Plumber Job Description: Write the Ad and a Practical Test With AI

Write the job description from the real working week: the jobs, the on-call duty, who owns the van and tools, the pay range and the checks you make. Then use AI to tidy the wording and strip biased phrasing, and pair the ad with a short practical test that every applicant sits and two people score against the same sheet.

AI earns its place at two steps and is risky at a third. It writes a clear ad quickly and drafts a test with a marking guide, but a qualified plumber must check the technical content, and it should never rank applicants. Keeping the scoring human is the fairness mechanism in this process, not an afterthought.

Follow me on Instagram@sagnikteaches

Start from the working week, not a template

The pages that rank for this search are mostly generic duty lists ("install, maintain and repair pipes") under a salary figure. Applicants skim them and move on, because every plumbing firm's list reads the same. What makes your ad different is the week itself: how many visits a day, what kind, who the plumber works with, when the phone rings at night, and what you provide. AI can only write that if you tell it, so start with a one-page brief.

Connect on LinkedInSagnik Bhattacharya

This is a filled-in brief for an illustrative five-van firm hiring its sixth plumber. Placeholders in square brackets are for you to complete.

Subscribe on YouTube@sagnikteaches

ROLE BRIEF (for the AI; no names or addresses)
Firm: five-van plumbing company, 12 people including office. Mostly domestic repairs and call-backs, bathroom replacements and some heating work; a few small commercial customers.
Role: qualified plumber, full time. Works alone on most jobs; paired with an apprentice on bathroom jobs.
A typical day: 4 to 6 visits, 20 to 40 minutes' drive between them, first job at 8 a.m., finish by 5 p.m. Jobs reach the app the evening before.
Weekly mix (last quarter): about 60% repairs and call-backs, 25% bathroom installs, 15% heating.
On-call: one weekend in five, Friday 5 p.m. to Monday 7 a.m. The on-call plumber triages by phone. Average [x] call-outs per on-call weekend over the last year; [y] after midnight.
Van and tools: company van, commuting use allowed; fuel card; own hand tools; company supplies press tool, drain machine, test pump and PPE.
Must have: recognised plumbing qualification; full driving licence; any registration required by law for unvented hot water work. Nice to have: gas registration.
Pay: [range] a year depending on qualifications; on-call allowance [amount] a weekend plus paid time on site.
How we hire: application, 20-minute phone call, paid practical test of about 45 minutes plus a short conversation, then an offer.
Adjustments: we ask every applicant whether anything would help them do the test.

The brief does three jobs. It stops the AI inventing duties (left alone it adds boiler servicing, water heater installs and "customer service excellence" that your firm may not do). It becomes the source for the test, so the test looks like the job. And it forces you to find out your own on-call numbers before an applicant asks for them.

The ad in eight blocks, in a fixed order

Applicants read an ad in a predictable order: is it worth reading, do I qualify, what are the hours like, what does it pay. Lay yours out the same way. If you also hire in heating and cooling, the HVAC technician job description tutorial applies the same ad-plus-test approach to that trade, and its structure is worth comparing with this one.

The pay line is where your ad can beat the salary pages that rank for this search, because they quote averages across every region and firm size. Set your range from your own payroll instead: the bottom is what you pay a newly qualified plumber now, the top is what your most experienced plumber earns, and the ad says what moves someone up it (an extra registration, running bathroom jobs alone, joining the on-call rota). Then read three or four live ads from the firms you compete with for staff, so you know whether your top figure would lose the people you want.

Give the brief to ChatGPT or Claude with an instruction that fences it in:

Write a job advert for a plumber from the brief below, in eight blocks: headline, opening, what you will do, what you need, hours and on-call, van tools and kit, pay and benefits, how we hire. Keep every fact, number and [placeholder] exactly as given. Do not add duties, benefits or qualifications that are not in the brief. Plain English, no hype, no slang, and speak to the reader as "you". Under 450 words. Offer three alternative headlines.

[paste brief]

Expect the draft to be tidy and slightly inflated. The habits to cut are always the same: "competitive salary" (you printed a range, so delete it), "fast-paced, dynamic environment", "passionate team player", and a benefits sentence the brief never mentioned. Check the numbers last. If the brief said one weekend in five and the draft says "regular weekend work", restore the exact wording, because that is the sentence an applicant will quote back to you.

