AI Resume Screening: How It Works and When to Opt Out

TL;DR: AI resume screening reads your resume in two steps: a parser turns the file into fields, then a matching model grades how well those fields fit the job. In the systems that document it, the grade sorts you and a recruiter decides. Should you opt out of AI resume screening when a job post offers it? Opt out if your titles or career path read badly to a matcher; stay in if your resume maps cleanly to the ad. Test the parse first.

What AI resume screening does with your application in 2026

Three separate steps sit behind the phrase, and each fails for different reasons.

Parser: software that converts your PDF or DOCX into structured fields (name, employers, titles, dates, skills). Matcher: a model that compares those fields with the job's requirements and assigns a grade or category. Knockout question: a yes/no application question (work authorization, licence, location) that a rule can use to reject you before anyone reads the resume.

The matcher is the part people mean by "AI screening", and two of the biggest applicant tracking systems document theirs. Greenhouse's Talent Matching FAQ (updated 5 October 2026) says the AI analyzes your resume against calibration criteria the employer sets (required skills, relevant experience, job titles, optionally industry, each weighted) and sorts you into categories such as Strong, Good, Partial and Limited. The same page states that "Talent Matching does not auto-reject or auto-advance any candidate." Workday's HiredScore grades each application from A to D. Pfizer's published bias-audit summary describes A as a strong match with the requisition and D as a poor one, and says HiredScore "does not automatically make or execute any decisions."

Vendor documentation describes the tool. Employers decide how to use it. In the Resume Genius 2026 Hiring Insights Report, a Pollfish survey of 1,000 US hiring managers, 19% said they use AI to purposefully screen out applications before a human reviews them, and 6% said AI moves candidates forward or rejects them with limited human review. So for most applications a recruiter still clicks the reject button, but the grade decides which pile they open first.

Bias is the reason regulators care. A University of Washington study ran more than three million resume-to-job comparisons through models from Mistral AI, Salesforce and Contextual AI: white-associated names were preferred 85% of the time against 9% for Black-associated names, and male names 52% against 11% for female names. Those were research models, not a named ATS, but they are the same family of technology.

Parser vs matcher: what breaks each one

The parser runs first, and its output is all the matcher sees. Greenhouse's Data Processing FAQ says the source data is "their submitted resumes" and that "structured application form data is not included in this calculation", so your resume carries the whole match. The same FAQ says the matching algorithm does not use names or contact information.

When parsing fails outright, Greenhouse does not guess. Its FAQ lists "rare resume formatting or technical problems" among the reasons a candidate gets the label "Needs manual review" instead of a match category. That is a softer outcome than the myth of instant rejection, and a worse one than a Strong match. Hiring managers notice formatting too: 53% of the Resume Genius sample said text-based PDFs with no images or complex formatting work best with their ATS. Our ATS-friendly resume format guide shows what extraction does to columns, tables and headers.

Matchers work on meaning, not exact strings. Eightfold's engineering blog says "every resume and job description is embedded into a high-dimensional vector space", so "Pandas" sits near "Python", not near "Panda". Greenhouse adds that several resume terms can map to one calibrated skill and "a longer list of matched terms doesn't always mean a higher match score." Repetition buys nothing; evidence does. Hidden white-text keywords are still text, so they land in the parsed profile a recruiter reads. Put the skills you can prove inside bullets, as the resume skills section guide lays out.

Run this check on the exact file you plan to submit:

1. 1

   Open the exported file

   Use the PDF or DOCX you will upload, not the editable source.

2. 2

   Select all and copy

   Press Ctrl+A or Cmd+A, then copy.

3. 3

   Paste into a plain text editor

   Notepad, TextEdit in plain-text mode, or VS Code. Ignore the styling.

4. 4

   Read top to bottom

   If sidebar items land inside job entries, dates drift away from their roles, or your name and email are missing, a parser is likely to see the same order. Rebuild those parts in a single column.

Should you opt out of AI resume screening?

The question comes from a link that now appears on some job posts. Greenhouse's Talent Matching setup page (updated 24 August 2026) describes an optional setting, "Enable AI opt-out for candidates". When an employer's site admin switches it on, candidates see "a link in the AI disclaimer on the job post" where they can manage their automated processing preferences. Greenhouse's documentation spells out what follows:

• You are not scored. Opted-out candidates "are labeled 'Needs manual review' and are not evaluated by Talent Matching."
• An existing score is deleted. If you opt out while the application is active, Greenhouse "deletes any existing AI-generated match score and reasoning for that application" and flags it for manual review.
• A rejection stays. If you were already rejected before opting out, the AI result may be retained under the employer's retention policy.
• The employer designs the manual queue. Greenhouse tells admins to work with legal and recruiting teams "to decide when this setting should be enabled and how your team will handle manual review of opted-out candidates."

That last point is the honest answer to the forum question. Nobody outside the company knows how fast its manual queue moves. Some employers state their policy. Okta's automated processing notice says its AI tools give recruiters a match score, that "opting out will not negatively impact your application", and that the application "will be reviewed manually by our team." Acquia's AI screening notice, effective 10 August 2026, covers all applicants and says a request to opt out or for an alternative review "will never negatively affect your application."

