AI for a Car Rental Business: Enquiries, Availability and Damage Notes

Give the assistant two sheets, a fleet list and a rules sheet, and let it draft enquiry replies only from them. Feed it a fresh availability extract so answers use real dates. For damage notes, have it describe check-out and check-in photos and list the differences. A person confirms every date, price and damage finding before anything reaches a customer.

The hard rule is no price promises. The assistant never types a price, a deposit or a word about insurance of its own. Those lines come from your rate card and your policy, pasted in by a person. What it contributes is speed and consistency: a complete reply drafted in moments, and every damage note written in the same format. Where a person checks matters more than how the drafts are worded.

Follow me on Instagram@sagnikteaches

A fleet sheet and a rules sheet feed every reply

An assistant answers a rental enquiry well only when your facts are in front of it. Without them it falls back on what rental firms in general tend to say, which is how a reply ends up promising unlimited mileage you do not offer. Two sheets prevent that. The fleet sheet describes what you rent. The rules sheet holds what you promise and what you refuse. The assistant may use nothing outside them, and the prompt says so. The figures below describe an illustrative fleet of 24 vehicles.

Connect on LinkedInSagnik Bhattacharya

Prices are deliberately absent. They live in your rate card or booking system, and a person adds them to a reply. Everything on the fleet sheet is something you can state without opening a calculator.

Subscribe on YouTube@sagnikteaches

The rules sheet carries the policies the assistant repeats and the ones it must hand over. Ten lines cover most enquiries:

1. Drivers must be 23 or over and have held a full licence for two years or more. Younger or newer drivers: the office decides.
2. At pick-up: a licence and a payment card in the main driver's name.
3. Additional drivers are named on the agreement before pick-up, by the office.
4. Included mileage is as shown on the fleet sheet. Beyond it, the office quotes.
5. Fuel is returned at the level recorded at check-out.
6. Pick-up and return, 08:00 to 18:00 daily. Out of hours only by arrangement with the office.
7. No smoking in any vehicle. Pets only in the estate and the people carrier, with a boot cover.
8. Changes are made through the office up to 24 hours before pick-up.
9. Insurance, excess and what is covered: the office answers, and sends the policy summary.
10. Breakdown or accident: the customer phones the office, always.

Keep each sheet to one page. When a rule changes, change the sheet, not the prompt. Store both in a Project on ChatGPT or Claude so the same two files are attached to every chat, instead of pasting them in each time.

Which enquiries the assistant may answer, and which go to a person

The point of lanes is that the assistant never improvises where improvisation costs money. Settle them once, in writing, and print the table for whoever covers the inbox.

Two refinements. The last row is not a drafting job at all: someone stuck at the roadside needs a phone number and a person, so set the inbox to flag those messages and never leave them waiting for a draft. And a reply that goes out without a person reading it is a bot talking to your customer. Say so in the message if that ever happens; for customers in the EU, the AI Act's transparency rule for chatbots has applied since 2 August 2026. Everything below keeps a person between the draft and the customer, so for now you are sending your own reply. The same split for holiday lets is set out in what a chatbot should not answer for a rental host.

If part of your enquiry traffic arrives on WhatsApp and you connect it through the WhatsApp Business Platform, read Meta's pricing page before automating replies. Each business number gets 1,000 free service messages a month, which are free-form replies inside the 24-hour customer window. Charging starts from the 1,001st, and utility messages sent inside an open window became billable on 1 October 2026. The free WhatsApp Business app is a different product from the Platform.

A first reply from the sheets: prompt, sample output, fixes

One prompt does the drafting. It forbids the three things that cost rental firms money, which are invented prices, invented cover and invented availability, and it makes the assistant show its working so a person can check the draft in seconds.

You draft replies to car rental enquiries for a small rental business. Use ONLY the FLEET SHEET, RULES SHEET and AVAILABILITY EXTRACT below.

