Set it up as a pipeline where AI drafts and organises and people decide. Write the role brief and scorecard first, draft the advert with AI, collect every application in one place, use AI to summarise each against your written criteria (never to reject automatically), run structured interviews, and keep a record of each decision and why it was made.
The line that matters is between help and judgement. AI is excellent at turning notes into an advert, summarising 140 applications into comparable one-pagers and drafting candidate emails. It shouldn't score, rank or reject people on its own, for two reasons. It can reproduce biases in ways that are hard to spot, and recruitment is a legally sensitive use of AI: data-protection law such as the GDPR restricts purely automated decisions about people, and if you recruit in the EU, the EU AI Act classes AI used in recruitment as high-risk, with those obligations due from 2 December 2027. For a small team, the practical rule is simple: a person reads every application before it's turned down.
Seven stages and where AI helps in each
| Stage | AI does | A person does |
|---|---|---|
| 1. Role brief and scorecard | Turns rough notes into a structured brief; suggests criteria | Decides what the job really needs |
| 2. Advert | Drafts it from the brief; checks wording | Edits for accuracy, pay and tone |
| 3. Applications | Acknowledges receipt; files documents | Designs the form; owns the inbox |
| 4. Screening | Summarises evidence against each criterion | Reads summaries and originals; decides the shortlist |
| 5. Interviews | Drafts questions, scorecards and scheduling emails | Interviews, scores independently, debriefs |
| 6. Work sample and references | Drafts the task brief and reference questions | Marks the task; speaks to referees |
| 7. Offer and onboarding | Drafts the offer email and a first-fortnight plan | Makes the offer; checks contract terms |
Two stages have their own tutorials in this series: writing job descriptions with AI for stage 2 and building interview questions and scorecards for stage 5. This one covers the process that joins them up.
Stage 1: a role brief before anything else
Every later stage uses the brief: the advert is written from it, screening is measured against it, interview questions come from it. Spend an hour on it. Here's an illustrative brief for a nine-person web design studio hiring a junior front-end developer:
ROLE: Junior front-end developer (full-time, hybrid, 3 days in studio)
PAY: $42,000-$48,000 depending on experience
WHY THE ROLE EXISTS: two senior developers spend ~30% of their time on
small site updates for care-plan clients; this role takes most of that.
IN THE FIRST SIX MONTHS THEY WILL:
- Handle routine care-plan updates (content, layout fixes, plugin updates)
- Build 2-3 small brochure sites from our design files with review
- Join the support rota for non-urgent tickets
MUST HAVE (screening criteria):
1. Can build responsive layouts in HTML and CSS from a design file
2. Working JavaScript for everyday front-end tasks
3. Has shipped at least one real site (paid, voluntary or personal)
4. Writes clearly to non-technical people (clients email us directly)
NICE TO HAVE (not screened on): experience with our CMS, accessibility
testing, any agency experience.
NOT REQUIRED: a degree, a specific number of years.Note the "not required" line. Written down, it stops the advert, the screening prompt and the interviewers from quietly adding requirements. Keep the must-haves to four or five; each one becomes a screening criterion and a set of interview questions.
Stage 2: one advert, adapted for each place it goes
Draft the advert from the brief, then have AI adapt it for each channel rather than posting the same block of text everywhere. A job board listing needs the title, pay and must-haves in the first few lines because that's what appears in search results; your own careers page can carry more about the team and how you work; a post in a developer community or a local business network should read like a person talking, not a listing. One prompt handles it: "Adapt this advert for [channel], keep every fact identical, change only length and tone."
Some job platforms now draft adverts for you. LinkedIn, for example, offers a "Write with AI" option when posting a job, working from the title, company and job type, though it isn't available on every account. Those drafts are generic by design, because the platform knows nothing about the role beyond its title. Paste in your own advert written from the brief instead, or use the platform draft only as a checklist of anything you forgot.
Whatever the channel, check three facts survive every version: the pay range, the working pattern (hybrid days, hours) and the must-haves. When AI shortens an advert, the pay range is the thing it most often drops, and it's the thing candidates most want to see.
