Use AI to draft three kinds of message from facts you hand it each morning: availability answers from a snapshot of free bikes, waiver reminders from a count of unsigned riders per booking, and rebooking offers once you have made the weather call. You check and send every draft, and only you decide whether a ride goes ahead.
Rental shops get into trouble when they ask the tool to know things it cannot see, such as whether two e-bikes are still free at 3 pm, and it answers confidently anyway. The fix is dull and reliable: give it fresh facts, give it written rules, and keep a person on the send button.
Every figure below belongs to an invented shop: 14 adult hybrids, 6 e-bikes, 8 children's bikes and 4 clip-on child seats, rented in three-hour blocks at $15 for a hybrid, $30 for an e-bike, $10 for a children's bike and $5 for a seat. Swap in your own fleet and rates. The method stays the same.
Build a free-bike snapshot before you write a single prompt
An availability answer can be no better than the list behind it. A workable snapshot is a plain-text block that you rebuild each morning, and again before any batch of replies after lunch. It takes about ten minutes in the first week and two or three once it is a habit. Copy it from your booking calendar, or from a spreadsheet that mirrors the calendar if your booking tool can export one.
Five details decide whether the snapshot is safe to rely on:
- Free to promise, not free in the yard. Take out the bikes you hold back for repairs or walk-ins before you count. If one hybrid and one e-bike are never sold online, the snapshot should show what is left after that.
- Counts per start time. "Free for the full three hours from 10:00" is the number a customer needs, not how many bikes you own.
- Extras on their own line. Child seats, trailers and helmets run out separately from bikes, and families with small children hit that limit first.
- A build time. The AI quotes it in every availability answer, so staff can see at a glance how stale the facts are.
- No customer details. Counts and slot times only. No names, phone numbers or emails.
This is a Saturday extract from the invented shop. Your real one would cover the next seven days.
AVAILABILITY SNAPSHOT - built Mon 08:10 - free to promise (repair buffer already removed)
Saturday, three-hour blocks, free for the whole block
Start 09:00 hybrids 5 e-bikes 3 kids 6 child seats 2
Start 10:00 hybrids 3 e-bikes 2 kids 5 child seats 1
Start 13:00 hybrids 7 e-bikes 4 kids 6 child seats 3
Start 16:00 hybrids 9 e-bikes 5 kids 7 child seats 4
No starts at 11:00 or 12:00 (returns and staff lunch)On a busy day I'd treat any snapshot older than two hours as stale. Two hours is my own working limit rather than an industry figure, so set yours by watching how quickly your calendar fills on a normal Saturday.
Standing instructions that stop it promising bikes you don't have
Put the rules in a Project, so you paste only the new snapshot each day. ChatGPT and Claude both have Projects, and a Gemini Gem works the same way, though Google is turning Gems into "skills" from November 2026. Avoid building on ChatGPT's custom GPTs, because OpenAI has scheduled their shutdown for 11 December 2026. Upload one-page files next to the instructions: your rates, a size chart, your terms for changes and cancellations, and your weather rules. ChatGPT allows only 5 files per project on Free (25 on Go or Plus), so keep each document short. The instructions below name six files, so on Free put the included-items and not-offered lists into one.
You draft replies for a bike rental shop. A person reads and sends every message. You never contact customers.
Facts you may use: (1) the AVAILABILITY SNAPSHOT pasted in this chat, (2) the files in this project: rates, size chart, terms, weather rules, included items, not offered. If a fact is in neither, write "I'll check and come back to you" and add a line for staff starting STAFF CHECK.
Availability: quote counts only from the newest snapshot and name its build time. Never say "booked" or "reserved". Say "available when we last checked at [time]" and give the deposit link from the rates file.
Prices: copy from the rates file and show the working, for example 3 x $15 + 1 x $30 + $5 = $80.
Never: predict the weather; promise a refund or credit beyond the terms file; explain what a waiver means; give fit or safety advice beyond the size chart; mention helmets, locks or insurance unless the included-items file lists them.
If the message mentions an injury, a complaint, a damaged bike, a refund dispute or a legal threat, write only HUMAN ONLY.
Tone: friendly, short, plain English. First names only. Sign off "[your name], [shop name]".Three lines in that block carry most of the weight. "Show the working" makes arithmetic slips visible, because a total sitting beside an e-bike priced as a hybrid is easy to spot. The STAFF CHECK line gives the AI a polite way to say it doesn't know, so it has no reason to invent, and you will see it often in the first week. The HUMAN ONLY rule keeps injuries, complaints and refund disputes away from a drafting tool altogether.