On-call wording that gets honest applicants

On-call decides whether some of the best applicants apply, and it is the part firms describe most vaguely. Get the facts first. The number to find is how many call-outs an on-call weekend really produces, and AI can count it for you. Export the last year's after-hours calls (date, time and type; delete names and addresses), paste them into a chat, and ask:

From this list of after-hours calls, count the call-outs per weekend, how many were after midnight, and how many were genuine emergencies (burst pipe, no water, flooding, a blocked toilet with no other toilet). Give the average and the busiest weekend. If a row is unclear, list it separately and do not guess.

Check a sample by eye and work out the average yourself; chat tools can miscount rows or misread a call type. Then the ad can say "on average one or two call-outs a weekend, rarely after midnight", but only if your numbers say so.

Do a quick sum before you publish. With five plumbers on a one-in-five rota, each is on call 10 or 11 weekends a year (52 divided by 5 is 10.4). At an average of 1.5 call-outs a weekend, that is about 16 call-outs a year per plumber, if last year's pattern holds. "About ten weekends a year" is a better line for an ad than "regular on-call".

Also state what happens after a late night. A line such as "if a call-out ends after 11 p.m., you start at 10 a.m. the next working day, paid as normal" tells applicants you have thought about it. Rest and working-time rules differ by place, so check yours with an adviser before you promise anything.

Van and tools: say who owns what

An applicant who reads "van provided" fills in the gaps with their own hopes. State each arrangement on its own line.

If the plumber brings hand tools, list about ten items (spanners, mole grips, pipe cutters, hacksaw, files, tape, spirit level, torch and so on). A person with half a kit can then decide whether to apply instead of guessing.

Consider how one loose line costs a firm a good applicant. The ad says "must have a clean driving licence". An experienced plumber with three penalty points reads it, assumes they are out, and never applies, while the firm's van insurer would have accepted up to six points (illustrative figures). State the real rule: "a full driving licence; our van insurer's limit is [x] penalty points". The applicant can then judge, and you do not lose someone for a rule you do not have.

Qualification checks that do not turn into guesswork

Say what the work legally needs first, then what you would like. A firm doing taps, toilets and leaks needs less paperwork than one fitting unvented cylinders or working on gas. The ad should name the exact qualification and any registration required for the work the new plumber will do, not a loose phrase such as "fully qualified".

Keep must-haves and nice-to-haves apart, and be strict with yourself about what counts as a must. "Experience with [brand] boilers" in the must-have list screens out plumbers who would learn in a week. Put it under "nice to have", and train for it.

Then make the checks the same for everyone, in the same order:

1. On the application, ask applicants to state their qualification and any registration numbers.
2. Verify registrations with the issuing body's public register, or ask the applicant to confirm them through the body's own process.
3. At interview, see the original certificates and note the date you saw them. Keep a copy only if you have a reason and a rule for how long you hold it.
4. Driving licence: check through the licensing body's process, with the applicant's consent.
5. References: ask two referees the same three questions each.
6. Right to work: do whatever the law requires where you hire, and ask an adviser if you are unsure.

AI has no role in these checks except one: ask it to turn your checklist into the wording for the application form and the interview notes, so every applicant receives the same questions in the same words.

Words to cut: a before-and-after table and an audit prompt

Biased wording is rarely deliberate. It is inherited from ads the writer has seen. The table covers the phrases I would take out of a plumbing ad; the general case is covered in the tutorial on writing job adverts with ChatGPT without biased language, and the pay line has its own checklist in pay transparency job postings.

Keep a physical requirement only if it is part of the job and you can state it exactly, for example "regularly lift and carry up to 20 kg and work in lofts and crawl spaces". A vague "must be fit and strong" excludes people who could do the work with the equipment you supply, and it tells nobody what the job involves.

Ask AI to audit the draft, with a prompt that stops it rewriting your facts:

Review this job advert for wording that could put off qualified applicants for reasons unrelated to doing the job: gendered or age-coded words, unnecessary physical, licence or language requirements, slang, hostile phrases, and requirements that are really preferences. For each problem, quote the phrase, say why, and suggest a replacement that keeps the meaning. Do not change any fact, number or pay detail. Then list anything a reasonable applicant would still want to know and cannot find.

