AI for Job Descriptions, Screening Notes and Interview Questions

Hiring paperwork in a business without HR

In a 30-person engineering firm, hiring falls to whichever manager needs the person. They write the advert from an old one, skim 80 applications between other work, take patchy notes in interviews and struggle to explain afterwards why one candidate got the job. None of this is negligence. It is hiring done in the gaps of a busy week.

AI can take a lot of the writing load off, and done carefully it can make hiring fairer, because structure is what makes hiring fair and AI is good at producing structure. Done carelessly, it automates bias at speed. The difference lies in which tasks you hand over.

A useful way to divide the work: AI writes and organises, people read and decide. Drafting an advert, tidying notes and preparing questions are writing tasks. Choosing who to interview, who to reject and who to hire are decisions about people's livelihoods, and they should rest with someone who can explain their reasoning to the candidate, to a tribunal if it ever came to that, and to themselves.

Job descriptions that attract the right people

Give the assistant the real substance and ask it to write, rather than asking it to invent a role.

• What the person will actually do in their first six months, in plain terms
• The genuinely essential skills, separated from nice-to-haves
• Salary range, location, hours and flexibility, stated honestly
• What the team and working environment are really like

Then ask it to review the draft for jargon, unnecessary requirements such as degrees for roles that do not need them, and gender-coded or age-coded language like 'digital native' or 'young, dynamic team'. Models are good at spotting this kind of wording, and removing it widens the applicant pool.

Watch for the opposite problem. Unedited AI job adverts read like every other advert: 'fast-paced environment', 'passionate self-starter'. Specific detail about the job attracts better applicants than polished generalities.

Screening notes and structured records

Where AI helps most in screening is in organising the reviewer's own judgement, not replacing it.

1. Agree three to six criteria from the job description before reading any applications
2. The reviewer reads each application and jots quick notes against those criteria
3. The assistant turns rough notes into a consistent record, with evidence for each criterion
4. The reviewer checks the record and makes the decision

This keeps a person reading every application while producing records you can compare fairly and explain later if a candidate asks. The same approach works for interview notes: rough notes in, structured summary against the criteria out, checked by the interviewer the same day.

Interview questions with scoring guides

Structured interviews, where every candidate gets the same core questions scored against the same guide, are widely regarded as fairer and more predictive than free-flowing conversations. Small businesses rarely use them because preparing them takes time. AI removes most of that effort.

Review the questions for anything that strays into protected characteristics, such as family plans, age or health, and make sure the exercise is proportionate to the role.

Where AI in hiring becomes risky

The riskiest use is also the most tempting: asking AI to rank or reject CVs. The problems are well documented.

• Discrimination. Models can penalise career gaps, non-standard education or names and wording associated with particular groups. Under the UK Equality Act and similar laws elsewhere, the employer is responsible for discriminatory outcomes regardless of which tool produced them.
• Automated decision rules. UK and EU data protection law restricts significant decisions made solely by automated means and gives candidates rights to information and human review.
• The EU AI Act classifies AI used to screen, rank or evaluate job candidates as high-risk, with obligations for deployers, including human oversight and transparency, phasing in.
• Unexplainable rejections. If a candidate asks why they were rejected and the honest answer is 'the tool scored you low', you have a problem.

Use AI to help people make hiring decisions more consistently. Do not use it to make the decisions.

Our longer post on using AI for recruitment carefully covers the governance side in more depth.

Candidate data and practical safeguards

Applications are personal data, sometimes including sensitive information. Use an approved business AI tool with appropriate data terms, not personal accounts. Tell candidates in your privacy notice how AI is used in your process. Delete applications and notes in line with your retention policy.

When SpiderHunts helps an employer bring AI into hiring, we build around the reviewer: criteria defined first, AI structuring notes and drafting questions, every decision recorded by a named person. That work usually sits within our AI integration service, and we will decline to build fully automated rejection because it is rarely defensible for a small employer.

Frequently asked questions