How to Hire Cleaners for Your Cleaning Business: Ads and Screening

Hire reliable cleaners with a specific job ad, a short application form that scores every applicant on the same few questions, a ten-minute call, a paid trial shift and two reference checks. AI can draft the ad and summarise answers, but a person makes every decision, and any AI step that reads applications needs a bias test first.

Most cleaning owners lose good applicants at two points: an advert so vague that the wrong people apply, and a slow reply that lets the right ones take another job. The sequence below fixes both and gives you something you can score. Employment rules differ by place, so the legal points are flagged where they matter, and the details belong with a qualified adviser.

Follow me on Instagram@sagnikteaches

Decide the job before you write the ad

An advert can only be as specific as the job behind it. "Cleaner" covers at least four different jobs, and the people who suit one often do not suit another.

Connect on LinkedInSagnik Bhattacharya

• Regular domestic round: the same homes each week, weekday daytime hours, steady and repetitive, trust with keys.
• End-of-tenancy and deep cleans: irregular days, heavy physical work, tight deadlines, often in a pair.
• Commercial evenings: early or late hours, usually alone, alarms and access cards to handle.
• Holiday-let changeovers: weekend peaks and a hard finish time before the next guests arrive.

Pick one job per advert and fill in a six-line job card before you write a word: hours and days; pay and how often it is paid; travel (is travel between jobs paid, and how do people reach the first job); equipment and products (who supplies them); what good work looks like, in your own standard; and the start date. The advert, the form questions and the trial shift all come from that card.

Subscribe on YouTube@sagnikteaches

The example in this tutorial is an illustration: a domestic cleaning company with five cleaners and 62 regular clients that needs two more people for weekday mornings. The job card says Monday to Friday, 9:00 to 15:00, 20 to 25 hours a week, $16 to $18 an hour depending on experience, paid weekly, travel between jobs paid, products and equipment supplied, and two weeks working in a pair before going out alone.

Decide whether you are hiring employees or using self-employed subcontractors before you advertise. The rules on who counts as which differ from place to place, and the penalties for treating someone as self-employed when the law sees an employee can be significant. If you set the hours, direct how the work is done and supply the products, many places would treat the person as an employee. An accountant or employment adviser can settle this in one conversation, and that is cheaper than settling it later.

A cleaner job ad, before and after an AI rewrite

Here is the kind of advert that fills an inbox with the wrong people: "CLEANERS WANTED!! Energetic, reliable cleaners needed ASAP for busy cleaning company. Must have car. Good pay. Call now." It gives no pay, no hours and no process. "Energetic" can read as a hint about age, and "must have car" excludes people who could reach the job another way. Give the AI the job card instead of asking it to be creative.

Write a job advert for a domestic cleaning company hiring two cleaners.
Facts: Monday to Friday, start between 9:00 and 9:30, finish by 15:00, 20 to 25 hours a week; pay $16 to $18 an hour depending on experience, paid weekly; travel between jobs is paid; products and equipment supplied; first two weeks working in a pair with an experienced cleaner; a paid 3-hour trial shift before any offer; two references needed; we cover [districts]; apply by answering six questions at [link], which takes about five minutes.
Rules: plain words; no hints about age, gender or nationality; no "energetic"; no "must have a car" (ask how they would reach the job instead); no perks I have not listed. Under 220 words.

A reply, illustrative and trimmed:

Cleaners wanted: weekday mornings, domestic homes

We are a domestic cleaning company looking for two reliable cleaners to join a team of five. Hours are Monday to Friday, starting between 9:00 and 9:30 and finishing by 15:00, 20 to 25 hours a week.

Pay is $16 to $18 an hour depending on experience, paid weekly. Travel between jobs is paid. We supply products and equipment.

You will work in a pair with an experienced cleaner for your first two weeks. Before any offer we pay you for a 3-hour trial shift, so you can see the work and we can see how we work together. We ask for two references.

We cover [districts]. You need to reach the first job by 9:00; tell us on the form how you would get there.

To apply, answer six short questions at [link]. It takes about five minutes. A free uniform and a long-term career path are waiting for you!

The last line is the AI filling a gap. The facts mention no uniform and no career path, so both go, and so does the exclamation mark. Everything above it is usable. Check any AI-drafted advert against these seven questions before you post it.

