Set up a company AI account before their first day, spend twenty minutes on day one going through a one-page summary of your AI rules, then give them three approved tasks for week one with a colleague checking the output. Widen their access once they've shown they follow the rules, and record the sign-off.
The first fortnight sets habits. A new starter who spends it pasting work into a personal chatbot account will keep doing so long after you've written a policy. Doing this properly takes about four to five hours of a colleague's time spread over the first month, and the one-page rules, task table and sign-off checklist below make it repeatable for every hire after this one.
Before day one: accounts, access and a prepared buddy
Most onboarding problems with AI happen in the gap between "you start Monday" and "your account's ready Thursday". People want to be useful, the tool they know is on their phone, and the customer email needs answering. Close the gap before they arrive:
- Create the company AI account. Add them to your business plan (ChatGPT Business, Claude Team, Microsoft 365 Copilot, Gemini in Workspace, whichever you use) so their work sits in an account you control and can close. Setting up company AI accounts instead of personal logins explains why this matters when someone later leaves.
- Turn on two-factor sign-in for the account on day one, not "later". Turning on two-factor authentication for every AI account covers each tool.
- Check file permissions. Microsoft 365 Copilot and Gemini in Workspace work with whatever files the user can already open, so a new starter with access to the whole shared drive has an AI that can summarise the whole shared drive. Give folder access by role.
- Add them to shared resources. Shared projects, Gems (which become skills from November 2026) or Notebooks for their role, the prompt library, and the templates they'll use.
- Brief the buddy. Pick one colleague who does similar work and ask them to set aside about 15 minutes a day in week one for checking.
- Print or share the one-page rules and the first-week tasks, both below.
Day one: the rules in twenty minutes
Your full AI usage policy may run to several pages. New starters need the short version first, in plain language, with the three or four rules that matter most in bold. Go through it face to face or on a call, then ask them to sign or reply to confirm they've read it. If you haven't written the full policy yet, writing an AI usage policy for a small business is the place to start.
OUR AI RULES ON ONE PAGE (read, then sign at the bottom)
1. USE ONLY OUR ACCOUNTS
Approved tools: [company assistant] and [meeting notes tool].
Never use a personal AI account or a free app for work.
2. WHAT NEVER GOES INTO ANY AI TOOL
Key-safe codes, alarm codes, passwords, bank details,
customers' health information, anything a client has marked
confidential.
3. WHAT CAN GO IN, ON OUR ACCOUNTS ONLY
Customer names and job details needed for the task.
Our price list, templates and product information.
4. YOU CHECK EVERYTHING BEFORE IT LEAVES
Prices, dates, quantities, model numbers and names are checked
against the job system, every time. AI drafts; you decide.
5. SAY WHEN AI HELPED, IF ASKED
If a customer asks, tell them honestly that we use AI to help
draft messages, and that a person checks each one.
6. WHEN AI GETS SOMETHING WRONG
Don't send it. Tell [buddy / office manager] the same day,
even if you caught it. That's how we fix the instructions.
7. WHO TO ASK
[Name], or post in [channel].
Name: ____________ Signed: ____________ Date: ________The examples in rule 2 matter more than the headline. "Never enter confidential data" is abstract; "never enter key-safe codes" is something a new coordinator immediately recognises from their first job sheet. Pick the examples that actually occur in your business. Filled in for three other kinds of business (illustrative), rule 2 looks quite different:
| Business | Rule 2 examples a new starter will recognise |
|---|---|
| Dental practice | Medical history forms, X-ray images, anything from a patient's treatment notes, payment card details taken over the phone |
| Payroll bureau | Salaries, bank account numbers, tax reference numbers, sickness and maternity records, anything from a disciplinary file |
| Locksmith and security firm | Master-key system charts, alarm codes, safe combinations, photos showing a customer's lock or keys clearly enough to copy |
The locksmith's last example is the kind nobody writes until it's pointed out: a close-up photo of a key is enough for someone to cut a copy, and job photos are exactly what new staff paste in when asking AI to write up a visit.
