Refolk
11 min readMarkdown

Colorado Owes You an AI Rejection Reason in 30 Days. Here's the Letter

From Jan 1, 2027, Colorado's SB 26-189 lets applicants rejected with AI help ask for an explanation, data corrections and human review. Here's how to ask.

Most rejections show up with no reason attached, and when software made the cut you can't even tell whether a person read your resume. Starting January 1, 2027, Colorado's SB 26-189 changes that for applicants rejected with help from AI: you can ask for a plain-language explanation within 30 days, correction of wrong data, and a human review. California's automated decision-making rules give applicants a different set of rights. The catch is that enforcement is paused and you can't sue. Send the request anyway, and treat whatever comes back as feedback on your resume.

What SB 26-189 gives rejected applicants from January 1, 2027

Starting January 1, 2027, anyone rejected in a Colorado hiring decision that AI materially influenced can ask for a plain-language explanation of the technology's role, delivered within 30 days. They can also ask to correct inaccurate personal data and request meaningful human review and reconsideration, where commercially reasonable.

Colorado signed SB 26-189 on May 14, 2026, rewriting its earlier AI law. A consequential decision is one that decides your access to, eligibility for, or opportunity in something important, and hiring is the clearest example. A deployer is the employer using the AI tool.

The rights a rejected applicant can invoke:

  1. An explanation, in plain language and within 30 days, of how the technology influenced the outcome.
  2. Correction of personal data that is factually wrong.
  3. Meaningful human review and reconsideration, where commercially reasonable.

Buchalter's reading sets a high bar for "meaningful." The reviewer has to be able to override the decision, look at the relevant evidence, be trained for the job, and not just go along with the system's output. A recruiter who signs off on an AI-generated ranking without real deliberation doesn't meet the standard.

30 days
Deadline for a plain-language explanation under Colorado SB 26-189

Applies to rejections where AI materially influenced the outcome, starting January 1, 2027.

Who is covered in Colorado and California

You're covered in Colorado if you live there, or if a business operating in Colorado evaluates you for an opportunity in Colorado. California's CCPA ADMT rules work differently: covered employers must give notice before the tool runs, let you opt out in many cases, and explain the output if you ask.

Under Colorado's law, the protected group includes employees, applicants who reside in Colorado, and anyone whose opportunity in Colorado is evaluated by a business operating there. So a Texas resident applying for a Denver role can plausibly qualify. A Colorado resident applying for remote work at a company doing business in the state can too.

California puts the obligation at the front of the process instead of after the rejection.

RightColorado SB 26-189California CCPA ADMT
Notice before useNotice duties exist, scope contested in rulemakingPre-use notice required
Opt out of the toolNot among the listed rightsYes, unless a named exception applies
ExplanationWithin 30 daysOn request, of how the tool reached its output
Human reviewWhere commercially reasonableA genuine human appeal can replace the opt-out
Correct your dataYesNot part of the ADMT notice package

In California, if an employer offers a genuine human appeal, it often no longer has to offer an opt-out. When a California posting mentions an appeal process, take it seriously, because it may be the only lever you get.

Which screening tools count, and which slip out

Tools that reject candidates on their own, such as knockout questions that disposition you automatically, fall squarely inside the law. Tools used solely to summarize, organize, translate, draft, route, or present information for a human reviewer are excluded.

As of November 2026, the vendors split on this question:

ToolCan reject without a person?What it does
HumanlyYesConfigurable rules disqualify candidates who miss knockouts, written back to the ATS
Paradox (Workday)YesConversational screening can disposition candidates who miss requirements
Indeed screener questionsYes, reversibleMissing a required answer moves you to a "Rejected" list the employer can undo
Greenhouse Talent MatchingNoLabels candidates Strong, Good, Partial, or Limited Match
Ashby AI-assisted reviewNoAssists review, does not advance or reject

A knockout question is a required screening question, such as work authorization or shift availability, where the wrong answer filters you out automatically.

"A human made the final call" doesn't settle it

The carve-out covers tools used solely to present information. Greenhouse Talent Matching does more than that. It sorts you into a category, highlights resume terms, and lists matched and missing skills. A label like "Limited Match" is a classification, and a recruiter working through a long list may never open the files at the bottom.

