Before publishing an AI-drafted job description, ask the hiring manager to justify each requirement against the work the person will actually do. Separate what a new starter must already be able to do from knowledge the organization can teach. Then carry those decisions into the advert, application questions and selection process.
An assistant can turn a short brief into a fluent description. That fluency does not establish whether a degree, software package, experience threshold or availability requirement belongs in the role. Nor does removing awkward wording establish that the remaining requirements are necessary.
Our editorial recommendation is to review requirements as decisions about the job, before reviewing them as sentences. The worked example below is fictional. It illustrates a proposed drafting and review process, not a hiring experiment, a validated selection method or an independently tested AI product.
Faster posting is a different outcome from better hiring
In More, but Worse: The Marketplace Effects of AI-Generated Job Posts, Emma Wiles and John J. Horton study an online labor-market field experiment. New employers were randomly offered AI-written drafts when starting their first posting. The experiment took place in 2023; the author-hosted draft inspected for this article is dated 4 September 2026.
The authors report that access increased posting and reduced writing time, without a discernible increase in matches. They also present evidence that treated posts were more generic and less informative. This is evidence about a particular marketplace and intervention, not proof that all current assistants worsen recruitment or that every extra requirement is AI-generated.
The practical distinction is useful: completing an advert quickly is not the same as giving a candidate accurate information about the work. Our proposed review addresses that information problem. It has not been tested as a remedy in the cited experiment.
UK guidance on inclusive job descriptions, published on 4 March 2026, recommends reviewing long lists of must-haves, using specific behaviors and competencies, and distinguishing essential from desirable criteria. It also calls for reviewing AI outputs. This is guidance applying to England, Scotland and Wales, not a universal legal checklist or a guarantee of hiring outcomes.
Establish the work before asking for the description
Consider a fictional service-operations coordinator. The manager initially supplies only a title and asks an assistant for a professional job description. A draft might plausibly add a degree, several years in a particular industry, advanced database skills and broad availability. In this example those additions are invented possibilities, not outputs observed from a named model.
Instead, the manager prepares a short role brief describing the recurring work:
- Check an order-exception queue, identify the appropriate owner and follow an agreed escalation procedure.
- Compare order records with a reference spreadsheet and flag discrepancies for review.
- Prepare a daily handover showing unresolved items, owners and next actions.
- Work within the team’s confirmed coverage arrangements, with exceptions handled through the manager.
The brief should identify what is already decided and what remains open. Include reporting lines, decision authority, actual tools, available training and confirmed working arrangements. An old job description can be an input, but should not silently become the authority for today’s role.
Ask someone who understands the work to check this account. In a small business that may be the manager and a colleague who covers the queue; in a larger organization it may involve the role owner and HR. Do not claim that consultation happened merely because the assistant writes “reviewed by the team.”
Review the requirement, its reason and its timing
Use a short working table to examine additions before they become public promises or applicant filters. The following decisions are hypothetical and specific to the fictional role.
| Proposed requirement | What the work needs | Proposed review decision |
|---|---|---|
| Advanced SQL | Compare records using an approved spreadsheet; no database querying is assigned. | Remove unless the manager identifies actual querying duties. |
| Experience with the exact ticketing product | Route exceptions and keep handover records accurate. | Describe the transferable workflow; confirm that product training is available. |
| Five years in the industry | Recognize common exceptions and know when to escalate. | Ask what capability the duration is meant to establish; define that capability directly. |
| Excellent communicator | Write a handover that names the issue, owner and next action. | Replace the broad label with the observable task. |
The table is not a rule to remove every credential or experience requirement. Some roles have genuine professional, technical or regulatory prerequisites. The accountable manager and HR should establish those requirements from appropriate sources, rather than accepting or deleting them on the assistant’s authority.
Timing matters too. A skill can be necessary for the role without being necessary on the first day. If product knowledge is to be taught during onboarding, say so only after confirming the training exists. Conversely, “we can teach it” is not a sufficient reason to omit a capability when no one has committed time or support to teach it.
For every essential criterion, ask which task depends on it, why it is needed on entry and how the selection process could obtain relevant evidence. A criterion that cannot be explained may need more role analysis before it becomes a rejection rule.
Give the assistant a bounded editing assignment
Once the role brief and requirement decisions are approved, ask the assistant to draft from those materials. Specify that it should preserve the difference between essential, desirable and trainable capabilities; flag missing information; and list any proposed additions separately for review.
Do not authorize it to fill gaps with invented benefits, salary ranges, remote-working arrangements or career promises. “Make this attractive” should mean improving clarity about the confirmed opportunity, not changing the offer.
Review the draft against the brief rather than reviewing only its tone. Can an applicant tell what they will do? Has a desirable capability become mandatory? Has a transferable skill become a requirement to know one vendor’s software? Has a particular coverage window expanded into unrestricted availability?
The CIPD’s inclusive recruitment guide recommends objective criteria instead of selecting for cultural fit. Its discussion warns that similarity to existing employees can become a substitute for suitability. For the fictional role, describe the work behavior that matters rather than asking the assistant to find words for an undefined “ideal personality.” This is professional guidance, not validation of a particular automated screening approach.
Keep the advert and selection process aligned
A corrected advert can coexist with an older application form that still demands the removed qualification. A recruiter briefing or applicant-tracking template can preserve it too. Reviewing the public text alone would miss that inconsistency.
Before release, compare the approved criteria with required form fields, screening questions, recruiter instructions and interview materials. Retain an explanation of substantive changes so another reviewer can understand why an old requirement was removed, narrowed or retained. This is a proposed document-control practice, not permission to automate candidate rejection.
Where an assessment is needed, have the responsible people choose an appropriate, accessible way to examine the capability. The fictional handover task suggests the kind of evidence that might be relevant, but a sample exercise still needs proper design and review. An AI-generated rubric does not become a validated assessment merely because it has numbered score bands.
Measure the information candidates receive
For a first trial, choose one role and retain the source brief, draft, requirement decisions and published version. Record substantive corrections as well as drafting time. Did reviewers remove invented duties? Did candidates repeatedly ask about an arrangement the advert left ambiguous? Did a screening question contradict the approved criteria?
Use those observations to improve the process. Application volume alone cannot show whether the description represented the role accurately or whether selection became fairer. Keep any analysis of candidate data within the organization’s approved HR and privacy processes.
This approach complements task-level workforce planning: understand the work before turning a broad label into a consequential decision. For an AI-drafted job description, the immediate next step is small and concrete. Take one requirement, identify the task it supports and have an accountable person decide what evidence justifies keeping it.