Pipeline
Write the Or-Equivalent Line So a Screener Can Use It
An or-equivalent line on an AI requirement does nothing unless the posting says what will be accepted. Name two or three substitutes. A project the candidate can walk through decision by decision, a task they used to do by hand that a model now drafts, and one occasion when a model was wrong and they caught it. Then screen against that list. A blank AI line often records where somebody worked, not what they can do.
The takeThe equivalence clause is where a hiring team's honesty gets tested cheaply. Writing one costs nothing. Honoring one costs an argument with a manager who wanted a familiar company name at the top of the resume. So the clause survives in postings that never meant it, and a candidate has no way to tell the two kinds apart from outside. If you would not defend a substitute at the debrief, cut the line and require the experience openly. A requirement you enforce treats people better than an exception you advertise and refuse.
Where Olive fits
Open a role and see what the work shows
Once a candidate with no employer to vouch for the tool is on a shortlist you already have, Olive assesses the work itself: a 40-to-60-minute occupational assignment done with an AI assistant, returned as six findings a human reviewer wrote or edited, each carrying the moment in the session it rests on. The candidate is granted the same report.
Assess your shortlistWhy does the missing AI line describe the employer, not the candidate?
Because the tool has to be in the building before it can reach a resume. As of 3 May 2026 the Census Bureau's Business Trends and Outlook Survey put AI use among US businesses at 19.8%, rising to 37% among firms with at least 250 employees 2. The unit counted there is the firm rather than the worker, and most US firms are small.
An outright ban is the rarer version of the story.
The Pew Research Center surveyed 5,273 employed US adults from 7 to 13 October 2024. A majority, 55%, rarely or never used AI chatbots at work, and among workers who were not chatbot users 36% gave as a major reason that there was no use for the tool in their job 1.
An employer prohibition came last on the same list, at 9% 1.
But both of those reasons point the same way. A blank line under AI can record where somebody worked, and the posting is where you decide whether that record ends the application.
Many candidates read a requirement list as a rule about eligibility, not as a description of the work.
Tara Sophia Mohr, writing in Harvard Business Review in 2014, asked over a thousand mostly American professionals why they had not applied to a job whose qualifications they did not fully meet. The top answer was that they did not believe they would be hired without the qualifications and did not want to spend the effort finding out. It was twice as common as any other answer: 41% of women, 46% of men 4.
Doubt about doing the job well was the least common answer on that list.
Name the substitutes in the posting, not in the debrief
Say what you will take, in the posting, in one line, or the equivalence clause does nothing. Three substitutes are worth considering. A project the candidate can walk through decision by decision, a task they used to do by hand that a model now drafts, and one occasion when a model handed them something wrong and they caught it.
- A project they can walk through decision by decision. A class assignment, a volunteer build, a household spreadsheet rebuilt from scratch. Size does not matter; the account of the choices does.
- A task they used to do by hand that a model now drafts. Somebody who did the work manually knows what the good version looks like, which is the only reason they can tell when a draft is wrong.
- One occasion when a model was wrong and they caught it. Ask what tipped them off, and what they checked it against.
A person whose employer never bought a seat can produce all three, because none of them needs a corporate account. Each one leaves somebody with decisions they can account for.
Two substitutes are enough. A list of five reads as a wish list, and a screener working a long stack will use the first line and skip the rest.
Put the line where the requirement is, not in a paragraph underneath it.
How far is it from an or-equivalent line to a screener's decision?
A clause the screen ignores changes nothing, and one large measurement of that gap was made on degrees. The Burning Glass Institute and Harvard Business School matched 11,332 roles at large firms that had dropped a degree requirement against roughly 65 million US career histories. About 45% of those firms showed no meaningful difference in who they hired afterwards 3.
The firms that moved had changed something besides the text. About 37% of all the firms studied raised their share of workers hired without a bachelor's degree into the analysed roles by nearly 20%, a relative rise and not 20 points, and about a fifth made short-term gains that did not stick 3.
Across the whole population the net change came to 0.14 percentage points, which the authors put at fewer than 1 in 700 hires that year 3.
Degrees are not AI experience, and the study tracked requirements that were dropped outright.
What the study measures is the distance between a sentence in a posting and who actually gets hired, and crossing that distance is the entire job of an or-equivalent line.
So the second half of the job happens at the screen. Skills-based hiring without a skills test is the general form of the same failure: remove the proxy, then never add a way to observe the thing it stood for.
If your own staff are not cleared to use the tool, Hiring for AI you have not approved internally is the condition to settle before the req opens, because the posting would publish a contradiction and a strong candidate may ask about it in the screen.
What has to come back before a substitute counts?
A substitute counts when it produces the evidence the job line was standing in for: a decision the person made, a source they checked before relying on it, and an output they refused. Job history was never the thing you wanted, only a cheap way of assuming those three had happened somewhere, and it stops working wherever the employer never adopted the tool.
Put a stage behind each substitute before the posting goes out.
Write it as two columns: the requirement on the left, the place in the process that will produce evidence for it on the right. Any line with an empty right column comes off the list.
Nothing about that test is special to AI. Every requirement needs a stage that tests it is the same discipline across a whole req. The AI line is only where it costs the most, because the stage has to produce evidence from somebody whose employer never bought a seat.
For a project, the stage is twenty minutes on the candidate's own decisions. Ask which claim the work rests on, what they checked it against, and what they threw away.
What you are listening for is specifics on those three questions, and a follow-up when the first answer stays general.
A published substitute is also a published script. Watch what the line does to those answers over a quarter: once every candidate offers the same substitute in the same words, everyone passes and the question has become a password.
So keep the twenty minutes on the decisions rather than on the list. A candidate can rehearse which substitute to offer, but not what they checked it against and what they threw away.