Week one: three approved tasks with a buddy check
Rather than "feel free to use the AI", give each new starter a short list of tasks they're approved to use it for in their first week. Three is enough. Their buddy checks every AI-assisted output before it goes anywhere, then gives brief feedback: what was right, what needed changing, and which rule it touched. Here's how that might look at an illustrative heating and air-conditioning installer:
| Role | Week-one approved tasks | What the buddy checks |
|---|---|---|
| Service coordinator | Drafting booking confirmations; summarising a long customer email thread; drafting a reply to a routine service enquiry | Dates, times and addresses match the job system; no promises about arrival windows beyond policy |
| Engineer | Dictating job notes into the phone app; turning notes into a customer-friendly visit summary; looking up a fault code explanation to check against the manual | Model numbers and readings are correct; the summary doesn't state a diagnosis the engineer didn't make |
| Estimator | Drafting a quote cover letter; summarising a site survey into a checklist; rewording technical specification text for a homeowner | Quantities and equipment match the survey; nothing is priced by the AI |
The buddy's feedback is where the learning happens, so keep it specific. At an illustrative small hotel, a new reservations assistant used AI to draft a reply to a guest asking about early check-in:
AI-ASSISTED DRAFT
"Early check-in from 11am is no problem at all, we'll have your
room ready for you on arrival."
BUDDY'S NOTE (three lines, sent the same morning)
Right: friendly tone, answered the actual question.
Change: early check-in is subject to availability and costs $30
before noon; check the policy sheet, not the guest's request.
Rule touched: 4 (check everything before it leaves).
SENT VERSION
"We'll do our best to have your room ready early. Check-in before
noon is subject to availability and costs $30; we'll confirm on the
morning of your stay."Three lines take the buddy two minutes, and the assistant now knows the policy, the habit and the rule it links to. That's what a week of checks is for.
Keep the tasks low-risk and frequent, so they get real practice without anything expensive depending on it. Nothing in week one should involve the AI producing a price, a legal statement or anything a customer would treat as a commitment.
Weeks two to four: earn wider access
From week two, the buddy moves from checking everything to checking a sample, and the approved task list grows. A simple sign-off at the end of the first month makes the step explicit rather than drifting. The new starter is signed off when they can tick all of these:
- At least ten AI-assisted outputs reviewed, with no factual error reaching a customer in the last five.
- Can explain, without notes, the rules on what never goes into an AI tool, with two examples from their own work.
- Has reported at least one AI mistake they caught, or confirmed they haven't seen one. (Asking the question normalises reporting.)
- Knows where the shared prompts and templates are, and has used them rather than writing everything from scratch.
- Has used only company accounts since day one. Ask directly; most people answer honestly when it's asked without blame.
For the hotel's reservations assistant, the completed sign-off read like this:
AI SIGN-OFF: reservations assistant End of week 4
Outputs reviewed: 16 (errors reaching guests in last 5: none)
Rules from memory: yes; examples given: "card numbers from the
booking form", "guests' ID scans"
Mistakes reported: 2 (a wrong rate for a group booking; a
breakfast time from last season)
Uses shared prompts: yes, for confirmations and FAQs
Company accounts only: yes, asked directly
Signed off by: front office managerAfter sign-off they can use AI on any task within your policy, and the buddy check becomes the normal review your business applies to everyone. Put the date of sign-off in their training record.
Experienced hires bring AI habits with them
A new starter from a larger firm may have used tools you don't have, or followed stricter rules than yours. Someone from a smaller one may have used free tools for everything. Either way, ask about it in week one: which tools did you use, for what, and what did you find useful? You'll often learn a good prompt or two.
