Methodology
Our Methodology
What a Posting Scan Covers
A scan reads the language of one job posting for potential risk indicators under:
- Title VII of the Civil Rights Act: language tied to sex, race, color, religion or national origin
- Age Discrimination in Employment Act (ADEA): age-coded language
- Americans with Disabilities Act (ADA): physical requirements and accommodation wording
Selection requirements such as a diploma or degree are listed for human judgment: under the federal Uniform Guidelines, whether one is appropriate depends on the job, so a person decides rather than the scan.
What needs selection data
A posting can't show who applied or who was hired, so these belong to the AI Hiring Tool Review:
- NYC Local Law 144 readiness for automated hiring tools (the law's required audit is done by an independent auditor)
- Four-fifths (impact ratio) analysis of selection rates
How Scoring Works
Inputs
The score comes only from fixed rules that read the posting’s wording. The same posting always gets the same score and the same findings. AI writes the explanation and the draft rewrite; it never adds, removes or rescores a finding.
Weights
Each finding has a severity. Each one adds risk by taking a share of the room that is left, so the score rises with every finding but a handful of findings never reaches 100.
- High: 45% of the remaining room. For example age-coded words such as “young” or “recent graduate”, native-speaker requirements, citizenship requirements.
- Medium: 25% of the remaining room. For example blanket physical requirements such as “must be able to lift”, gender-coded words such as “rockstar”.
- Low: 10% of the remaining room. For example weaker age signals on their own, such as “energetic”.
- An explicit exclusion of a protected group (for example “whites only” or “no disabilities”) always scores at least 75.
| Findings | Score | Level |
|---|---|---|
| One Medium finding | 25 | Low |
| One High finding | 45 | Medium |
| Two Medium findings | 44 | Medium |
| Two High findings | 70 | High |
| Two High and one Medium | 77 | High |
| Two High and two Medium | 83 | High |
What moves the score
Only findings: more of them, or more severe ones, raise it. Removing or rewording a flagged phrase lowers it.
What doesn’t
- Items listed under “Needs human judgment” (driving, diplomas, experience ranges, background checks, drug tests, shift length)
- A missing EEO or accommodation statement
- Pay, posting length, links or contact details
- The AI-written explanation and draft rewrite
What it doesn’t measure
The score reflects the posting’s wording only. It is not a legal conclusion, and it doesn’t measure intent, who applied or was hired, state or local rules, or whether a requirement is essential to the job.
The Four-Fifths Rule (AI Hiring Tool Review)
The impact ratio compares selection rates between groups. It needs real selection data, so a posting scan doesn't calculate it; the AI Hiring Tool Review does, with the hiring data you provide.
IMPACT RATIO
When this ratio falls below 0.80, it may signal adverse impact. The four-fifths rule is a rule of thumb federal agencies use to flag possible adverse impact; it is not a legal conclusion.
What You Receive
Every free scan gives you:
- Risk score (0–100) with severity classification
- Flagged language with specific regulatory citations
- Why-it-matters summary
- Risk summary by law
- A draft rewrite, with the open items a person decides before posting
- Downloadable PDF report — structured for counsel review
Sample · fictional posting. A free scan of one posting gives you a live report you can download as a PDF.