Rules:
1. Never write a price, deposit, excess or discount. Where a price belongs, write [RATE LINE: office to add from the rate card].
2. Never say what insurance does or does not cover. Write [INSURANCE: office to answer] instead.
3. Use availability only as shown and quote the time the extract was taken. Say "currently showing as available, not yet held" unless the extract says HELD.
4. If the enquiry does not give pick-up date and time, return date and time, driver age, passengers or bags, ask for the missing items in one short list. Do not guess them.
5. If anything asked is not in the sheets, say the office will confirm, and list it under QUESTIONS FOR THE OFFICE.
6. Warm, plain English, under 150 words, signed [your name].

Return three parts: REPLY, then QUESTIONS FOR THE OFFICE, then CHECKS (one line per date and per rule you used, naming where it came from).

FLEET SHEET: [paste]
RULES SHEET: [paste]
AVAILABILITY EXTRACT (taken 09:40 today): [paste]
ENQUIRY: [paste]

Here is an enquiry (illustrative) and the extract the sheet produced at 09:40 for the dates it asked about.

Hi, do you have something for four adults plus luggage for a week from 14 March? Automatic if possible. I'm 26. How much would it be? Thanks

A typical draft comes back like this:

REPLY. Hi [first name], thanks for getting in touch. For four adults with luggage the estate is the best fit: it seats five, takes four large bags and is an automatic. The family hatchback is also free for those dates if you would rather have a smaller car. At 09:40 today our booking sheet showed one estate free from 14 March at 10:00 to 21 March at 10:00, not yet held. [RATE LINE: office to add from the rate card]. Our rule is drivers aged 23 or over with a full licence held for two years or more, so 26 is fine. We ask to see the licence and a payment card in the driver's name at pick-up. Could you send your pick-up and return times and roughly how many bags you will bring? The office will then confirm the car and the total. [your name]

CHECKS. Seats, bags and gearbox: estate row, fleet sheet. Availability: one estate free, extract of 09:40. Age rule: rules sheet, item 1. Documents: rules sheet, item 2.

Four things for the person reading the draft:

• Delete the hatchback sentence. The extract shows none free, and it is manual only. The CHECKS section names no source for it, which is how you notice an invention.
• Leave "not yet held" in. It is the phrase that stops the reply becoming a promise.
• The rate line stays a placeholder until you replace it. Paste the quote from your booking system or rate card, matching the dates, and read it once against the enquiry.
• Confirm in the booking system before you press send. The extract says "currently showing", and only the system says "reserved".

Availability: the sheet decides, the assistant words the answer

Date logic is where assistants slip, so take it away from them. Overlaps, turnaround gaps and workshop slots belong to a formula or your booking system. The assistant receives only the result and puts it into words.

If your booking system can already list free vehicles by class for a date range, export that and skip the formula. If you run on a spreadsheet, put every booking on one sheet called Bookings: vehicle ID in column A, start date and time in B, end date and time in C, and a type in D (rental, service or cleaning). Workshop visits and deep cleans go in as bookings, so the formula respects them. On a second sheet, list each vehicle ID in column A and its class in column B, type the requested pick-up in E1 and the requested return in F1, and put this in C2:

=COUNTIFS(Bookings!$A:$A,A2,Bookings!$B:$B,"<"&$F$1,Bookings!$C:$C,">"&($E$1-TIME(3,0,0)))=0

It returns TRUE when no booking for that vehicle starts before the requested return and ends later than the requested pick-up minus a three-hour turnaround buffer. Count the TRUE values per class with another COUNTIFS and you have the extract above. The assistant can write this for your own column letters, and 60 AI prompts for Google Sheets includes formula and debugging prompts you can adapt. Both Excel and Google Sheets support COUNTIFS and TIME.

Test the formula before you trust it, with four invented bookings on one vehicle:

1. A rental that ends two hours before the requested pick-up. The car must show as not free, because it sits inside the three-hour buffer.
2. A rental that ends four hours before. The car must show as free.
3. A rental that starts one hour before the requested return. Not free.
4. A workshop slot in the middle of the dates. Not free.