Opting out does not skip the ATS, the knockout questions or the recruiter. It swaps a category label for "Needs manual review".

Your rights: notice, alternatives and human review (status 13 October 2026)

Most US candidates have no legal right to opt out. Where a law applies, it mostly gives you notice, sometimes a right to ask for something else, and rarely a guaranteed human.

New York City. DCWP's AEDT page sets the 10-business-day notice and the one-year bias audit. Rule § 5-304 in DCWP's adopted rules says the notice must explain how to request an alternative selection process or accommodation "if available", and adds: "Nothing in this subchapter requires an employer or employment agency to provide an alternative selection process."

Illinois. The amendment took effect on 1 January 2026 and requires notice of AI use in recruitment and hiring. On 2 June 2026 the Department of Human Rights postponed its proposed notice rules, so the format of that notice is still unsettled, per Ogletree Deakins.

Colorado. The General Assembly's SB26-189 page says the act "repeals and reenacts" the 2024 provisions for automated decision-making technology in consequential decisions, including employment. After an adverse outcome, the deployer owes a plain-language description of the tool's role within 30 days, and you can request "meaningful human review and reconsideration." There is no private right of action; the attorney general enforces it.

European Union. Annex III point 4(a) lists AI for recruitment or selection, including screening applications, as high-risk. The amended Article 113 (Digital Omnibus, Regulation (EU) 2026/1744) moved Annex III obligations to 2 December 2027. Then employers must tell you a high-risk system is used (Article 26(11)) and explain its role in an adverse decision (Article 86). The ban on AI that infers emotions in the workplace, Article 5(1)(f), has applied since 2 February 2025.

In the courts, Mobley v. Workday tests whether a software vendor can be liable for screening outcomes. On 16 May 2025 Judge Rita Lin of the Northern District of California granted preliminary certification of an age-discrimination collective action, as Fisher Phillips reported. No court has ruled on the merits.

How to write a resume that passes the parser and the matcher

Formatting fixes the parser, content fixes the matcher. Items 1–4 are parser work, items 5–7 matcher work.

1. Single column. No sidebars, tables or text boxes for core content.
2. Standard headings. "Experience", "Education", "Skills". A parser maps known labels to fields.
3. Contact details as plain text in the body. Email, phone, city and LinkedIn URL on one line, without icons and not only in the page header.
4. One date format. "Jan 2022 – Present" throughout, each date range next to its role.
5. Title wording the matcher can map. If your official title was "Associate III" and the ad says "Data Analyst", write "Data Analyst (official title: Associate III)". Only where it is true.
6. Named tools inside bullets. "Built churn dashboards in Tableau from PostgreSQL" gives the matcher two skills with evidence; a skills block alone gives it a list.
7. Outcomes with numbers. "Cut month-end close from 9 days to 5" tells a human reviewer what the grade cannot.
8. Length that fits your history. One page for early-career resumes, two for senior ones; the CV length guide gives the thresholds.

Then check the file before it reaches an employer's matcher. The free ATS checker shows the fields a parser pulls from your resume and flags the layout problems in the table above, so you can fix them before deciding whether to opt out.

Common questions about AI resume screening

Will opting out of AI screening hurt my application?+

It depends on the employer's manual-review process, which only the employer knows. Greenhouse labels opted-out candidates 'Needs manual review' and leaves the review process to the employer. Okta and Acquia state in their notices that opting out does not count against you. If the notice says nothing, ask the recruiter.

Is it legal to use AI to screen resumes?+

Yes in most of the US, with conditions in some places. New York City requires a bias audit and notice, Illinois requires notice, and Colorado's 2026 act adds explanation and human-review rights after an adverse outcome. In the EU, recruitment AI is high-risk under the AI Act, with obligations from 2 December 2027.

Do AI screeners check whether my resume was written by AI?+

Some employers screen for bot-generated applications. Acquia's notice describes 'application integrity screening' that flags signals of fraudulent, duplicate or automated applications. That targets mass-submitted applications; a matcher like Talent Matching scores the content against the job.

Does a human see my resume if the AI grades it low?+

In Greenhouse and HiredScore, a low grade sorts you lower; neither vendor's documentation says the tool rejects you. Whether a recruiter reaches the lower categories depends on volume and the employer's process.

Key takeaways

• "AI screening" is two systems: a parser that builds your record and a matcher that grades it. A broken parse reaches the matcher as missing data, so fix the file first.
• In Greenhouse and Workday HiredScore, the documented output is a category or grade for a recruiter, not an automatic rejection; knockout questions are where rule-based rejection happens.
• The "opt out of AI review" link on a job post is usually an employer-enabled ATS setting. In Greenhouse it removes your score and puts you in a manual-review queue whose speed the employer decides.
• Opting out makes sense when your titles or career path would grade poorly against weighted criteria; staying in makes sense when your resume maps cleanly to the ad.
• Legal rights are local: notice in New York City and Illinois, explanation and human review after an adverse outcome in Colorado, and EU high-risk duties from 2 December 2027.