To draft a grey fleet policy with AI, first decide the rules yourself (who may drive on business, which insurance cover is required, how licences are checked, what mileage rate you pay), then paste those decisions into ChatGPT, Claude or Copilot and ask for a sectioned policy in plain English. Have a qualified adviser review it before staff sign.
The AI is good at structure and wording and poor at knowing your insurer's terms or the legal duties that apply to you. Left to guess, it writes confident clauses such as "the company insures all vehicles used for work", which would be wrong for almost every small firm. So the order matters: your decisions, then the draft, then the human check.
What the policy has to cover, and what it does not
A grey fleet is made of vehicles that belong to employees but get used for work: the installer who drives to three sites in her own hatchback, the broker who visits clients in his own saloon, the designer who carries samples to a showroom. The business does not own, insure or maintain those cars, yet it still has duties towards the people driving them and towards anyone they might hurt.
Commuting to a usual place of work is normally not business use. Driving from that workplace to a customer, from home to a first site on a day when you start away from base, or between two sites, normally is. A good policy states that line in a sentence so nobody argues about it after an accident. If you also run company vehicles, keep that policy separate; the automation side of those is covered in fleet admin for small businesses.
The policy is not a legal opinion and the AI is not your adviser. What it can do is turn a messy list of rules in your head into a document with headings, owners and checkable steps, which is most of the work.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowDecide the rules before you open the chat tool
Most weak policies are weak because nobody decided anything. The model fills each gap with the most common clause it has seen, and that clause may not match what you do. Fill in this sheet first. It takes about forty minutes for a small firm and it is the real work of the project.
| Decision | Options to choose between | Who decides |
|---|---|---|
| Who may drive their own car on business | Anyone with a licence; only staff with a signed approval; only after a trial period | Owner |
| Insurance level required | Social use only is not enough; business use cover required; plus carrying tools or stock if relevant | Owner, confirmed by the staff member's insurer |
| Licence check | At hiring only; every year; every six months for high-mileage drivers | Owner and adviser |
| Mileage payment | The approved tax-free rate; a lower company rate; a fixed car allowance | Owner, with accountant |
| Vehicle condition | Employee self-declaration yearly; photo of tyres and lights; proof of any required roadworthiness test | Owner |
| Fines, damage, excess | Employee pays all; company pays nothing on private policies; case by case | Owner, reviewed by adviser |
| Passengers and loads | Allowed with seatbelts; staff only; no tools over a set weight | Owner |
| Phone and journey rules | No handheld use; no calls while driving; maximum driving hours in a day | Owner |
Where you are unsure, write "ask adviser" in the cell. Those are the questions you will carry to the legal review later, and they are better as a short list than as a surprise in a finished draft.
A security installer with nine staff, three in their own cars
Illustration only. A security-installation firm has nine staff. Five are installers who drive between customer sites. Three of those use their own cars most days and two use company vans. The owner has never written anything down, pays mileage from a spreadsheet at a rate she remembers from last year, and has never asked to see anyone's insurance.
She fills in the decision sheet above. Her answers: only approved drivers may use their own car; business-use cover is required and must include carrying tools; licences are checked at hiring and every year; the approved tax-free rate is paid, nothing more; each driver self-declares the car's condition every January with photos of the four tyres; the employee bears fines and the repair excess; no handheld phone use; no more than nine hours on the road in any day.
Her sums for why it matters: the three own-car installers average 14,000 business miles each a year, 42,000 miles in total. At an illustrative payment of $0.60 a mile that is $25,200 a year paid out with no document behind it. For comparison, the cost of a single accident in a car that turns out not to be covered for the journey could dwarf that, because the first question afterwards is whether the car was insured for the use. The policy costs her a day of drafting and an hour a quarter to maintain.
The facts sheet you paste into the AI
Give the model your decisions as plain statements, with no personal data. This is the filled-in version for the installer firm. Replace the details with yours.
You are helping a small business write a grey fleet policy: rules for employees who use their own cars for work journeys. Write for non-lawyers. Use British English, short sentences, numbered sections, and "must" or "must not" where a rule is firm.
Business facts (do not add to these or assume others):
- 9 staff; 3 use their own cars regularly for site visits.
- Business use cover is required, including carrying tools.
- Licence is checked at hiring and every 12 months with the employee's consent.