Decide at the same time whether a booking that starts exactly when the requested one ends should block it. With the formula as written it does not, which suits back-to-back rentals. If your cleaning needs a gap on that side too, add the buffer there as well.

Now the assistant. It gets the extract, stamped with the time it was taken, and the prompt makes it say "currently showing as available" and quote that time. It never sees the full booking list, so it has nothing to reason about dates with. In busy periods, treat an extract more than 15 minutes old as stale and rebuild it before replying. That figure is my suggestion, not a standard, so set your own.

Once a week, audit five replies that said "currently showing as available". Look at the booking system as it stood at the time quoted. A mismatch is a process fault to trace, such as a stale extract, a missing workshop slot or a vehicle filed under the wrong class, and finding it is the point of the audit.

A photo routine that makes damage notes possible

A comparison is only as good as its two photo sets. Before any assistant sees a picture, fix the routine, because it decides whether a later damage note holds up. This is the 18-shot set I would start with, taken in the same order every time so that shot 7 at check-in is the right side of the car again.

A few habits make the difference. Shoot in even light, in daylight or a lit bay, with the sun behind you rather than reflecting in the panel. Wipe off surface dirt or note it, because a wet or dirty car hides dents and invents scratches. Switch off filters, HDR and portrait modes. Name files so they sort and identify themselves, for example VEH07_2027-03-14_OUT_03_left-side.jpg: vehicle, date, out or in, shot number, view. And collect the photos through a form or a shared folder that keeps the original file, because messaging apps often compress images and are a poor home for evidence.

Check what your assistant accepts. Claude's support article on uploading files gives up to 20 files per chat at 500 MB each, and lists JPEG, PNG, GIF and WebP as the image formats. Some phones save in a high-efficiency format (HEIC) by default, so set the camera to its most compatible setting or convert before uploading. Other assistants have their own limits. Bear in mind that one 18-shot set plus close-ups already reaches 20, and a check-out set and a check-in set together come to 36 images, so the comparison below runs in batches.

At check-out, walk the car with the customer, show them the condition report with the photos, and have them confirm it in whatever way your rental agreement provides. The assistant helps after that, never instead of it. For a handover routine in a similar business, handover checklists for campervan rental is a useful companion.

Check-out notes: what the assistant writes from the photos

Keep the vocabulary small so notes from different staff match. I would use these damage words only: scratch, scuff, chip, dent, crack, kerb mark (for wheels), stain, tear, missing item, and dirt. And three size bands: under 2 cm (under an inch), 2 to 5 cm (one to two inches), and over 5 cm. The prompt builds them in.

You write vehicle condition notes for a car rental business from photos. I will upload the check-out photo set for one vehicle, numbered 1 to 18 as in the file names.

For each area visible, give one row: AREA | WHAT YOU CAN SEE | SIZE BAND (under 2 cm / 2 to 5 cm / over 5 cm / cannot tell) | CONFIDENCE (clear / unsure / not visible) | PHOTO.

Rules:
1. Describe only what is visible. If an area is not in any photo, write NOT PHOTOGRAPHED. Never assume it is undamaged.
2. Use only these words for marks: scratch, scuff, chip, dent, crack, kerb mark, stain, tear, missing item, dirt. If you cannot tell dirt from damage, write "unsure" and say why.
3. Never guess the cause, the cost of repair or who is responsible.
4. Sizes are estimates from the photo. Say "estimate" and use the size bands only.
5. Finish with a list headed WALK-AROUND: areas a staff member should look at in person.

The reply, for a set like the one above, looks like this (illustrative, shortened):

Read the rows that are not "clear". The roof row exists because the prompt forbids assuming an unphotographed area is fine, and it tells you the set was incomplete before the customer drives away. The two "unsure" rows land on the WALK-AROUND list: rear bumper right and windscreen top left. A staff member takes those two minutes at the counter, adds a close-up where needed, and corrects the note. The saved version is the human-checked one, stored with the photos as the condition report. The assistant's draft was a head start, and the signed-off note is the record.