Before it touches a real customer, send it the awkward messages. Six tests take about fifteen minutes and expose most of the problems:
| Test message | What a good draft does |
|---|---|
| "Do you have 12 bikes tomorrow at 10?" (the snapshot shows 7 free) | Offers the 7, suggests a split start or a second day, shows the sums and names the snapshot time. |
| "Is it going to rain on Saturday?" | Declines to predict. Quotes the forecast you pasted, with its source and time, or points to your weather rules. |
| "My son fell off one of your bikes last week. Can I have a refund?" | Writes HUMAN ONLY and nothing else. |
| "Can my dog ride in a basket?" (in no file) | Says you will check, adds a STAFF CHECK line and invents no rule. |
| "What's your best price for six bikes all day?" | Quotes the rates file only, with working, and offers no discount you haven't written down. |
| "I'm 195 cm tall. What size do I need?" | Uses the size chart. If 195 cm sits outside it, adds STAFF CHECK instead of guessing. |
Monday to Sunday: one week of enquiries, worked through
Here is a week for the invented shop. The AI produced every draft in the middle column. The right-hand column is what a person caught before anything went out. The figures and the shop are illustrative.
| Day | The enquiry and the draft | What the person caught |
|---|---|---|
| Mon | A group wants 4 adult bikes and 2 child seats on Saturday at 10:00 for three hours. The draft offered 3 hybrids, 1 e-bike and 1 seat at 10:00, or 4 hybrids and both seats at 09:00. | The 10:00 option priced the e-bike at the hybrid rate. The printed working ("4 x $15") exposed it: the true total is $80, not $65. |
| Tue | A parent asks about a bike for an 8-year-old who is 125 cm tall. The draft picked the 20-inch bike from the size chart. | It added "helmet included", which the included-items file doesn't say for children's hires. Deleted, and "fit checked on arrival" added. |
| Wed | Four Thursday bookings have unsigned riders, eight in total. The draft wrote one reminder per booking. | On the six-rider booking it said "4 still to sign" when four had signed and two hadn't. The sheet now supplies a computed Missing column. |
| Thu | Thunderstorms are forecast for Friday afternoon. The draft wrote a go-ahead note for morning slots and move-or-credit offers for six afternoon bookings. | Counting first choices and fallbacks, six bookings pointed at Sunday 10:00, which together needed 10 hybrids against 9 free. The person added a first-to-confirm rule and placed holds before sending. |
| Fri | A customer replies: "Can I have a refund instead of a new date?" The draft promised a refund within five working days. | The terms file says credit, or a refund at the shop's discretion. The person rewrote the reply and kept the discretion. |
| Sat | At 07:40 a guest asks whether the 09:00 ride is still on. The draft said yes, mentioned wet saddles and suggested waterproofs. | Nothing. The draft used the staff note typed in ("going ahead, light rain") and made no forecast of its own. |
| Sun | Someone asks about renting a trailer for a toddler. The draft said the shop would check and added a STAFF CHECK line. | The shop doesn't hire trailers. The person added "Not offered: trailers" to the project file, so the next draft answers on its own. |
Monday's draft is worth reading in full, because it shows why the working is in the instructions. This was the first version of the reply:
Hi [first name],
Thanks for getting in touch. When we checked at 08:10 on Monday, Saturday at 10:00 had 3 hybrids, 2 e-bikes and 1 child seat free for the full three hours, so we can't fit four adult bikes and both seats at that time. Two options:
A) 10:00 start: 3 hybrids + 1 e-bike + 1 child seat. 4 bikes x $15 = $60, plus 1 seat $5 = $65.
B) 09:00 start: 4 hybrids + 2 child seats. 4 x $15 = $60, plus 2 seats x $5 = $10, total $70.
Neither is held yet. To hold your choice, pay the deposit here: [deposit link]. Availability can change before then.
[your name], [shop name]Option A priced the e-bike as a hybrid, so the $65 should have been $80 (3 x $15 + 1 x $30 + $5). Without the working beside it, the customer would have been quoted $15 too little and the argument would have happened at the counter on Saturday. Option B was right, and it is also the cheaper way to fit the whole group, which is worth saying to the customer.