1. Does it state the pay, the days and the hours?
2. Does it say what you supply, so people know what to bring?
3. Does it say what happens after someone applies: a call, a paid trial, references?
4. Could someone apply from their phone in five minutes?
5. Is there any wording that hints at age, gender, nationality or family status?
6. Does it describe a normal shift well enough that a stranger could picture it?
7. Is every promise one you will keep in the first month?

For a deeper test of the language, how to write job adverts with ChatGPT without biased language lists the phrases that quietly narrow your applicant pool. Post where cleaners already look: the large job boards, community groups for the districts you cover, your website, a line on the van, and your current team. Ask each cleaner you employ whether they know someone, and pay a referral bonus after the new person completes eight weeks, not on their first day. Check each job board's current rules and costs before you pay for anything.

The application form: six questions and a score for each answer

Cleaning applicants often have no CV, and a form puts every one of them through the same questions in the same order, which is what makes the comparison fair. It also takes five minutes on a phone, so you hear from people who would never write a formal application. Six questions are enough.

Questions 1 and 2 are pass or fail because this job has fixed hours and a fixed start; if yours is flexible, change them. Every question has to relate to the work, so cut anything you could not defend as job-related. Questions 3 to 6 give a score from 0 to 8, and in the example round anyone who passed 1 and 2 and scored 5 or more was invited to a call. Here are four scored forms, anonymised and illustrative.

Applicant 07 gave a concrete example of looking after a neighbour's flat, with the things they actually did. Applicant 19 wrote a vague answer to question 3 but handled the two scenario questions well, so a call is worth it. Applicant 31 left question 3 blank. Scoring stops at a failed practical question, so 12 is not scored, but you can ask whether they would like to be kept on file for a different shift.

Every applicant gets an answer within two working days, including the ones you decline, with a short, kind message. AI drafts the decline and next-step messages well; change them so each one is true. If you hold a few as reserves, say so plainly rather than leaving them waiting.

The time saving is real but modest. Reading 38 forms at about six minutes each is 3.8 hours. If AI first summarises each form into a few factual lines and you check the summary against the original answers, it takes about 2.5 minutes each, or 1.6 hours. How to do that without letting the AI choose is the next section. A wider set of scorecards is in how to build interview questions and scorecards with AI.

Using AI on the answers without letting it choose

Use AI to summarise and to flag gaps. Do not use it to rank, score or reject. Four habits keep that line clear:

1. Strip names, photos, ages, addresses and phone numbers before the AI sees anything. Label each form with an ID such as 07 and paste only the six answers.
2. Ask for facts, in quotes where possible, and tell it not to rank or recommend.
3. Score the answers yourself against the table first, then read the AI's summary to see what you missed.
4. Write down the reason for every decision in job-related terms ("cannot reach the first job by 9:00"), not impressions.

The prompt:

Below are one applicant's answers to six questions for a cleaner job. Summarise the facts in three bullets: availability, how they would reach jobs, and what they said about looking after other people's property. Quote their words where you can.
Do not rank, score or recommend. Do not infer age, gender, nationality, health, family situation or accent. Flag any question left blank or answered in under five words.

A typical reply, illustrative:

- Availability: Monday to Thursday, 9:00 to 15:00; not available Fridays.
- Travel: bus, then a 10-minute walk; says they would arrive by 8:50.
- Property: looked after a neighbour's flat for three weeks while they were away, watered the plants and kept the spare keys somewhere safe.
Seems enthusiastic and probably a parent, so school hours will suit. Blank answers: none.

The last two lines are the AI breaking the instruction. "Seems enthusiastic" is an impression, and "probably a parent" is a guess about family situation, which has no place in a hiring file. Delete both, or re-run with a firmer instruction. The first bullet matters more: the job card says Monday to Friday, so "not available Fridays" is a real gap against the requirement. Whether that is a deal-breaker or a negotiation is your decision, not the tool's.

Applicants' details are personal data. Consumer chat plans from the main vendors have a switch in their privacy settings that controls whether your chats are used to train models, and business plans do not train on your content by default. Check the setting you are on before you paste anything, and keep to anonymised answers either way. For a wider look at the screening side, bias checks for AI CV screening covers the same ground for recruitment agencies.

A bias test you can run in half an hour

If you remove names and personal details before the AI sees an application, much of the risk goes. You should still test your own setup once, because a prompt that looks careful can behave oddly and a name can slip into a pasted answer. Take two real application answer sets, one strong and one average, and run these three tests.