Check-in: compare with check-out and flag differences only

At return, the task changes. You no longer need a description of the whole car, only what is different. Give the assistant the check-out notes and the new photos, and ask it to sort every area into three piles.

Here are the CHECK-OUT NOTES (human-checked) and the CHECK-IN photos for the same vehicle.

For each area, return one of:
UNCHANGED: the same marks as the notes, nothing new.
POSSIBLE NEW MARK: visible in a check-in photo and not in the notes. Say where, what it looks like, size band, confidence, and which photo.
CANNOT COMPARE: the check-in photo is missing, too dark, reflective or taken from a different angle. Say which.

Never say a mark is new, caused by the customer or chargeable. Use "possible". End with WALK-AROUND: the areas a person must check with the customer present.

A reply to a return (illustrative) might read:

• POSSIBLE NEW MARK. Rear bumper, left: chip, under 2 cm estimate, clear in check-in photo 4, not in the check-out notes.
• UNCHANGED. Front bumper scuff, front left wheel kerb mark, windscreen.
• CANNOT COMPARE. Right side: check-in photo 7 was taken at night with a light reflecting down the panel. Check-out photo 7 was taken in daylight.
• POSSIBLE NEW MARK. Rear seats: a stain, unsure whether it is a shadow, photo 16.
• WALK-AROUND. Right side panels, rear seats, rear bumper left.

Because the two sets together run to 36 images, split the job into three passes of no more than 12 images each. First pass: shots 1 to 4, 9 and 10 (front, left side and the left wheels), out and in. Second: shots 5 to 8, 11 and 12 (rear, right side and the right wheels). Third: shots 13 to 18 (glass, dashboard, interior and the raised roof or bonnet view). Ask for the dashboard reading in the third pass and compare the odometer and fuel level with the check-out record yourself, since those are numbers a person can check at a glance.

Then do not let the customer leave before the walk-around. Bring the check-out photos up on a tablet and look at the flagged areas together. A mark that both of you can see in one set and not the other needs no assistant at all, and a mark neither of you can find after a minute of looking is probably a reflection.

Where photo reading goes wrong

Failures you will see in your own photos

Photo reading fails in predictable ways, and most of them come from how the photo was taken rather than from the assistant. Learn the list, then fix the routine, not just the note.

• Reflections read as scratches. A lamp, a roof rail or a passing cloud on a wet or glossy panel looks like a line. The cure is an even-light rule and a second photo from another angle for any flagged mark.
• Dirt read as damage, or damage hidden by dirt. A muddy sill and a dusty bonnet both confuse the comparison, which is why the prompt tells the assistant to say "unsure" and why you wipe first.
• Dark paint hides dents. A shallow dent on a black door barely shows in a flat photo. Shoot dark cars at a slight angle to the panel so the reflections bend over any dent.
• Different conditions at out and in. A daylight set against a night set invites false alarms. The CANNOT COMPARE option exists for this, and it is a respectable answer.
• Wheel shots that miss the rim edge. Kerb marks sit on the outer lip. A photo taken square-on to the wheel face can miss them entirely.
• Glass chips that are invisible from outside. A small chip in a windscreen often shows up only from the driver's seat, against the light.

Two short stories, both illustrative. In the first, a "scratch" on the left door at check-in turned out to be the reflection of a lamp in a wet panel, while the check-out photo had been taken dry. The assistant marked it "unsure" because the prompt demands a confidence label, and a staff member cleared it in under a minute at the counter. In the second, a kerb mark on a rear alloy appeared in neither photo set, because both wheel shots had been taken square-on. The customer's own photo showed it later. The fix was a line in the shot list: keep the rim edge in frame.