Three things to settle early. Personal subscriptions stay personal: work goes in company accounts, even if their personal tool is better at something. Nothing comes with them from a previous employer: no saved prompts or chats containing that employer's client details. And if they prefer a tool you don't use, the route is your tool-approval process rather than a quiet workaround, because unapproved tools are how client data ends up in free AI services.
The second of those is easy to break without meaning to. Take an experienced estimator joining an illustrative joinery firm who, to get started quickly, pastes a long set of instructions from her old job into the company assistant's shared project. A week later a draft quote offers "our usual 12% trade discount for returning builders". The joinery firm has no such discount; it was her previous employer's term, sitting in the pasted instructions along with two of that firm's client names. The buddy caught it before it went out. The fix was simple (strip the instructions back to your own rules and prices), and the lesson went into the week-one conversation for every experienced hire after her: bring your skills, not your old files.
Temps, contractors and apprentices
Short-term and junior starters need the same rules but a different set-up.
Temps and seasonal staff. If someone is with you for a few weeks, decide at the start whether they use AI at all. If they do, give them a proper seat for the period and remove it on their last day; never hand over a colleague's login or a shared account, because you lose any record of who did what. If the role is simple enough that AI adds little, say "no AI tools in this role" plainly, which is clearer than leaving it unmentioned.
A garden centre taking on five temps for the eight weeks before Christmas shows how the decision usually splits. Four of them work the tills and help customers load trees into cars, and AI adds nothing to either, so their induction sheet says "no AI tools in this role" in one line. The fifth handles online orders and delivery emails, the same work the permanent office assistant does with AI. She gets a seat for two months (on ChatGPT Business, about $50 at the $25 monthly rate), the one-page rules and the same three week-one tasks as a permanent starter, compressed into her first three days. Her seat is removed on her last afternoon, and the removal goes on the same checklist as returning her keys.
Contractors and freelancers. They may use their own AI tools on your work. Your contract with them should say which tools are acceptable for your data, whether your client information may go into them, and that they follow your rules while working in your systems. The one-page rules can go in the contractor pack.
Apprentices and trainees. Give them the same first-week tasks, but keep the buddy checking everything for longer and make them do some tasks without AI first. Someone learning the trade needs to know what a correct booking or job summary looks like before they can judge an AI's version of it.
At an illustrative electrical contractor, an apprentice writes her first ten job summaries by hand, and only then starts turning her notes into AI drafts. The difference shows on her third AI-assisted summary. Her notes say she tested the two new kitchen circuits; the draft says "all circuits tested and found satisfactory". It reads well and it's wrong, because nobody tested the rest of the house, and a customer could read it as a clean bill for the whole installation. She spots it because her own handwritten summaries never said "all". Without those ten, she'd have had nothing to compare the draft against, and her supervisor added a line to the task sheet: summaries state exactly which circuits were tested.
Worked example: an HVAC installer hiring a service coordinator
Take an illustrative twelve-person HVAC installer bringing on a new service coordinator. The office manager owns onboarding, and the existing coordinator is the buddy.
| When | What happened | Colleague time |
|---|---|---|
| The Friday before | Company assistant seat added; two-factor set up; access to the service folders only; added to the shared "customer replies" project | 45 minutes |
| Day one | One-page rules gone through and signed; 40-minute walkthrough of the assistant using real past enquiries | 60 minutes |
| Week one | Three approved tasks; buddy checks every output | About 15 minutes a day, 75 in total |
| Weeks two to four | Sample checks twice a week; approved list extended to supplier emails and appointment reminders | About 90 minutes in total |
| End of week four | Sign-off against the checklist | 20 minutes |
That's just under five hours of colleague time. The firm learned why the Friday set-up matters the hard way, with the previous hire: their account wasn't ready until day three, and on day two they pasted a job sheet, key-safe code included, into a free chatbot on their phone to tidy up a confirmation message. Nothing came of it, but the key-safe code was changed and the "before day one" step became fixed. With this hire, the buddy caught two errors in week one, both appointment windows copied from the customer's request rather than the job system, and none reached a customer. By week four the coordinator was using the shared reply prompts for most routine enquiries. If you'd also like to use AI to plan the wider induction itself, see planning a new starter's first two weeks with AI.