The repeat run comes first because it tells you how much the output varies when nothing has changed. A difference between the name-swapped versions only counts if it is bigger than that. If a test fails, tighten the prompt and run it again, narrow the AI's job to something simpler, or take that step back by hand. Note what you tested and when, and run the three tests again after you change the tool or the model, since both change often.

The ten-minute call and the paid trial shift

The call has three jobs: hear how the person talks about work, check the practical points the form could not, and let them ask questions. Keep it to 10 to 12 minutes and score three things from 0 to 2 afterwards: reliability signals (do the details of their working week hang together), honesty under pressure, and how they respond to being shown a different way of doing something. AI can draft scenario questions and the points a good answer includes, as long as you fence it in.

Suggest four scenario questions for a phone call with a cleaner applicant, for a domestic cleaning job where the cleaner holds a client's key. For each, give two or three points a good answer would include.
Questions must relate only to the job: nothing about family, health, age, background or plans to have children.

1. A client's ornament breaks while you are dusting. What do you do?
   Good answer: tells the client or the office straight away, describes it plainly, does not hide it or blame anyone.
2. You are given a key and an alarm code. How do you look after them?
   Good answer: keeps the key out of sight and unlabelled, never shares the code, reports a loss at once.
3. A client asks you to clean something you were not booked for. What do you say?
   Good answer: stays polite, checks with the office before agreeing, does not promise time you do not have.
4. You find the job is much bigger than the notes said. What next?
   Good answer: contacts the office, agrees priorities with the client, does not leave unannounced.

The fence matters. Without the last line of the prompt, AI drafts of call questions can drift into childcare, health or "where are you from", and none of that is job-related. Cut any question like that, and make sure every question you keep would be fair to ask of every applicant.

The trial shift is where you learn most. Make it three hours, paid at your normal rate, alongside a trusted cleaner. Run it at a client's home only if the client agrees in advance, or at an empty property or your own premises. Never leave a candidate alone in someone's home and hand over no keys or alarm codes at this stage; the tutorial on scheduling teams, keys and recurring visits covers how to hand keys over properly once someone is hired. The mentor scores the shift on their own, then you compare notes. The standard you score against should be your written one, and writing cleaning procedures and checklists with AI shows how to produce it.

In the example round, an offer needed 11 or more with no zeros, subject to references. Applicant 07 scored 12, and the mentor noted that the tap bases were missed and fixed at once when shown, which is exactly the sort of detail a number alone would hide. Keep the notes; they feed the first-week plan, and building a cleaning company training programme with AI shows how to turn them into training.

Reference checks that get a useful reply

Ask the candidate's permission first, and ask for two referees: the latest employer or a long-term client, and one other person who has seen them work. Email first, then phone if there is no reply in two days. Use the same five questions for every candidate, so the answers can be compared.

Subject: Reference request for [candidate first name]

Hi [referee first name],

[Candidate first name] has applied for a cleaner role with [business name] and gave your name as a reference, with their permission. Could you spare ten minutes this week for a call, or reply to the five short questions below?

1. What was their role, and over what dates?
2. How reliable were they with time and attendance?
3. How would you describe the quality of their work?
4. Were they trusted with keys or property, and how did that go?
5. Would you employ them again?

Thank you,
[Your name]

On a call, add two more: how did they handle feedback, and is there anything you should know to support them well. Many past employers will confirm only dates and job title, and that is normal, so do not treat a short reply as a warning on its own. The trial shift covers what a thin reference cannot. AI can draft the email and tidy your call notes into three lines. It cannot weigh what a referee left unsaid, so that part stays with you. Keep the notes with the application record, and ask your adviser how long records of unsuccessful applicants may be kept.

One hiring round in numbers: 38 applications, two cleaners

Here is the whole funnel for the example company. The figures are illustrative; they show the shape, and your own counts will differ.

Now the owner's time, in minutes, by hand and with AI doing the drafting and summarising only.

AI saved 173 minutes, just under three hours, and the job still took the owner about eight hours. The calls, the trials and the references are people work, and they are where the hiring quality comes from. The cash cost is the trial pay and the mentor's time: six shifts of three hours at $16 an hour, the bottom of the advertised range, is $288 for the candidates, plus 18 hours of a cleaner's time at $18 an hour, which is $324, so $612 in all and $306 per hire. If a hire leaves in the first month, you pay most of that again, which is the case for the trial and the references.

