Policy
Job Posting: AI Requirement Language Block
A recruiter, who pastes one block into the posting and deletes the tool bullets it replaces.
Four alternative wordings for the AI line in a job posting, ordered from a duties line that commits you to nothing to a stated requirement with a named check behind it. Each block carries the obligations it creates, so the strength you pick matches the evidence you actually hold. A table of lines to delete covers tool names, years-of-experience minimums and verdict language. The duties the wording was written against are listed at the end, and counsel decides which reach you.
Scope
Four wordings for the AI line in a posting, weakest first. Pick the one your evidence supports, paste it, and delete the lines it replaces. Each block carries what it commits you to, because a stated requirement is only as good as the file behind it. None of these is a complete posting and none is advice: this is clause text for your own counsel to edit.
The blocks
| Block | Use it when | What it asserts |
|---|---|---|
| One, duties line | The work uses an assistant and no bar has been written down | Nothing. It describes the job |
| Two, task requirement | A task list and a written bar already exist | A stated bar |
| Three, task requirement with a named check | The exercise behind the check already exists | A stated bar and how it gets tested |
| Four, process rule | AI use inside your own loop needs stating | A term of the process, not a bar |
Block one: duties line
Posting text, for the responsibilities section. You will work with an AI assistant on [the exposed task]. A normal week includes taking a draft it produced, checking it against [the source of truth], and saying which parts you kept.
What it commits you to.
- Nothing at the application stage. The line describes the work and removes nobody from the pile.
- An interview that can ask about that task, because the posting named it.
- No tool. A candidate who uses a different one still meets the line.
Use it when the team wants people who work well with an assistant, nobody has written down what that means, and nothing measures it yet. It says the true thing.
Block two: task requirement
Posting text, for the requirements section. You have taken output an assistant produced on [the exposed task], checked it against [the source of truth], and can say which suggestions you rejected and why. No particular product is required.
What it commits you to.
- A task list for this role naming where an assistant actually gets used, written from your own work rather than copied from another employer's posting.
- A written bar saying what clearing that task looks like, in terms someone else could observe.
- The same evidence, produced the same way, for everyone who reaches the stage where the line is applied.
Missing any one of those three, the line is a preference with paperwork behind it. Move it back to block one.
Block three: task requirement with a named check
Posting text, added to block two. Candidates who reach [the stage] complete a timed exercise in this role's own material, with an assistant available. It replaces [the first call] rather than adding a round.
What it commits you to.
- The exercise exists before the posting goes live. A named check nobody built is a claim, not a check.
- An answer key written before the first invitation goes out: what counts as a source opened, a suggestion refused, a number recomputed.
- A description rather than a verdict. Say what the candidate will do, how long it takes, and who reads the result.
Naming the check is the part that changes what arrives. Federal assessment guidance gives two counters to self-reported experience: creating an expectation that responses will be verified, and verifying them 1. The first is free and belongs in the posting. The second is why the exercise has to exist first.
Block four: the process rule
This is a different sentence from the requirement, and mixing the two is the common error. A requirement asserts a bar a candidate clears before applying. A process rule says what happens inside your loop and needs only to be true. Keep it in the process section with the stages.
Posting text, for the process section. The take-home is AI-open: use the tools you use. The call after it spends twenty minutes on the decisions you made in it. The live exercise runs without an assistant.
What it commits you to.
- A posture for every stage, decided before the wording. Application, assignment, live exercise and reference conversation do not need the same one.
- Interviewers who ask what the tools did. Skip the prompt logs: a log is a compliance artifact nobody reads, and the answer worth having is about judgment.
- No claim about authorship. Untrained readers asked to tell machine-written text from human writing performed at chance 2, and the tools sold for that job are neither accurate nor reliable 3. A rule published before anyone applied is a term of the process, and you enforce it by talking to the candidate about the work.
If an automated tool runs on applicants, the posting is often where notice lands. Write it as its own sentence naming the stage, and keep it out of the requirements list. Which notice duties reach you is a question for counsel, and the duties this wording was written against are at the end.
Lines to delete
Delete these on the same pass that adds a block above. A tool line surviving three bullets below the new one puts the old bar straight back.
| Line | Why it does nothing | What replaces it |
|---|---|---|
| Experience with [tool name] | True for anyone who has opened the tab once, so it removes nobody | The task the tool gets used on |
| AI-fluent, AI-native, prompt engineering experience | Adjectives, not requirements | The task, written as work |
| Five years of AI experience | Narrows by calendar rather than by capability | The task, with no number attached |
| Scored by AI, automated assessment | Reads as a verdict handed down before anyone has looked | What the candidate will do, how long it takes, and who reads the result |
- A written claim about AI fluency is cheap to produce. In a field experiment across nearly half a million jobseekers, candidates given algorithmic writing help on their resumes were hired more often, with no sign employers were less satisfied 4. By the time the claim reaches you it carries almost no information.
- Do not wire the line into the applicant tracking system as an automatic knockout question. Three tutoring companies settled federal claims that their application software automatically rejected applicants above an age cutoff 5. That rule was hand-written rather than a model, which is the point: a criterion nobody examined, applied automatically, is the shape of the problem.
- Do not hide an instruction in the posting to catch AI-written applications. It buys nothing a published rule does not buy honestly.
- Delete the requirement where the role never runs the task. A line earns its place when an assistant could draft a large share of the role's exposed task, that task runs weekly rather than annually, and a wrong answer reaches a customer, a regulator or a board before anyone catches it. A role meeting one of those gets tool access and a short policy, not a requirement.
