating Job Admin permissions to restrict who can create, edit, and delete calibrations.
Key Question: Which users can view the results of calibrations? Can those same users create, update, or delete the calibration?
| Example standard | How it might be implemented |
|---|---|
| Trained users only | Access to Talent Matching is limited to individuals who have completed required training or policy documents. Users without confirmation cannot view or manage calibrations. |
| Recruiter-managed calibration | Recruiting team members manage calibrations. All users in the review stage can view match scores. |
| Broad hiring team access | Any hiring team member at the appropriate Job Admin level can create, edit, or delete calibrations. |
Example discussion topics:
- Which roles (for example, recruiters vs. hiring managers) can create or edit Talent Matching calibrations?
- Which roles can view match scores without being able to change the calibration?
- How will your team request and approve changes to calibration permissions over time?
6. Define policy around overriding match scores
Overrides are corrective actions used when manual review identifies missing or misinterpreted information, or when candidates opt out of AI processing and must be evaluated using your non-AI process.
Key Question: Under what conditions may match scores be overridden, and how is this documented?
Example discussion topics:
- What is a defined list of examples where an override may or should be used?
- What information must be documented in Greenhouse when a manual override is completed?
- What training can you provide to your teams to ensure overrides align with best practices, rather than subjective judgement?