Rules that depend on where you employ people

Nothing above is legal advice, and the rules differ enough between places that a short conversation with an employment adviser or accountant is worth having before your first hire. These are the questions to put to them.

• Employee or self-employed: which one applies to the way you will work with cleaners, and what each means for pay, insurance and records.
• Trial shifts: how trial periods, and pay for them, are treated where you employ people.
• Right-to-work checks: what you must check, how, and what you must keep.
• Background checks: what is lawful for someone who will hold keys, what your insurer expects, and whether the candidate must consent.
• Pay information: whether you must state pay in the advert, and what you must put in writing on the first day.
• Record-keeping: how long you may keep applications from people you did not hire, and how you delete them.

Automated screening has rules of its own. Some places require employers to tell candidates before an automated tool is used on them, and some require an independent bias audit of the tool. Keeping AI to drafting and summarising, with a person making every decision, is the safest position, but whether a particular product counts as an automated decision tool is a question for your adviser, not for a vendor's marketing page.

If you recruit people in the EU, the EU AI Act matters here. Its Annex III lists AI systems intended for the recruitment or selection of people as high-risk, including systems that place targeted job advertisements, analyse and filter applications, or evaluate candidates. The Digital Omnibus on AI, in force since July 2026, pushed the high-risk duties for those systems back to 2 December 2027, so check the dates again before you rely on them. Separately, AI that infers people's emotions in the workplace has been banned since February 2025, and the Commission's non-binding guidance reads "workplace" widely enough to include recruitment. The practical rule is simple: do not use video or voice tools that claim to score a candidate's mood, confidence or personality.

Whatever the rules say, a plain line on the form is fair to candidates: "We use AI to help summarise written answers. A person reads every application and makes every decision." Add it only if it is true, and keep it true.

Questions cleaning owners ask when hiring

Should I hire cleaners as employees or subcontractors?

The rules differ by place, and the penalties for treating someone as self-employed when the law sees an employee can be large. Authorities often look at who sets the hours, directs the method and supplies the equipment. If you schedule people, supply products and expect set standards, many places would treat them as employees. Ask an accountant or employment adviser before you write 'self-employed' in an advert.

Can I let AI reject applicants automatically?

It is a poor idea for a small employer. Automatic rejection removes the human check that catches mistakes and bias, and some places require notice to candidates or an audit for automated screening tools. In the EU, AI used to filter or evaluate candidates is classed as high-risk. Use AI to summarise and draft, and keep a named person responsible for every yes and no, with a job-related reason written down.

Should the trial shift be paid?

Yes, as a default. Some places treat unpaid trial work as employment that must be paid, and a paid shift also attracts candidates who are already working elsewhere. Tell the candidate the length in advance (about three hours), pay your normal rate, never leave them alone in a client's home, and tell the client a trainee is coming. Ask your adviser how trial periods work where you employ people.

How many references should I ask for?

Two: one from the most recent employer or a long-term client, and one other person who has seen them work. Get the candidate's permission first. Many past employers confirm only dates and job title, which is normal, so do not treat a short reply as a warning on its own. The trial shift covers what a thin reference cannot tell you.

Further reads

• Lone Worker Risk Assessment for Cleaners: Early Starts and Client Sites — Protect new cleaners who work alone on early starts.
• Cleaning Business Management Software: Scheduling, Timesheets and AI — Scheduling and timesheets once the team grows.
• AI Bias in Small Business Decisions: Hiring, Pricing and Credit — A wider look at bias in hiring, pricing and credit.
• How to Get House Cleaning Clients: A Local Plan Built With AI — Fill the new cleaners' hours with regular clients.
• How to Use AI in Hiring for a Small Shop: Ads, Questions, Scorecards — The same ad and scorecard method for a small shop.
• Can AI Screen CVs Fairly? What Small Employers Need to Know — What small employers need to know about AI CV screening.
• Cleaning Business Marketing Ideas: A Month of Posts With ChatGPT — A filled-in ChatGPT brief, a 16-post calendar, honest before-and-after rules, offers that fit your capacity and a way to read the month.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: EU AI Act text (Annex III point 4; Article 5(1)(f)) and implementation timeline as published on artificialintelligenceact.eu; law-firm commentary on the Commission's guidance on prohibited practices. Rules on employment status, trial work, right-to-work and background checks, and automated screening tools differ by place: ask a qualified adviser. All worked figures are illustrative.