Where applications go: a form beats an inbox
Applications arriving as email attachments in someone's inbox are hard to track and easy to lose. Use an application form (any form tool will do) that asks for a CV upload, a portfolio link where relevant, and two or three short questions tied to the must-haves, such as "Tell us about a site you built: what did you do, and what would you change?". Answers to questions like that tell you far more than a CV, and they give the AI something specific to summarise.
Whether you need an applicant tracking system (ATS) depends on volume:
- One to three hires a year: form, spreadsheet, shared folder and a business AI assistant. Nothing else.
- Several hires a year, or hundreds of applicants per role: an ATS starts to pay for itself. Some have free tiers; Zoho Recruit's free corporate plan, for example, allows one active job per recruiter licence.
- Continuous hiring: a full ATS with AI features. Workable's entry plan was listed at $299 a month for companies of 1-20 employees when checked in September 2026, with AI credits included and its screening agent as an add-on. Read AI recruiting tools for small businesses before paying for one.
Whichever route, use a business plan for any AI that touches candidate data, not a personal account. ChatGPT Business and Claude Team both need at least two seats; a sole owner hiring once may find Microsoft 365 Copilot or Gemini in an existing Workspace plan simpler.
Screening: summaries against criteria, not scores
This is where AI saves the most time and where it needs the tightest rules. Ask for evidence, not verdicts:
You are helping a hiring manager review applications for the role below.
For this application, go through each MUST HAVE criterion and:
- quote the evidence from the application (CV, answers, cover note)
- or write "No evidence found in the application"
Do NOT give scores, rankings, or a recommendation.
Do NOT comment on: name, age, gender, nationality, photos, address,
career gaps, which school or university, or writing style beyond
criterion 4.
Then list anything the reviewer should check that you couldn't read
(links, attachments that failed to open).
Role brief: {{brief}} Application: {{application text}}An illustrative summary for one applicant:
Candidate 047
1. Responsive layouts: "Rebuilt my local football club's site from a
Figma file; mobile-first CSS grid" (answer 1). Portfolio link given.
2. JavaScript: "Added a fixtures filter in vanilla JS" (answer 1). CV lists
a JavaScript course.
3. Shipped site: football club site, live (CV and answer 1).
4. Writing for non-technical people: answer 2 explains a fix to the club
committee in plain terms - quoted in full below.
Couldn't check: portfolio link (not opened).What the reviewer does next: opens the portfolio (the AI couldn't), and reads the original application for anything borderline. A summary like this takes about a minute to review, against five or six minutes to read a CV and answers cold.
A realistic screening mistake
In an early run without the "don't comment on career gaps" rule, the AI wrote "limited recent experience: no roles listed for 2023-2025" for a candidate who had spent two years as a carer and built three sites for local charities in that time, listed further down the CV. A reviewer skimming summaries might have passed over them. The fix was the rule in the prompt, plus the habit below. Checks for this kind of problem are covered in bias checks for AI CV screening.
The rule that keeps screening fair
A person reads at least the original answers, not just the summary, for every application they decide to reject. With a good form this takes a minute or two each. It's the single safeguard that most reduces the risk of AI quietly filtering people out, and it's what lets you say honestly that people make every decision.
When applicants use AI too
Expect a good share of applications to be written with AI. That isn't a reason to reject anyone; the role may well involve using AI. But it changes what your form can tell you. Generic questions get generic, polished answers that all sound alike. Compare two illustrative answers to "Why do you want this role?":
A: "I am passionate about creating engaging, user-centred digital
experiences and I am excited by the opportunity to grow my skills in a
dynamic, collaborative studio environment."
B: "I've been updating my aunt's bakery site for two years and the part
I enjoy is the fiddly stuff - getting the menu to stop breaking on
phones. A care-plan role would be that, every day, with people to learn from."Answer A could come from any of 50 applicants; answer B could only come from one. Design questions that ask for specifics AI can't supply without the candidate's own experience: a real project, a mistake they fixed, a decision they'd make differently. Then let the screening prompt quote what they wrote. The specific answers stand out in the summaries on their own, without you needing to guess who used AI.