Close the loop: the next hire and the eventual leaver
After each onboarding, spend ten minutes updating three things: the one-page rules (did the new starter ask about something they don't cover?), the first-week task table (were the tasks too easy or too risky?), and the before-day-one checklist. Each hire makes the next one faster.
After the HVAC installer's coordinator was signed off, the ten minutes produced three changes. Rule 3 gained a line, because she'd asked in week one whether customers' photos of their boilers could go into the assistant: "Equipment photos are fine; crop out house numbers, people and anything on the walls." The week-one task table swapped "summarising a long customer email thread" for "drafting appointment reminders", since the thread summaries had needed the most correction and weren't low-risk enough for a first week. And the before-day-one checklist now says "check the shared project's files are current", because the project had still held last winter's call-out rates when she joined.
Onboarding and offboarding are two ends of the same process. Everything you set up before day one, the seat, shared projects, folder access and any connected apps, is a list of things to remove when that person leaves. Keep it with their record, and use it with the staff offboarding checklist for AI tools when the time comes.
Further reads
- How to Roll Out an AI Policy So Staff Actually Follow It — Make the rules behind the one-pager stick for everyone.
- How to Build a Shared Prompt Library for Your Team — Give new starters prompts that already work.
- AI Literacy Requirements: What Your Staff Need to Know — What every member of staff should understand about AI.
- How to Test Job Candidates' AI Skills in an Interview — Find out a hire's AI habits before they start.
- How Much Time and Money Does AI Staff Training Take? — Size the training time for new and existing staff.
- How to Set Up Single Sign-On for Your Team's AI Tools — Tie AI accounts to your main logins so access is simpler.
- How to Handle Staff Who Over-Rely on AI — Signs of AI over-reliance, a conversation script, a three-rule standard to put in writing, and when a pattern needs a formal process.
- How to Scale AI From One Workflow to the Whole Business — How to copy your first working AI workflow across the business: a readiness gate, a playbook template, adjacency rules and a pet shop's nine months.
- How to Train Staff to Use AI in a Small Business — A three-layer training plan with learning outcomes, a 75-minute session agenda, a copyable error-spotting exercise and a language-school example.
- Reduce Owner Dependency: Use AI to Capture What Only You Know — Get what's in your head into a form staff can use: an interruption log, an AI interview prompt, decision rules with thresholds, and a shared assistant.
- Does a Five-Person Business Really Need an AI Policy? — Why a five-person business needs a one-page AI policy rather than a handbook: six triggers, a complete template, and when one page stops being enough.
- How to Build a Kitchen Training Manual With AI — Capture how your kitchen really works, let AI shape it into station cards and quizzes, and keep every temperature and allergen under the head chef's control.
- Gemini Notebook (Formerly NotebookLM) for Small Business Teams — Set up a shared Gemini Notebook for onboarding, policy lookups and customer answers, with plan limits, sharing settings and a music school worked example.
- Who Owns AI Chat History When an Employee Leaves? — What happens to a leaver's AI chats on ChatGPT Business, Claude Team, Copilot and Gemini, who can read or export them, and a handover plan that works.
- How to Hire Your First AI Automation Specialist — Size the backlog, pick the right shape of role, write an outcome-based advert, test with a paid task and plan the first 90 days, with a campsite group's sums.
- Should a Small Business Hire a Prompt Engineer? — Why prompt work rarely fills a job in a small firm, the five thresholds that change the answer, and a skills test if you recruit anyway.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Learn, Data, Privacy, and Security for Microsoft 365 Copilot (permission-based access); OpenAI help pages on managing members and seats in ChatGPT Business; Claude help pages on sharing projects in Team plans.