This was the fight at the Oct. 26, 2026 hearing on the attorney general's proposed rules. Business groups argued that the draft definition of "material influence" could trigger notice and review even when a person makes the final decision. That pushback tells you the drafted rules read broadly. So don't stop at asking whether AI rejected you. Also ask whether a match label, score, or ranking was applied before any human looked at your file.

Why large regulated employers are where knockouts live

Auto-rejection risk is most likely at large enterprises, because that's where Workday is, and Workday owns Paradox. In Refolk's index of professional profiles, US recruiters who list Workday as a skill outnumber those who list Greenhouse 4.4 to 1.

The index counts US profiles with "Recruiter" or "Talent Acquisition" in the title, segmented by which ATS skills they list and whether they mention high-volume hiring:

SegmentProfilesSample employers
Lists Workday skill523Wells Fargo, Boeing, Lifepoint Health, C&S Wholesale
Mentions "high volume"127ManpowerGroup, Bank of America, IBM, Indeed Flex
Lists Greenhouse skill118Meta, Palo Alto Networks, Lumos, Paraform
4.4x
US recruiters listing Workday vs. Greenhouse in Refolk's index

The Workday group skews toward banks, aerospace, and healthcare. Greenhouse skews toward tech.

These are filtered-index counts, not a census, so treat them as ratios rather than market sizes. A recruiter listing Workday doesn't prove their employer runs Paradox. Still, the pattern matters. Paradox has belonged to Workday since October 1, 2025, and the Workday-heavy employers in the sample are banks, aerospace firms, and hospital systems. Recruiters who mention "high volume" roughly equal the Greenhouse group (1.08x), so high-volume pipelines are not a niche.

What this means for you: if you apply to a large regulated employer, expect a conversational screener or knockout form, and expect it to be the kind of tool SB 26-189 covers. Applications to a Greenhouse or Ashby shop are more likely to involve match labels, which puts you in the contested "material influence" zone.

Before answering knockouts, it helps to know which hard requirements you actually meet. Refolk scores your fit against the posting before you apply, so a hard requirement won't catch you off guard in a chatbot.

The catch: enforcement is paused and you can't sue

Your rights begin January 1, 2027, but nobody is enforcing them yet, and you have no private right to sue under SB 26-189. That still doesn't make the request pointless.

The specifics:

  • Enforcement stay. An April 2026 court order in xAI v. Weiser (Colorado AG Phil Weiser is the defendant) holds off enforcement until 14 days after the court rules on xAI's planned preliminary injunction motion.
  • Cure period. Even once enforcement starts, the AG usually has to give an employer 60 days' notice to fix a violation.
  • No private lawsuit. Applicants can't sue under this statute.

Here's why you should send the request anyway. The law says it doesn't limit existing rights or remedies under state or federal law, discrimination law included. Deployers have to keep records for at least three years after the decision. A dated written request establishes that you asked, what you asked for, and when. If the employer answers, you learn something. If it ignores you, that silence is on record for whenever enforcement starts or a discrimination claim comes up.

A dated request costs you almost nothing. The employer's answer, or its silence, is evidence either way.

How to word the request

Email the recruiter or hiring contact on your rejection, copy any privacy or HR address listed in the posting or privacy policy, and ask for each right by name. Keep it short and factual, cite the statute, include the requisition ID, and keep it neutral: a calm request reads as diligence, an accusatory one gets a form reply.

A template you can adapt:

Subject: SB 26-189 request, [Job Title], Req [ID], application dated [date]

Hello [Name],

I applied for [Job Title] (Req [ID]) on [date] and received a rejection on [date]. I am a Colorado resident [or: this role is based in Colorado]. Under Colorado SB 26-189, I am requesting:

  1. A plain-language explanation of how any automated decision-making technology, including scoring, ranking, match labels, or screening questions, influenced this decision.
  2. Correction of any inaccurate personal data used to evaluate me, including how my resume was parsed and my answers to any screening questions.
  3. Meaningful human review and reconsideration of my application by someone with authority to change the outcome.

Please confirm receipt. I understand the explanation is due within 30 days.

Thank you, [Name, email, phone]

For California, swap the statute line for "Under the CCPA automated decision-making technology regulations, I request an explanation of how the tool reached its output and access to any human appeal process."