The work sample and the interview are where polish stops helping. A candidate who wrote a flawless application with AI and can't talk through their own project will show it within ten minutes, which is another reason not to over-weight the written stage.
Interviews, work samples and references
Once you have a shortlist of five to eight, the process becomes more human, and AI moves to preparation:
- Scheduling: send a booking link for pre-set slots rather than emailing back and forth. AI can draft the invitation, including what the interview involves and how long it takes.
- Structured interviews: the same core questions for every candidate, drawn from the must-haves, with a scorecard each interviewer fills in alone before discussing. The detail is in the scorecards tutorial linked above.
- A short, paid or time-boxed work sample for roles where it's fair: for the junior developer, "fix these three layout bugs on a test page, up to 90 minutes". AI can help write a clear brief and a marking guide; a developer marks it.
- References: AI drafts questions tied to the must-haves ("How did they explain technical problems to clients?"). Speak to referees yourself.
Keep AI out of the judgement: don't feed interview recordings to a tool and ask who performed best. If you record interviews for note-taking, get the candidate's consent first, and don't use any tool that claims to read emotions or personality from video.
Candidate emails: fast, and checked by a person
Candidates judge you on communication, and small employers are often the worst at it because nobody owns it. AI makes it easy to be good. Prepare templates at the start:
- Acknowledgement (automatic on form submission): thanks, what happens next, when they'll hear.
- Not progressing after screening (sent by a person, in batches): short, kind, no reasons unless you can give useful ones.
- Interview invitation: format, length, who they'll meet, booking link.
- Not progressing after interview: personal, with one or two pieces of specific feedback.
The post-interview email is where AI helps most and where it needs checking. An illustrative draft from the interviewer's scorecard notes:
Hi [first name], thank you for coming in on Tuesday and for the work on the layout
task. We've decided to offer the role to another candidate whose client-
facing experience was a closer match for how much of this job is spent
emailing clients directly. Your CSS work on the task was the strongest we
saw; if you can, build confidence explaining technical fixes in plain
language, as that was the gap for this particular role. We'd be glad to
hear from you if we advertise again.What you'd check: that the feedback is about the criteria, not comparisons that reveal anything about the other candidate, and that nothing in it touches age, background, health or anything unrelated to the job. This draft passes. An earlier draft said "another candidate with more years in the industry", which quietly reintroduced a requirement the brief had ruled out; the interviewer removed it.
A nine-person studio's hire, start to finish
Here's the studio's hire laid out end to end, with illustrative numbers.
Last time, without a process: about 90 applications by email, the owner reading each CV (around 6 minutes each, 9 hours), three weeks to a shortlist, and two strong candidates lost because they'd accepted other offers by the time the studio called.
This time:
- Day 1: role brief (1 hour), advert drafted by AI and edited (45 minutes), posted.
- Days 1-14: 140 applications through the form, each acknowledged automatically.
- Days 8 and 15: screening in two batches, so early applicants didn't wait. AI summaries for all 140 (about 20 minutes to run in batches); the owner reviewed each summary side by side with the candidate's own form answers (about 2 minutes each, roughly 4.5 hours) and opened the portfolios of the 30 closest calls (about an hour).
- Day 16: shortlist of seven; invitations sent with a booking link.
- Days 18-22: 45-minute structured interviews with two interviewers; five candidates invited to the 90-minute work sample.
- Day 24: offer made; the rest told the same day, with feedback for the five who did the task.
About 6 hours of screening against 9 last time, for half as many applications again, and 24 days from advert to offer instead of about six weeks. The AI cost nothing extra: the studio already had a Team plan. The bigger gains were speed and consistency: every applicant was assessed against the same four criteria.
Telling candidates, keeping records, deleting data
A short line in the advert or the privacy notice is enough for most small employers. Illustrative wording:
How we use AI in hiring: we use an AI assistant to help summarise
applications against the criteria in this advert and to draft some of
our emails. A member of our team reads every application and makes
every decision. If you'd rather your application wasn't processed this
way, tell us when you apply and we'll review it entirely by hand.Keep a simple record for each role: the brief, the advert, the screening prompt you used, the shortlist decisions with a one-line reason, the scorecards and who made the offer decision. If a candidate ever asks how they were assessed, or challenges a decision, this is your answer.