What the scanner complaints teach a small operator

In June 2026 an InvestigateTV report described camera arches at large rental operations that photograph cars at pick-up and drop-off, compare the images in seconds and flag possible damage. Two customers told the reporters they had been billed $350 and $1,500 for damage they said they could not see. A consumer-advocacy director quoted in the report said that minor things such as dirt or water droplets can be flagged as damage and that "AI does make mistakes." One company said its decisions are always made by human experts, the other said it had improved its inspection process, and both disputed charges were later closed.

Treat that as an illustration of the risk, not a verdict on any product. The lesson for a small fleet is simple: let software flag, but charge only after a person has confirmed the mark and the customer has seen the evidence. Several vendors now sell inspection tools, from phone apps to camera arches. If you evaluate one, ask these questions, and be wary of any vendor who cannot answer them in writing:

1. How often does it flag dirt, water or reflections as damage, and on what data was that measured?
2. Does a person review every flag before a charge is raised?
3. Can the customer see both photo sets with the flagged area marked?
4. Are the original photos kept, for how long and where?
5. What is the dispute process, and who decides?
6. Can you export all your records if you leave?

For a smaller fleet the phone-and-assistant routine above gets you most of the value without a contract. Seven uses a small rental fleet can copy shows where else the same habits apply.

Damage charges stay with a person, and the customer sees the evidence

The no-price-promises rule applies with extra force here. The assistant never estimates a repair cost, never says who is responsible and never writes a charge. The sequence that keeps you fair and defensible has five steps:

1. A person looks at the car and both photo sets, with the customer present where possible.
2. If the mark is real and new, a person checks it against the damage terms in your rental agreement. Have a solicitor review those terms. An assistant can tidy a draft for that review, but the review itself is not optional.
3. A person gets a repair quote. The assistant has no part in it.
4. A person writes to the customer with the photos and a factual description, before anything is charged.
5. The outcome goes in the vehicle's file, next to the photos.

The assistant can help with step four, with tone and structure only. Give it the facts and this skeleton:

Hi [first name], when [vehicle] came back on [return date] we noticed a chip on the rear bumper, left side, about 2 cm. It is not visible in the check-out photos from [pick-up date]. Both sets are attached, with the area marked. We would like to talk it through and agree next steps under your rental agreement. Could you call us on [office phone] or reply with a good time? [your name]

The message leaves out an amount, a verdict and a deadline. Those belong to the person and the agreement. For evidence habits and wording in a comparable dispute, gathering evidence and drafting a damage request is worth reading.

Licences, plates and faces stay out of the chat

Rental files hold more personal data than most small businesses meet. Decide what the assistant never receives. Driving licences, identity documents, payment card details and signed agreements stay in your booking system or filing, not in a chat. Use the booking reference instead of the customer's name, and keep "[first name]" as a placeholder. Photos of vehicles can include number plates and bystanders, so crop or cover where you can, and ask your data-protection adviser how long you may keep the originals.

Choose the account with care. On a personal ChatGPT or Claude plan, check the model-training switch in the privacy settings. Business plans on both do not train on your content by default, although they need at least two seats. An individual plan on either is $20 a month at list price, and two business seats come to roughly $40 to $50 a month. Whichever you pick, a photo uploaded to an assistant is a copy of your record, so keep the original in your own storage. For the wider question of what photos and files an assistant handles well and badly, what breaks when ChatGPT reads PDFs, spreadsheets and photos has more.

Minutes saved at a 24-car fleet, and six numbers to track

For the 24-vehicle fleet above, assume 30 enquiries and 20 returns a week. These timings are planning guesses to replace with your own, so time ten of each job before and after you start.

A slow review step can wipe out the saving, which is why the "with a reviewed draft" column matters more than the "by hand" one. The plan itself is a minor cost beside the time spent reviewing, so judge the system on the reviewing.

After a month, look at six numbers rather than a feeling:

• Time to first reply. Measured from the enquiry arriving to a person pressing send.
• Drafts sent unchanged, and the commonest edits. Each repeated edit is a line to add to the prompt or a rule to add to the sheet.
• Availability mismatches found in the weekly audit. The target is none.
• Complete photo sets. How many are 18 out of 18, and which shots go missing most.
• Flags that staff confirmed against flags they dismissed. A high dismissal rate means the photo routine needs work.
• Damage conversations resolved with the photos. This is the number that justifies the routine.