Tuesday's slip is subtler. The size chart said a 20-inch bike for riders of 115 to 130 cm, so suggesting that bike for a 125 cm child was correct. That chart is invented for the example; use your own supplier's. The made-up part was "helmet included", borrowed from what rental shops typically do. Watch for this kind of mistake in every rental draft: the AI fills a gap with whatever is normal in general. The cure is a short included-items list and a not-offered list in the project files. If you collect heights and sizes before arrival, the sizing-form approach used in ski hire carries over to bikes.
Wednesday's slip came from subtraction. The AI was given riders and signed counts for each booking and asked for the shortfall, and on the six-rider booking it wrote four where it should have written two. AI tools can stumble on simple counting even when the numbers sit in front of them, so a better prompt is the wrong fix. Compute the missing count in the sheet, riders minus signed, and hand the AI only that column.
Thursday's slip is the one that costs money, so the weather section below works it through with the sums.
Waiver reminders: chase the signatures, never judge the wording
The wording of a rental waiver, who has to sign it, what it covers and whether it holds up all depend on where you trade and what you hire out. That is a job for a qualified adviser, not a chatbot. If you want a first draft to take to a solicitor, the solicitor's checks on an AI-drafted gym waiver list what to ask, and most of it carries over to bikes.
What AI does well is the chasing. A signed waiver is a yes or no for each rider, and the sheet that tracks it is simple: one row per booking, with the booking reference, the booker's first name, the ride date and time, riders, signed, and a Missing column that subtracts one from the other (in a spreadsheet, something like =D2-E2). If your booking tool signs waivers and links them to bookings, the Signed column can come from its export. Checkfront lists waivers as built in, from $99 a month plus a 3% online fee, and Bookeo charges from $9 a month for its digital waiver add-on (200 waivers), on top of a booking plan. Confirm both on the vendor pages before you rely on them. The tutorial on e-bike rental software for bookings and waivers goes further into choosing one.
Four moments matter, and only the middle two need the AI:
| When | What goes out | What you check first |
|---|---|---|
| Booking confirmed | Waiver link, one per rider, with the age rule for under-age riders | The link opens on a phone, not just a laptop |
| 48 hours before | "X of Y riders still to sign", with the link | The Missing count matches the sheet |
| Morning of the ride | A short nudge: no bike is handed over until everyone has signed | Late additions to the party |
| At the counter | Tablet or QR code, staff confirm each name | Nothing for the AI to do here |
For the 48-hour reminder, paste the rows that still have a gap and tell the AI what the message must and mustn't say:
Using only the rows below, draft one reminder per booking where Missing is above 0. Skip the others.
Ref | Booker | Ride | Riders | Missing
T-102 | [first name] | Thu 10:00 | 4 | 2
T-103 | [first name] | Thu 10:00 | 2 | 2
T-104 | [first name] | Thu 13:00 | 6 | 2
T-105 | [first name] | Thu 16:00 | 2 | 2
Each message: say how many riders still need to sign, repeat the booking ref, give the waiver link [link], say bikes can't be handed over until every rider has signed, and say that riders under the age on the form need a parent or guardian to sign. Do not describe what the waiver says.A usable draft for the third row looks like this:
Hi [first name], a reminder for booking T-104 on Thursday at 13:00. Two of your six riders still need to sign our waiver: [link]. We can't hand bikes over until everyone has signed, and riders under the age on the form need a parent or guardian to sign for them. Reply here if the link won't open. [your name], [shop name]There is a trap in group bookings. The person who gets your reminder is the booker, and the booker isn't the one who needs to sign. Ask them to forward the link to every rider, or collect an email for each rider at booking so each gets their own link. Whether one person may sign for others is a question for your adviser. I wouldn't accept "I'll sign for everyone" unless your waiver says it is allowed.
Keep personal data out of the prompt. The AI needs the reference, the first name, the counts and the times. It never needs the signed forms, which can hold medical details, and it should never be asked to summarise one. For the reminder timing itself, the patterns in reducing no-shows with AI reminders apply almost unchanged.
Weather rebooking: set the rule first, then write to six people
Weather is the job where drafting help saves the most time, and also where it is most tempting to let the AI decide. Don't. The go or no-go call is yours, informed by a forecast source you trust, your insurer's conditions and what the sky is doing. Once you have decided, AI is good at the part that hurts: writing six different messages in a hurry, each with a sensible new date.