[paste ad]

An illustrative answer, on a draft written the old way:

1. "strong, hard-working lads" - gendered and age-coded; "strong" is not a skill. Suggest: "able to lift and carry materials safely using the equipment we provide".
2. "must have a clean licence" - may exclude applicants your insurer would accept. Suggest: state the actual limit.
3. "experience with [brand] boilers" is listed as essential - this reads like a preference. Suggest moving it to "nice to have".
Still missing: pay range; what a typical call-out looks like; who the plumber reports to.

The audit is useful and incomplete. It did not question a line saying "must lift 25 kg", because it cannot know whether your jobs need that; you have to decide whether the requirement is real. And if you ask it to "make the ad more inclusive", it tends to add a sentence about a "diverse and welcoming team". That sentence changes nothing for an applicant. The pay range, the stated adjustments and a test everyone sits do.

The practical test: a 45-minute bench task and a short talk

A test is fair when every applicant gets the same task, the same time, the same tools and the same marking, and when it looks like the job. It is not a trick. It tells you whether someone who says they can do the work can do it, and it gives you something to compare other than a feeling about who interviewed best. The structure here has three parts and takes about an hour.

The bench task (about 45 minutes). On a test board in your workshop with a capped water supply, the applicant fits an isolation valve and flexible connector to a mixer tap, joins a short run of pipe with two fittings in the material you use, fills the run and leak-tests it to your own written procedure. There is no gas work and no electrical work in a test, ever. Give everyone the same tools and materials. Tell applicants the format and the tool list beforehand, and hand over the task sheet at the start, because reading a job sheet is part of the job.

The fault-finding talk (about 10 minutes). Read two scenario cards aloud, with the same follow-up prompts for each applicant. For example: "A customer says the hot tap in the kitchen runs lukewarm, though the boiler is fine. What do you check, and in what order?" and "A toilet cistern keeps refilling. Where do you start?"

The customer scenario (about 5 minutes). "You have opened a wall and found corroded pipework that needs replacing. It adds half a day. The customer is at home. What do you say?" The same conversation is the written change order in the tutorial on plumbing change orders drafted with AI, and an applicant who handles it well is already good at it.

If the whole process runs past an hour, pay applicants for their time at the rate you would pay on the job. Unpaid trial days exclude people who cannot take unpaid time off, and in some places they raise legal questions, so ask an adviser before you use them.

Seven criteria at three points each make 21. The pass mark in this example is 15, with no zero on safety. Two assessors score alone first and then compare. If their scores on one criterion differ by two points or more, each explains from their own notes, and neither changes a score unless they spot a real error.

Candidate C has the second-best total and still fails, because the no-zero-on-safety rule is the reason the sheet exists. Candidate B is not a bad plumber. The sheet shows exactly where the gaps were, which gives you something concrete to say in the feedback you owe them.

Asking AI for the test, the marking guide and model answers

AI is quick at this and unreliable on technical detail, so the rule is simple: it drafts, a qualified plumber approves. Use the same brief as for the ad, so the test matches the work.

You are helping a plumbing firm design a fair practical test for a qualified plumber. Our role brief is below. Produce:
(1) a bench task of about 45 minutes on a test board with a capped water supply, with no gas and no electrical work, plus the list of materials and tools we must provide;
(2) two fault-finding scenario cards for a 10-minute conversation;
(3) one customer-communication scenario;
(4) a marking guide with seven criteria scored 0 to 3, describing a 3 and a 0 for each;
(5) a model answer for each scenario in numbered steps.
Do not invent any figure. Wherever a pressure, time or clearance is needed, write [confirm] so a qualified plumber can fill it in.

[paste role brief]

This is an illustrative excerpt of one scenario card and its model answer:

FAULT-FINDING CARD 1: The hot tap in the kitchen runs lukewarm; the boiler is working normally.
Model answer:
1. Ask whether other hot taps in the house are also lukewarm.
2. Ask whether the tap is a mixer and whether the problem started after any work.
3. Replace the cartridge.
4. Check the isolating valves are fully open.
5. Check for cross-flow between the hot and cold sides.

Five things a qualified plumber would change before this reaches an applicant:

1. The order. The draft puts "replace the cartridge" third, ahead of the cheap checks. Your plumbers would check the isolating valves and cross-flow first and spend money on parts last. The marking guide must reward how your team actually works, not how a chatbot imagines it.
2. Stray figures. The prompt asks for [confirm] wherever a number is needed, and models usually obey, but they sometimes slip a figure in anyway. Scan the whole output for digits and replace each with your own procedure's number or the manufacturer's instruction.
3. Materials you do not have. The draft's list included a fitting your test board is not set up for. Adjust the task to the board, not the board to the task.
4. The timing. "45 minutes" is a guess. Put one of your own plumbers through the task with a stopwatch and add a quarter for nerves.
5. Customer wording. If the model answer to the customer scenario uses "stopcock" or "flux", rewrite it in plain words. The criterion rewards plain speech, so the model answer must show it.