- Mileage is paid at the approved tax-free rate per mile, claimed monthly on a form.
- Each driver completes a car condition declaration every January.
- The employee pays fines and the insurance excess; the company pays neither.
- No handheld phone use; a driver must not drive more than 9 hours in a day.
- Accidents must be reported to [manager name] the same day.
Where a point needs a legal or insurance answer I have not given, write [ASK ADVISER] and do not guess.
Write: purpose and scope; definitions (business use versus commuting); approval to drive; insurance; licence checks; vehicle condition; mileage claims; journey rules; accidents and breakdowns; fines and damage; records and data; breaches; review date.
The instruction to write [ASK ADVISER] instead of guessing is the most useful line in the prompt. It turns invented clauses into visible gaps you can see and chase.
What the AI sends back, and what to fix
An illustrative extract of a first draft, with the problems explained after it:
4. Insurance
4.1 You must hold insurance that covers business use for the whole time you drive on company business.
4.2 The company will reimburse any increase in your premium caused by business use.
4.3 You must show your certificate to [manager name] each year.
Clause 4.1 is fine. Clause 4.2 is an invention: the owner never said the firm would pay premium increases, and she would not want to. Clause 4.3 is acceptable but vague, since it does not say what "show" means (a scan, a photo, a download from the insurer's portal) or when. Delete 4.2 outright, and rewrite 4.3 as "by 31 January each year, or within seven days of a renewal or a change of car, send a copy of your current certificate to [manager name]". Always read the first draft hunting for reimbursements, promises and legal statements you did not supply.
The sections, and what each must say
| Section | What it must say | How you will know it is followed |
|---|---|---|
| Scope | Who is covered, which journeys count as business, the commuting line | Staff can state it back in one sentence |
| Approval | No one drives on business until approved in writing | Approval list in the tracking sheet |
| Insurance | Business-use cover, loads, annual proof, change of car or insurer | Certificate dated within the last 12 months |
| Licence | Valid category, consent to checks, duty to report penalties or bans at once | Check date and result logged |
| Vehicle | Legal and roadworthy, regular servicing, yearly declaration | Declaration on file each January |
| Mileage | Rate, claim form, deadline, what a journey record contains | Claims reconcile to diary entries |
| Journey rules | Phone use, hours, breaks, planning visits, weather | Spot check of schedules |
| Incidents | Who to call, what to record, no admission of fault at the scene | Incident form exists and is easy to find |
| Fines and damage | Who pays what | Adviser has approved the wording |
| Data | What you store about licences and certificates, who sees it, how long you keep it | Restricted folder, retention date noted |
Insurance checks that survive an accident
The insurance section is the one with money attached. Private car policies list permitted uses, typically social, domestic and pleasure, with commuting and business use as extras. Insurers' own guides say business use is needed even for the occasional trip to a client's office or a second site, and that driving outside the listed use can leave the policy invalid for that journey. If an employee has only social or commuting cover, a work journey could leave a claim refused and the driver personally liable. Your policy should require proof that business use is covered, and the proof should be a document, not a promise.
Ask the AI to draft the message the employee sends to their insurer. A request that actually gets a useful reply names the use precisely:
Subject: Confirming business use cover for my policy
Hello,
I use my car for my job, visiting customer sites a few times a week. Typical journeys are 10 to 60 miles, and I sometimes carry tools weighing up to 20 kg.
Please confirm in writing: (1) whether my current policy covers this use; (2) if not, what class of use I need and what it costs; (3) whether carrying tools is covered.
Policy number: [policy number]
Name: [your name]
Insurers answer the question asked, so the figure of "up to 20 kg of tools" matters. Whatever cover the employee ends up with, your policy should say they are responsible for paying for it. If that causes resentment, discuss the allowance with your accountant rather than quietly absorbing a premium increase for one person and not another.
Keep a one-row-per-driver sheet with the cover class, the certificate date and the next due date. If you already track subcontractor insurance this way, copy the layout from tracking insurance expiry for subcontractors; the columns are the same.
Licence checks without putting personal data into the chat
A licence check has a legal and a practical side. Practically, you need to know the person holds a valid licence for the vehicle category and see whether anything endorses it. Legally, licence details are personal data, so you need the employee's consent or another lawful basis, a stated purpose and a limit on who sees the result. That is exactly the kind of point to put on the adviser's list.