Decide how long you keep unsuccessful candidates' data and stick to it; a fixed period of a few months is common, but check what applies to you with an HR or data-protection adviser. Delete the AI tool's conversation history for screening sessions at the same time if the tool keeps it. For the underlying rules on candidate data in AI tools, see AI bias in small business decisions and the data-protection tutorial listed below.
Signs your process is filtering out good people
Small teams can't run statistics on a single hire, but they can run spot checks:
- Read ten random rejections in full after each role. If you'd have shortlisted any of them on reading the original, your criteria or prompt are too narrow.
- Compare "No evidence found" with the original for five applications. If the evidence was there in a different form (a portfolio, a voluntary project, a different job title), add examples to the criteria.
- Watch the source of your shortlist. If everyone shortlisted came from one kind of background or one job board, look at where the advert went and what the criteria reward.
- Ask the hire, three months in, what nearly put them off applying. The answer usually improves the next advert.
Once the offer is accepted, the process hands over to onboarding; planning a new starter's first two weeks with AI picks up from there.
Hiring-with-AI questions small employers ask
Do I have to tell candidates that I'm using AI?
Telling them is good practice everywhere and required in some places, particularly where data-protection or AI rules apply to recruitment. A short, plain line in the advert or privacy notice explaining what the AI does and that people make every decision is enough for most small employers. Check the specifics for where you hire with an HR or legal adviser.
Can I let AI reject obviously unsuitable applications automatically?
I wouldn't. Automated rejections are where bias and errors hide, and data-protection law such as the GDPR restricts decisions made solely by automated means that significantly affect people. Let the AI summarise, then have a person read every application you reject, even briefly. For a small team the time cost is modest.
Which AI tool should a small team use for hiring?
For occasional hiring, a business plan of a general assistant (ChatGPT Business, Claude Team, Gemini in Workspace or Microsoft 365 Copilot) plus a form and a spreadsheet covers everything. Consider an applicant tracking system once you're hiring several roles a year or getting hundreds of applications per role. Avoid consumer accounts for candidate data.
Should we use AI video-interview tools that analyse candidates?
Be very cautious. Tools that claim to read personality or emotions from video are hard to validate and easy to challenge. If you recruit in the EU, the EU AI Act has banned AI that infers emotions in the workplace, including recruitment, since February 2025. Recording an interview for notes is a separate question that needs consent.
Further reads
- AI or a New Hire? How to Decide Before You Recruit — Check the role is really a hire before you start recruiting.
- How to Evaluate the AI Features in Your Recruitment Software — Questions to ask before trusting an ATS's AI screening.
- How to Test Job Candidates' AI Skills in an Interview — Test how candidates use AI in the job itself.
- How to Write Job Adverts With ChatGPT Without Biased Language — Strip biased language from the advert before it goes live.
- GDPR and AI Tools: What a Small Business Must Do — The data-protection basics for candidate data in AI tools.
- How to Hire Your First Customer Service Person Alongside AI — A worked hire where the role is shaped around AI.
- Employment Contracts With AI: Drafting Steps and Legal Checks — Draft a clear employment contract with AI from a job facts sheet, then send the right clauses to a lawyer instead of paying for the whole thing.
- Can AI Screen CVs Fairly? What Small Employers Need to Know — How AI CV screening goes unfair, what the rules expect of small employers, and a six-step set-up with tests you can run in an afternoon.
- How to Write a Job Advert for an AI-Savvy Admin Assistant — A bakery's admin advert built line by line: which AI skills to ask for, wording that attracts the wrong people, and a task that shows who checks their work.
- Questions to Ask a Recruitment Agency About Its AI Screening — Stage-by-stage questions on an agency's AI screening, how to grade the answers, and how a grooming salon found good candidates its agency's filter had rejected.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Workable pricing page; Zoho Recruit pricing page; LinkedIn Help on AI-assisted job descriptions; AI Act service desk page on Article 5 and the Commission's guidelines on prohibited practices; project fact sheet on EU AI Act timelines. Checked September 2026.