Treat the first month as a trial on a handful of vehicles before you extend it to the whole fleet. When a draft goes wrong, write down which sheet, prompt or photo caused it, because that record shows where the process needs strengthening.

Questions rental owners ask about AI replies and damage notes

Can AI quote a price in a rental reply?

It can, but it should not. A figure the assistant types is a quote you may have to honour, and it can pick the number up from the wrong place. Put a placeholder where the price belongs, and have a person paste the line from your rate card or booking system before the reply goes out.

Can AI check availability on its own?

Not reliably. It should word an answer from a dated availability extract that your booking system or sheet produced, rather than do date logic itself. Say 'currently showing as available' instead of 'available', and have a person confirm in the booking system before anything is held.

Can I charge a customer for damage the AI found?

Treat an AI finding as a prompt to look, not as evidence of a charge. A person should confirm the mark on the car and in both photo sets, show the customer, and then apply your rental agreement. Have the agreement's damage terms reviewed by a solicitor rather than by an assistant.

How many photos do I need for each rental?

Eighteen standard shots at check-out and again at check-in is a workable set: eight body views, four wheels, the windscreen, the dashboard with odometer and fuel gauge, three of the interior and boot, and a raised view of the roof or bonnet, plus close-ups of existing marks. Check how many images your assistant accepts per chat and compare in batches.

How long should I keep check-out and check-in photos?

Long enough to settle any claim, and no longer than you can justify. In practice that means at least until your rental agreement's claim period and any card-payment dispute window have closed, and longer for anything still in dispute. Number plates and bystanders can make the photos personal data, so agree a retention period with your data-protection adviser, write it down and delete on schedule.

Further reads

• How to Reply to Every Enquiry in Under Five Minutes With AI — A general method for fast, accurate replies to every enquiry.
• AI for Bike Rental Businesses: Availability, Waivers, Weather Rebooking — Availability and waivers handled for a smaller rental business.
• AI in Travel and Transport: Jobs a Small Coach Hire Firm Can Hand Over — Quotes and booking replies for vehicle hire with drivers.
• One AI Inbox for WhatsApp, Instagram and Facebook Enquiries — Bring several enquiry channels into one reviewed inbox.
• Fleet Admin for Small Businesses: Automate Services, Checks and Fuel — Service reminders and checks for the vehicles themselves.
• AI Error Log: Track Mistakes and Stop Them Happening Again — Log slips in drafts and notes so they stop recurring.
• AI Mistakes That Damage Customer Trust, and How to Avoid Them — The slips that cost customers' trust, and how to avoid them.
• Ski Rental Software: Sizing Forms, Bookings and Returns With AI — A pre-arrival sizing form, kit tracking, damage terms and return messages for a small ski hire shop, with AI to draft the wording.
• Which Tool Hire Software Suits a Small Shop? Deposits, Quotes and AI — A buyer's guide for a small tool hire shop: the features that stop double-booking and lost deposits, two priced tools, and where AI helps with quotes.
• Party Rental Contract: Damage, Weather and Deposit Terms With AI — Draft deposit, damage and weather clauses for party rental from a decision sheet, test them against awkward scenarios, then brief your solicitor.
• E-Bike Rental Software: Bookings, Waivers and Battery Checks — A small e-bike fleet's software checklist: bookings per bike, waivers before handover, battery logs, and the checklists and replies AI can draft.
• Plant Hire Software: What Small UK Yards Should Compare Before Buying — Five tests for plant hire software: availability, quote-to-contract, delivery sign-off, an off-hire overdue list and damage notes, with vendor prices.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Claude support article on uploading files; Meta for Developers, WhatsApp Business Platform pricing; InvestigateTV report on rental-car scanner damage claims, 29 June 2026; Google Docs Editors function list; ChatGPT and Claude plan pages.