What the AI should do is write the procedure and the consent wording, never handle the licence itself. A short version it might draft:
Licence check consent
I agree that [company name] may check my driving licence record when I am approved to drive on business and then once a year. I understand the check is used only to confirm I am allowed to drive and to manage insurance, that only [manager name] will see the result, and that I must tell [manager name] straight away if my licence is endorsed, suspended or revoked.
Name: [name] Date: [date]
The tracking sheet carries four columns only: name, check date, result (valid or not valid) and next check. Do not paste licence numbers or scans into ChatGPT, Claude or Copilot to "summarise" them. If a staff member raises a worry about the data, what to check in an AI tool's privacy terms shows you where your inputs go and what the tool keeps.
Mileage claims you can audit
A mileage claim should let a stranger rebuild the journey. The minimum fields: date, start point, destination, purpose, miles, and the reason a direct route was or was not taken. Free-text lines such as "client visits" are where overpayment hides.
The AI is useful here as a checker. Export a month of claims to a sheet, strip names to initials, and give it a task like this:
Below is a month of mileage claims (driver initials, date, from, to, purpose, miles). Flag any row where: the same date has overlapping or impossible journeys; the miles are a round number repeated more than three times; the purpose is blank or only says "visit"; the total for one day is over 150 miles. Return a table of flagged rows with the reason. Do not guess at addresses.
Illustrative output for the installer firm's March claims:
| Driver | Date | Reason flagged |
| AB | 6 Mar | Two journeys of 58 miles each listed from "base"; second starts before first ends |
| CD | 14 Mar | Purpose "visit"; 162 miles |
| CD | 21 Mar | "50 miles" claimed on four separate dates |
None of these is proof of anything. They are three questions for a ten-minute chat, and that is all the AI is for. The original claim sheet must be kept the way your accountant requires. For expense approvals more widely, AI expense management for receipts, mileage and approvals goes further.
A quick sum to test the rate
Take one driver at 6,000 business miles a year. At an illustrative $0.60 a mile the firm pays $3,600. If your accountant says the approved tax-free rate is lower for miles above a threshold, the same driver might be paid less per mile after that point, and the policy has to state it. A fixed car allowance of $300 a month would cost the same $3,600 but pay a low-mileage colleague the same sum as the driver covering 14,000 miles, which staff notice. Settle the method with your accountant before the AI writes the clause, then paste the final rates in yourself.
Vehicle condition and journey rules
You cannot service an employee's car, but you can require evidence that it is roadworthy. The usual compromise is a yearly declaration with three or four photos: tyres with tread visible, windscreen, lights working, and the dashboard showing no warning lights. Add proof of any roadworthiness test required for cars of that age; the adviser will tell you which. The declaration lives in the tracking sheet next to the insurance date.
Journey rules turn good intentions into something checkable. The three that matter most are phones, hours and planning. Say plainly that handheld use is banned, that calls are taken only when parked, and that a visit schedule must leave realistic travel time. If you build schedules with software, give it a travel buffer between visits. A day with five sites in four hours is a policy breach caused by the planner, not the driver.
Incidents and breakdowns
Write a one-page incident card the driver keeps in the glovebox: stop safely, check for injuries, call emergency services if needed, take photos, swap details, do not admit fault, call the named manager the same day. Have the AI turn your bullet list into a card that fits one side of A5, and print it. The card is more likely to be used than a clause on page six of a policy.
Where the first draft went wrong for a mortgage broker
Illustration. A two-adviser mortgage broker used the AI to write its grey fleet policy because its advisers visit clients in their own cars. The draft said the company "provides breakdown cover and covers any loss of no-claims discount". Nobody had said that. It also invented a rule that drivers must "not exceed 8 hours of driving per week", which no adviser could meet, and said nothing about carrying client files in the boot. The owner caught the first two on the read-through because the prompt had demanded [ASK ADVISER] markers and the model had ignored them in two places. She sent the draft back with the instruction "list every sentence not supported by my facts sheet", and the AI returned the six sentences it had added itself. Two were deleted; the rest were turned into questions for her adviser. The missing section on client documents in the car became a rule: files stay in a locked case, never left in a parked car overnight.
The lesson is mechanical. After every draft, run the "unsupported sentences" check and delete or question each hit.