Read the AI line again in six months, the way a candidate will
A requirement policy decays unless somebody re-reads the postings it lives in. Between 2017 and 2021 the Burning Glass Institute measured the share of IBM's IT postings specifying a degree rising from 21% to 29% 6. The report still put IBM well ahead of its peers at stripping degree requirements from postings, and alone among large technology firms in how far it had acted on its public commitment.
Nothing in that measurement says the policy was abandoned. It says a stated rule about requirements does not hold itself in place, and that postings are written by many hands over years.
Writing the substitutes out carries one serious objection: specificity does not execute itself.
In a secondary analysis of the Behavioural Insights Team's 2022 online experiment with 10,468 working-age participants, the advert that spelled out required behaviours drew less willingness to apply than the vaguer version built on high-level skills 5.
Readers did not rate the specific version as any clearer than the others. The authors record the result as a failure of their own wording and ask for replication in real hiring.
Open the requisition again and read the AI line the way a candidate will.
Somebody who spent four years somewhere the tool never arrived sees one of two things: a door with a stated second entrance, or a door. Six months of screen notes will tell you which one you wrote.
Common questions
What can a posting accept instead of on-the-job AI experience?
Name two or three things and keep them concrete. A project the candidate can walk through decision by decision. A task they used to do by hand that a model now drafts for them. One occasion when a model produced something wrong and they caught it before it went anywhere. All three are available to somebody whose employer never bought a seat. Put the line beside the requirement rather than in a paragraph below it, and make sure the screen asks for the same three things.
Is "or equivalent experience" enough on its own?
No, because nobody is told what the clause will accept. An undefined or-equivalent line leaves each screener to invent a standard, which can reduce to the employer names on the resume, and the requirement then filters as though the clause were absent. A candidate cannot tell a real equivalence from a decorative one, so some of the people it was written for may never apply. If the words are not followed by two or three named substitutes, the line is doing no work.
Does a personal project count as equivalent AI experience?
Count a personal project when the person can account for the decisions inside it. A build with no defensible choices behind it is a screenshot. The same build becomes evidence when they can say what the model got wrong, what they checked it against, and what they threw out. Scale is not the test: a rebuilt household spreadsheet with one caught error in it carries more than a large project the candidate can only describe in general terms.
How do you stop screeners from ignoring the equivalence line?
Give screeners the line as a scoring question rather than as permission. A screener who has to record which substitute a candidate offered, and what came back when they were asked about it, cannot quietly fall back on the employer name at the top of the resume. Audit a sample of rejections after the first month against the published line. A clause nobody was ever asked to apply is the ordinary way an equivalence line dies.
What if a candidate's last employer banned AI outright?
Ask what the employer's ban actually covered, and what the person did about the work underneath it. A prohibition can be narrower than it sounds, such as no external models on client data or no unapproved tools on the network. Somebody paying attention still knows which tasks in their week a model would take, where the risk sat, and what the approved path was. In a Pew Research Center survey fielded in October 2024, among workers who had not used AI chatbots at work, an employer prohibition was the least-cited of the four major reasons offered.
Does naming substitutes widen the applicant pool?
Naming substitutes does not widen the pool on its own, and the evidence on requirement wording is mixed. A secondary analysis in the Behavioural Insights Team's 2022 experiment found the advert written as specific behaviours drew less willingness to apply than a vaguer one, and readers did not find the specific version any clearer. What the named list does reliably is give the screen something to apply and give the candidate something to bring. Treat pool size as a separate measurement, taken from applications rather than assumed from the wording.
References
- 1. Workers' experience with AI chatbots in their jobs ✓ pewresearch.org Supports the claim that an employer ban is the rarest of the stated reasons workers have not used AI chatbots at work: among non-users, 36% named having no use for the tool in their job as a major reason and 9% named their employer not allowing it, in a survey of 5,273 employed US adults conducted 7 to 13 October 2024.
- 2. Large Firms With at Least 20 Employees Biggest AI Users ✓ census.gov Supports the claim that most employers have no AI tool in the building at all: 19.8% of US businesses reported using AI as of 3 May 2026, rising to 37% among firms with at least 250 employees. The unit is the firm, not the worker.
- 3. Skills-Based Hiring: The Long Road from Pronouncements to Practice ✓ static1.squarespace.com Supports the claim that editing a requirement out of a posting does not by itself change who gets hired: about 45% of firms that dropped a degree requirement showed no meaningful hiring difference, 37% classed as leaders raised non-degreed hiring in the analysed roles by nearly 20%, and the net national effect was 0.14 percentage points, fewer than 1 in 700 hires.
- 4. Why Women Don't Apply for Jobs Unless They're 100% Qualified ✓ web.archive.org Supports the claim that candidates read a requirement list as an eligibility rule rather than as a description of the work: the top reason for not applying, twice as common as any other, was not expecting to be hired without the qualifications (41% of women, 46% of men), while doubting they could do the job was the least common.
- 5. Gender differences in response to requirements in job adverts: Research report ✓ bi.team Supports the complication that writing requirements as specific behaviours does not automatically read as clearer: in an experiment with 10,468 participants the specific-behaviours advert drew less willingness to apply than the high-level-skills version and failed its own clarity manipulation check. Secondary and exploratory analyses; the authors call for replication.
- 6. The Emerging Degree Reset: How the Shift to Skills-Based Hiring Holds the Keys to Growing the U.S. Workforce at a Time of Talent Shortage ✓ burningglassinstitute.org Supports the claim that a stated policy on requirements decays unless the postings are re-read: IBM's share of IT postings specifying a degree rose from 21% in 2017 to 29% in 2021, while the report still called IBM well ahead of its peers and alone in how far it had acted on its public commitment.
6 sources, numbered by first appearance. Every one was opened and checked against the claim it carries. 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.