Save everything with dates: the posting, the confirmation email, the rejection, and your sent request.

Which requests are most likely to work

A correction of a factual error is the strongest request you can make. Asking for human review is weakest exactly where rejection volume is highest.

Ranked from strongest to weakest:

  1. Correcting a wrong knockout answer or misparsed fact. A mistyped work-authorization answer, a wrong shift selection, or a start date parsed as 2012 instead of 2021 is "factually incorrect personal data." Indeed's Rejected list is reversible, so the employer can simply move you back.
  2. Getting the explanation. It's due within 30 days, and the "commercially reasonable" qualifier is attached to correction and review, not to the explanation.
  3. Human review for professional roles. When a role draws a modest number of applicants, re-reviewing one file is plainly reasonable.
  4. Human review in high-volume hourly hiring. The Colorado Chamber of Commerce argued at the October hearing that notice and review for large numbers of rejected applicants would be a significant burden. Expect employers to rely on the qualifier here.

Demand for this is real. Greenhouse's 2026 Candidate AI Interview Report reportedly found that 46% of US job seekers wanted the option to request a human interview. Wanting it doesn't mean getting it, and in high-volume pipelines the default is still a form rejection. You fill out the same kind of application for the same kind of role over and over, and a screener turns you down before any person sees your name.

Turning the explanation into a better resume

Whatever the employer sends back shows you how its software read your resume, and that's information you normally never get. Use it to fix the problems that will trip the next screener too.

What to check in the response:

  • Parsed dates. If the response shows wrong dates or a missing role, your formatting broke the parser. Put month and year on the same line as each title.
  • Matched vs. missing skills. Greenhouse Talent Matching shows highlighted terms plus matched and missing skills. A skill on your missing list that you do have means you described it in different words. Rename it to match the posting.
  • Knockout answers. If you were rejected on a requirement you meet, such as work authorization, the question was ambiguous or your answer was off. Answer more carefully next time and fix this record through the correction request.
  • Opt-out flags. Greenhouse marks candidates who opt out of matching as "Needs manual review." In California, where you can often opt out, that route can put your file in front of a person, which helps if your background doesn't map neatly onto keywords.

Doing that rewrite by hand for every posting gets old quickly. That's the work Refolk takes off you: paste the posting and get your own history back as a resume rewritten in that posting's vocabulary, plus a cover letter and a fit score showing which requirements still read as missing.

FAQ

Can I appeal an AI job rejection if a recruiter made the final decision?

Often, yes. The exemption only covers tools used solely to summarize or present information. If a tool ranked, scored, or labeled you (Strong, Partial, Limited Match, for example) before a person reviewed you, that's arguably material influence, and the Oct. 26, 2026 hearing fight was over exactly that. Ask whether any score or label was applied, and let the employer explain why it believes the exemption applies.

What happens if the employer ignores my request?

For now, very little happens directly. Enforcement is paused under xAI v. Weiser, the AG has to give a 60-day cure notice first, and you can't sue under this statute. But employers must keep records for three years, and existing discrimination laws still apply. Your dated, unanswered request becomes part of that record.

Does this apply to rejections before January 1, 2027?

No. Colorado's SB 26-189 rights start January 1, 2027, so a rejection that arrives before then isn't covered by them. If you're mid-process with an employer that does business in Colorado in late December, save your application materials. A rejection that arrives in January is covered, as long as the tool and the role fall within the law.

The data behind this

Put this to work

Paste your career in once. Every application after that is written for you.

Drop a resume or a LinkedIn URL. I rank the live openings against it, rewrite the resume and write a cover letter for the best of them, and fill in the employer's form when you press the button. You read, you decide what goes out.

  1. 01Drop your resume

    A PDF or a LinkedIn URL. About a minute, once.

  2. 02I rank the openings

    Every weekday morning, the live catalog scored against your history. Up to 20 worth your time, not two hundred links.

  3. 03Each one is written up

    Resume rewritten for the posting, a cover letter, a fit score. Press send, or let me fill in the form.

  • New matches ranked and written before you are up.
  • Every bullet stays inside what your history supports. Nothing invented.
  • Queued, submitted, interviewing, offer: one screen, not a spreadsheet.

Free, no card. Up to 25 applications sent a week, and nothing is sent until you say so.

Keep reading