What to hand the adviser for the legal review
Send a short pack, not a long email: the policy draft, the decision sheet with its [ASK ADVISER] cells, your mileage method, and three questions. First, does the wording on fines, damage and excess hold up for our staff contracts. Second, are our licence checks and record-keeping consistent with data-protection law that applies to us. Third, do we owe any extra duty because staff drive for work, such as risk assessment or training. An hour of an adviser's time on those is a better use of money than a long unfocused read.
Say frankly that the document was drafted with AI from your own facts sheet. It is not embarrassing and it saves their time: they are reviewing, not writing from nothing. Keep the answers, update the policy and note the change date in a version line. Where the policy sits inside a wider staff handbook, writing an employee handbook with AI and what to check shows how the pieces fit.
Rolling it out and proving it works
Hand the policy to each driver with a one-page summary the AI can write from the full text: five rules, three deadlines, who to call. Take a signed acknowledgement for the file. Staff who have driven their own cars for years may read this as distrust, so explain that it protects them too: cover confirmed on paper is what protects them after an accident. The same rollout habits appear in how to roll out a policy so staff follow it, which is written for AI rules but applies to any policy.
Then check it worked, without drama:
- Every approved driver has a certificate dated within twelve months.
- Every driver has a licence check date within twelve months.
- January declarations are all on file by the end of the month.
- Each month's mileage claims reconcile to diary or job records within a small tolerance you set.
- Someone other than the policy owner could find the incident card in under a minute.
Put the five checks in the calendar as a quarterly ten-minute review. Add two more rows to the tracking sheet, one for new starters and one for changes of car, because a new vehicle quietly invalidates last year's proof. When the review turns up a gap, correct the process first and the person second.
Keep the policy short enough that someone will read it on a bad day. Four pages that people follow beat fifteen that sit in a drawer, and the AI will happily shorten a draft if you tell it the page limit up front.
Questions owners ask before they publish a grey fleet policy
What counts as grey fleet?
Grey fleet means any vehicle owned or hired by an employee that they use for work journeys and claim mileage for. The test is who owns the vehicle and why it moves. Driving to your normal workplace is commuting; driving from there to a customer, a supplier or a second site is business use, and the policy should cover it.
Can the AI see licence numbers or insurance certificates?
It should not. Keep personal documents out of the chat tool entirely. Use the AI to write the procedure, the email wording and the tracking-sheet layout, and keep the real licence numbers, certificate scans and expiry dates in your own HR folder or sheet with restricted access.
How often should we re-check insurance and licences?
Insurance cover is usually renewed every year, so check the certificate at each renewal. Licence checks are commonly done at hiring and then at least yearly, with shorter gaps for drivers who do high mileage or have penalty points. Your adviser or insurer can confirm the interval that fits your risk.
Do we have to pay for staff car servicing?
Not as a rule. The mileage rate is meant to cover running costs, so most policies say the employee is responsible for servicing, tyres and roadworthiness checks, and the company checks that they happen. Wording on damage, fines and breakdown cover should be reviewed by a qualified adviser.
Further reads
- AI Route Planning for Small Delivery Fleets and Mobile Teams — Plan routes that cut business miles and fatigue.
- AI Software for Trucking Companies: What to Check for a Small Fleet — Software checks for firms that run their own vehicles.
- AI Employee Onboarding: Plan a New Starter's First Two Weeks — Add the driving rules to a new starter's first fortnight.
- How to Write a Health and Safety Policy With AI and What to Check — Write the wider health and safety policy this one plugs into.
- AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Build rotas that leave realistic travel time between shifts.
- How to Set Up Human Review for AI Work Without Slowing Down — Set up the human review step every AI draft needs.
- AI Job Scheduling and Dispatch for Trades and Field Service Teams — Schedule site visits with realistic travel time built in.
- AI for Fleet Managers: Seven Jobs Beyond Service and Fuel Reminders — Licence registers, accident forms, insurance renewal packs, repair quotes, penalty notices, replacement cases and board reports, with prompts and checks.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: an insurer's guide to classes of use and driving for work (Ageas) and a fleet-risk guide on document-check frequency (Cardinus), both read 5 November 2026; general employer duty-of-care principles. Illustrative figures are not official rates. Check your own insurer and tax authority.