Pipeline
Required or Preferred: Where AI Skills Belong in a Posting
For most roles, an AI-skills line belongs in neither column of a job posting: not required, and not preferred. Required means you would turn down a strong candidate on day one for lacking it, and that is rarely true of a general AI assistant: a new hire learns the tool inside a normal ramp, and the judgment that makes its output good is what you were hiring for anyway. Write the AI into the responsibilities instead, and if a qualification must exist, make it an observable behavior with a bar.
The takeThe required-versus-preferred debate is usually argued as an applicant-volume problem, and that framing lets both options off the hook. Preferred is only the safe choice where nobody screens on it, and a line nobody screens on is a line nobody needed. Once anybody sorts on it, the label on the section has changed and the selection has not. The middle option is where a team keeps the screen and quietly gives up the reason for it, and that is the version worth refusing.
Where Olive fits
Open a role and see what the work shows
A posting states a bar; no posting can show how a particular person works with an assistant. Olive is a 40-to-60-minute occupational assignment done with an AI assistant, written up by a human reviewer as six findings, each carrying the timestamped excerpt it rests on, and the candidate is granted the same report.
Rank your shortlistWhat does required actually commit you to?
Required is a promise about day one. It says a candidate without this cannot do the job on arrival and cannot be brought to it inside the ramp you already budget for. Apply that test to a general AI assistant and it usually fails, because the assistant is the teachable half and the judgment that makes its output any good is the slower, older thing you were already hiring for.
It is also a promise to a regulator, though almost nobody writes it that way. Federal selection law classifies a hiring step by its function in the decision. The Uniform Guidelines on Employee Selection Procedures, the 1978 federal regulation still in force, define a selection procedure as any measure, combination of measures, or procedure used as a basis for an employment decision, and say the term reaches the full range of assessment techniques, from paper-and-pencil tests through informal or casual interviews and unscored application forms 1. A qualification you sort on sits inside that definition, and so does the phone screen where you ask about it.
There are roles where required is the right word:
- The tool is the job. An AI product role, an internal enablement role, a role whose output is the model configuration itself.
- The tool is not substitutable. A licensed clinical documentation system, a proprietary trading stack, a workflow a regulator names by product.
- Nobody can be brought to it after the offer. Access is gated, onboarding to the system takes months, or the ramp is measured in quarters rather than weeks.
Outside those three, required is a preference written in the language of a floor, and the cost of that lands on candidates who read the posting literally and self-select out.
Why preferred is not the safe default
Because preferred still gets used. A line in the preferred column reaches the same recruiter, the same screening rubric and the same knockout question, and once anyone sorts on it the label has changed nothing about what the applicant experienced. The safety in preferred comes from not screening on it, which is a discipline nobody wrote down and nobody audits.
The familiar argument for a lean required list is about applicant volume: that some candidates, women especially, hold back unless they match nearly everything. Hold that as a hypothesis about behavior rather than as a finding you can build policy on. The federal statute underneath both columns is firmer ground. Under 42 U.S.C. 2000e-2(k), added by the Civil Rights Act of 1991, an employer has to demonstrate that a challenged practice is job related for the position in question and consistent with business necessity, once a complaining party has first shown that the practice causes a disparate impact on the basis of race, color, religion, sex or national origin 2. Age claims run under the ADEA and disability claims under the ADA, on standards of their own, and age is the ground an AI-experience line is most likely to touch.
That describes what somebody would have to prove and what you would then have to show. It obliges nobody to audit a posting in advance, and what it eventually looks at is whichever practice actually removed applicants, wherever that line happened to sit in the document. So the question worth carrying into the draft is whether this line is taking people out of the pile, and where it is, the word above it does not change the practice somebody could point at later. This is public legal ground, not legal advice.
Two neighbouring questions matter more than the label: whether a required AI line screens out the people you want, and what makes an AI requirement defensible once somebody asks you to explain it.
Write the AI line into the duties instead
The responsibilities section carries no screening burden, so that is where an honest account of AI in the work belongs. Write what the person does: drafts the first pass of the analysis with an assistant, checks the numbers against the source system, and can say which parts of the draft got thrown away. A candidate reads that and knows the job. Nobody comes out of the pile because of it.