Before the posting goes live
| Step | What it needs |
|---|---|
| Name the exposed task from the role's own work | The federal occupation task lists are a starting point, written by someone with no stake in your requisition 6 7 |
| Write the bar in terms someone else could observe | Somewhere the posted wording can be checked against it later |
| Decide the posture for every stage | Application, assignment, live exercise and reference conversation, settled before the sentence is written |
| Confirm the named check exists | If it does not, delete the sentence naming it |
| Delete the tool bullets the new line replaced | The same pass, not the next one |
| Send the final wording to counsel | What the applicant tracking system does with the line, and which notice duties reach the states and cities you post into |
The duties behind the wording
These are the duties the blocks above were written against. Which of them reach you is a question for counsel.
- Federal. Once a complaining party shows a particular practice causes disparate impact, demonstrate that it is job related for the position in question and consistent with business necessity. 42 U.S.C. 2000e-2(k), effective November 21, 1991.
- Illinois. Artificial intelligence with a discriminatory effect, and zip codes used as a proxy, are civil rights violations; an employer must also notify the employee that AI is used. 775 ILCS 5/2-102(L), effective January 1, 2026.
- New York City. A bias audit within the prior year, a public summary of its results, and notice to the candidate 10 business days before the tool is used. NYC Admin. Code 20-870 et seq., effective January 1, 2023.
- Texas. Developing or deploying an artificial intelligence system with the intent to unlawfully discriminate against a protected class is prohibited; disparate impact alone is not sufficient to show that intent. Tex. Bus. & Com. Code 552.056, effective January 1, 2026.
Limits
- No block here is a complete posting, a complete policy, or advice. Each one is clause text for your own counsel to edit.
- No wording here asks a candidate to disclose a disability, a health condition or an accommodation need.
- Nothing here produces a number about a person, and nothing here judges one.
- Nothing here tells you whether your own wording is lawful. That question takes your facts and your states, and it takes counsel.
Take it
The file and the credit
The publishing entity legal name and postal address are not filled in yet, and both sit inside the disclaimer every packaged format renders. No file is emitted until they are.
Credit line, to paste beside anything you quote from this document.
<!-- Olive template. Licence: https://olive.is/licence/template/1-0 -->
<p><a href="https://olive.is/answers/tools/job-posting-ai-requirement-language/" rel="nofollow">Olive</a>, Job Posting: AI Requirement Language Block, version 1.0.0, checked 2026-08-26.</p>Important notes
Not a substitute for the advice of an attorney. This is a starting document for your own attorney to edit. It applies no law to your facts. No attorney has reviewed it for your state or your facts, and Olive is not your lawyer.
Published by Olive Independent Study, Inc., [DELAWARE INCORPORATING ADDRESS], United States. Contact hello@olive.is. A person reads every complaint and answers within ten working days. All concerns that Olive has engaged in the unauthorized practice of law are referred to the North Carolina State Bar, wherever the complaint came from.
Checked 2026-08-26 against the sources listed in this file. Version 1.0.0.
This text disclaims no warranty, caps no liability, waives no remedy, and names no court or state for a dispute. Those absences are deliberate.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.
Packaged files
- The publishing entity legal name and postal address are not filled in yet, and both sit inside the disclaimer every packaged format renders. No file is emitted until they are.
Checks
- Sources last re-opened August 26, 2026.
- Next review due February 26, 2027.
Where it came from
Read the argument behind it
- How Do You Write an AI Requirement Without Flooding the Pipeline? Name a tool and every applicant who has opened it qualifies. Name the task the job runs, plus the sentence saying the claim gets checked, and the pool sorts itself.
- Put the AI Rule in the Job Post and Watch Who Still Applies One sentence in the process section: which stages are AI-open, and what the next call asks about. Published before anyone applies, it is the only version you can enforce.
- What Makes an AI Requirement in a Posting Defensible In the US you can require AI experience. What binds is the job-relatedness showing behind it: the tasks it came from, a written bar, and evidence applied evenly.
References
- Assessment and Selection: Training and Experience (T & E) Evaluations U.S. Office of Personnel Management, 2026. opm.gov Supports the two counters to self-reported experience named in block three: creating an applicant expectation that responses will be verified, and carrying out the verification.
- All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Text Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL-IJCNLP 2021), 2021. aclanthology.org Supports the claim in block four that an unaided reader cannot tell machine-written text from human writing, which is why a rule works as a published term rather than as an inspection.
- Testing of Detection Tools for AI-Generated Text arXiv preprint of the paper in International Journal for Educational Integrity, 2023. arxiv.org Supports the claim in block four that the available tools sold for identifying machine-written text are neither accurate nor reliable enough to carry a decision about a candidate.
- Algorithmic Writing Assistance on Jobseekers' Resumes Increases Hires Wiles, Munyikwa and Horton, National Bureau of Economic Research (NBER Working Paper 30886; published in Management Science, 2025), 2023. nber.org Field experiment in an online labor market with nearly half a million jobseekers. Supports the claim that a written fluency claim is cheap for any applicant to produce and therefore carries little information.
- iTutorGroup to Pay $365,000 to Settle EEOC Discriminatory Hiring Suit U.S. Equal Employment Opportunity Commission, Newsroom press release, 2023. eeoc.gov Supports the worked example of an automatic age cutoff inside application software producing federal enforcement. The consent decree carries no admission of liability.
- 13-2051.00 - Financial and Investment Analysts O*NET OnLine, U.S. Department of Labor, 2026. onetonline.org One of the two published federal occupation task lists named as the starting point for an exposed task. Accessed 24 August 2026.
- 15-1252.00 - Software Developers O*NET OnLine, U.S. Department of Labor, 2026. onetonline.org The second published federal occupation task list named as the starting point for an exposed task. Accessed 24 August 2026.
7 sources, numbered by first appearance. How Olive sources claims