Run the same prompt again after your edits and ask the AI to "find any step where the marking guide and the model answer disagree". It is a good second reader for internal consistency, even though it cannot judge the plumbing.

Fairness checks: before posting, with the first five, after the hire

Before you post. Run the audit prompt. Ask someone outside plumbing to read the ad for anything unclear. Test every "must": it should be a legal or safety requirement, or a skill you cannot teach in a month. Confirm that the pay range is one you would really pay.

With the first five applicants. Use the same room, board, materials and time for everyone. Keep two scorers, and have them write their scores before they talk. Ask the adjustments question of every applicant, not only those who mention a need. Keep the notes.

After the hire. Count how many applicants reached each stage (applied, phone call, test, offer) and note why people dropped out. If most leave after the phone call, the call needs work. If they withdraw after reading the test details, the test is too long or too vague. Look at the stage, not at the applicants. These are plain counts; if you want fairness monitoring by group, ask an adviser how to do it lawfully.

Keep applicant data out of chatbots. Names, CVs and test scores do not belong in a consumer chat window. The business plans of ChatGPT and Claude do not train on your content by default, according to the vendors, but nothing in this process needs a CV in a chat, so do not paste one. Everything above works with the brief, the ad and the marking guide alone.

If you hire in the EU. Annex III of the EU AI Act lists AI used to recruit or select people as high-risk, including systems that place targeted job adverts, filter applications or evaluate candidates, and the duties for those stand-alone systems have been pushed back to December 2027. The process here uses AI to draft an ad and a test, and people to select, so AI stays outside the selection decision. Confirm that reading with an adviser. The Act also bans AI that infers emotions in the workplace, other than for medical or safety reasons, which is what some video-interview tools claim to do; the video interview checklist for trades covers what to ask a vendor.

The same four documents then serve every later hire: the brief, the ad, the test and the scoring sheet. Update the brief each time with fresh on-call numbers and the real mix of work, and the ad and test follow. The tutorial on building interview questions and scorecards with AI shows the same method for roles that have no bench test.

Hiring questions plumbing firms ask

Should the plumber job ad show the pay range?

Yes. Give a range tied to qualifications, plus what on-call pays. An ad without pay information wastes time on both sides, and some places now require pay details in job ads, so check the rules where you hire. State the range you would honestly pay, not a hopeful top figure you never reach.

How long should the practical test be?

About 45 minutes on the bench plus a 15-minute conversation covers the job without exhausting anyone. If the whole process runs past an hour, pay applicants for their time. A test that needs a whole unpaid day is a trial shift under another name, and it screens out people who cannot take unpaid time off.

Can I use AI to screen plumber CVs?

It is better not to. AI can copy bias from its training, and if you hire in the EU, tools that filter or rank applicants sit on the AI Act's high-risk list. Use AI for the ad, the test and the marking guide, and let people shortlist against one written checklist.

What if an applicant needs changes to the test?

Ask everyone in the invitation whether anything would help them do their best work, and agree reasonable changes such as extra time, a seat, or instructions in a different format. Keep the criteria the same for every applicant. Note what was agreed, and ask an adviser which rules apply where you hire.

Further reads

• How to Write Job Descriptions With AI That Attract Good Hires — The general method for job descriptions that attract the right people.
• AI in Recruitment and Hiring: A Bias Test Checklist for Small Agencies — A repeatable bias test for any AI step in your hiring.
• Using AI in Recruiting a Small Team: Job Ad to Shortlist, Step by Step — The whole small-team process from ad to shortlist, step by step.
• AI Bias in Small Business Decisions: Hiring, Pricing and Credit — Where bias creeps into hiring, pricing and credit decisions.
• AI for Tradies: Eight Admin Jobs to Hand Over Before You Hire Help — Admin jobs to hand over before the new hire arrives.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the text of the EU AI Act (Annex III, employment and workers' management; Article 5 on emotion recognition) and the AI Act Service Desk timeline; OpenAI and Anthropic statements on business-plan data use.