It also tells the reader something a qualification never can. 'Familiarity with AI tools' says nothing at all. 'You will draft client memos with an assistant and be accountable for every citation in them' says what a Tuesday looks like, and it selects in the only way a posting honestly can, by making the wrong candidates lose interest.
AI language has reached ordinary postings, and it is still a minority of them. As of December 2025, Indeed Hiring Lab found about 9% of human resources postings and about 15% of marketing postings carried an AI-related term, against nearly 45% in data and analytics 3. Those numbers count vocabulary. A mention does not mean the role tests for AI skill, or that anyone has written a policy for it.
When the tools do come up in the draft, name the behavior rather than the product. Product names in a duties section date quickly, and they hand every applicant the vocabulary to mirror back.
Set an observable bar if it has to be a qualification
Write a behavior somebody could watch: has drafted work of this kind with an assistant and can walk through what was changed and why. Some requisition systems will not accept a role without a qualification line, and some roles genuinely need one, so that is the sentence to write when the line has to exist. Then state where the bar came from, and apply it to every applicant the same way.
Three artifacts make that real, and none of them takes long:
1. A task list. The specific tasks in this role where an assistant gets used, written down before anybody drafts the qualification line. 2. A written bar. What clearing it looks like on those tasks, in terms somebody could observe, agreed before the first resume arrives. 3. Per-candidate evidence. One record per applicant showing how they were measured against that bar, produced the same way for everyone who reached that stage.
Skip any of the three and the qualification is a preference with a stern label on it. The first artifact is the one teams skip, and it is the only one that makes the other two mean anything: a bar derived from a competitor's posting is not derived from the work.
Working the bar out for one specific role is its own exercise, covered in setting a defensible proficiency bar and in writing the requirement so it is not legally vague. Whichever way you go, run the final wording past counsel.
Common questions
Can a posting have no AI line at all?
Yes, and for many roles that is the right call today. A posting is not an inventory of every tool in the building. If an assistant is simply one of the things a competent person will use, the way a spreadsheet is, the duties section can describe the work without naming it. Add the line when a candidate would be surprised on their first week, when the role changes how the team uses AI, or when someone would decline the job on learning it.
Does moving the AI line to responsibilities hurt search ranking on job boards?
The terms are still in the posting wherever the line sits; what changes is what the applicant reads as a gate. The trade is real in one direction: fewer applicants filter themselves out, which means a slightly larger pile. If pile size is the binding problem, the answer is a better first stage, not a stricter qualification line that you are not prepared to defend.
What if the hiring manager insists on required?
Ask whether a strong candidate lacking it would be unhirable on the first day, and write down the answer. If they can name the tasks that fail without it, that list is the start of a defensible requirement and required is probably right. If the answer is that they want strong candidates, that is a preference about quality and belongs in the duties. Either way the conversation produces the artifact you need later, which is a written reason the bar exists.
Should required qualifications list a number of years?
Not for AI tool use. A year count reports accumulated exposure, and it only reports anything once a tool has been stable long enough for exposure to accumulate. Recency plus a specific piece of work carries more and dates better. If the requisition system forces a number into the field, attach it to the occupation: four years in claims adjudication, not four years of AI.
How many required qualifications should a posting carry?
Few enough that you could defend each one from the tasks the role actually contains. Read yours line by line and ask which ones you have ever turned somebody down for. The lines nobody has ever enforced are the ones costing you applicants for free, and they are also the ones that read worst if a rejected candidate asks where the bar came from.
References
- 1. 29 CFR Part 1607 - Uniform Guidelines on Employee Selection Procedures (1978), sections 1607.16(Q) and 1607.3(A) govinfo.gov Supports the claim that a stated qualification you sort on is a selection procedure under federal law, alongside tests and informal interviews.
- 2. 42 U.S.C. 2000e-2(k) - Burden of proof in disparate impact cases uscode.house.gov Supports the job-relatedness and business-necessity showing an employer owes once a complaining party demonstrates a practice causes disparate impact, and the Title VII characteristics the section reaches.
- 3. January 2026 US Labor Market Update: Jobs Mentioning AI Are Growing Amid Broader Hiring Weakness hiringlab.indeed.com Supports the sector figures for how many HR, marketing and data postings carried an AI-related term as of December 2025.
3 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
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.