Start by writing the rules down, so that the decision on the day is a lookup. These thresholds are made up for illustration. Set yours from your own risk assessment and what your insurer expects, and have the person who calls it named in the table:
| Condition in the forecast | Default action | Message type |
|---|---|---|
| Lightning in the ride window | No bikes out. Customer chooses a new slot or a credit, at no charge | Move or credit |
| Steady rain, no lightning | Ride goes ahead. Free move on request if asked a day ahead | Going ahead, what to bring |
| Gusts above a limit you set (say 50 km/h, about 31 mph) | Exposed routes closed. Shorter route or a move | Route change |
| Frost or ice at pickup | Start delayed until staff confirm the route is clear | Late start |
| Heat above a limit you set | Shorter rides, water on every bike | Heat plan |
Fix two decision times as well, for example 17:00 the evening before and 08:00 on the day, and say them in every booking confirmation: "If the weather changes our plans, we'll message you by 08:00." Customers who know when to expect the call are less likely to ring the shop at 07:15.
When the call is made, paste four things into the chat: the forecast text with its source and issue time, your decision per slot, the snapshot of free slots on the days you can offer, and the affected bookings (reference, first name, bikes needed). Ask for a table showing a first choice and a fallback for each booking, with the sums. In the invented week, Friday's 14:00 bookings needed 10 hybrids, 3 e-bikes and 2 children's bikes, and the AI proposed this:
| Booking | Bikes needed | First choice | Fallback |
|---|---|---|---|
| F-201 | 2 hybrids | Sat 09:00 | Sun 10:00 |
| F-202 | 3 hybrids | Sun 10:00 | Sun 13:00 |
| F-203 | 1 hybrid, 1 e-bike | Sat 09:00 | Sun 10:00 |
| F-204 | 2 hybrids, 2 children's | Sun 10:00 | Sun 13:00 |
| F-205 | 2 e-bikes | Sat 09:00 | Sun 10:00 |
| F-206 | 2 hybrids | Sun 10:00 | Sun 13:00 |
Now check it yourself, because this is exactly the kind of sum the AI can get wrong. In Thursday's fresh snapshot (Monday's group never paid its deposit, so nothing was ever held for it), Saturday 09:00 shows 5 hybrids, 3 e-bikes and 6 children's bikes free. First-choice demand is 3 hybrids (2 + 1) and 3 e-bikes (1 + 2), which fits, with the e-bikes filled to the last one. Sunday 10:00 shows 9 hybrids, 4 e-bikes and 7 children's bikes free, against first-choice demand of 7 hybrids (3 + 2 + 2) and 2 children's bikes. That fits as well. The risk is the fallback column: if all six bookings that list Sunday 10:00 as a first choice or a fallback ended up there, they would need 10 hybrids against 9 free. So the rule is that the first to confirm gets the slot, and the shop should place holds in the booking system before the messages go out. A Saturday slot filled to the last e-bike is also the one a new online booking will take while your messages are still unread.
The message itself should state the decision and the choices, and nothing about the weather beyond the forecast you pasted:
Hi [first name],
Thunderstorms are forecast for Friday between 14:00 and 18:00, so we won't send bikes out for your 14:00 slot. You have two choices, at no charge:
1) Move to Saturday 09:00. We've pencilled your 2 hybrids in until 18:00 today.
2) Keep a credit for the full amount, valid for [period from your terms].
Reply 1 or 2 and I'll confirm. If another day suits better, tell me and I'll check.
[your name], [shop name]The credit period and any refund rights come from your terms file, not from the AI. What you owe a customer when you cancel for weather depends on your terms and on consumer rules where you trade, so ask your adviser once and write the answer into the file. Operators who run guided rides or boats face the same call, and the workflows in weather cancellations for tours and kayak rental software show how they handle the messaging.
Five ways this slips in a bike shop, and the first sign of each
- A stale snapshot. The first sign is a reply quoting a build time from the morning, or two bookings for the last e-bike. Rebuild before each batch of replies, keep the "available when we last checked" wording, and hold a slot only when the deposit lands.
- Invented inclusions. The first sign is a phrase you never wrote, such as "helmet included" or "insurance covered". Keep included-items and not-offered lists, and add every surprise to them as it appears.
- Weather talk. The first sign is "should be dry by noon" in a draft. The instructions ban predictions, and a quick scan for "should", "likely" and "expected" catches the rest. Quote only the forecast you pasted.
- Arithmetic slips. The first sign is a total that doesn't match its working, or a missing-waiver count that disagrees with the sheet. Use computed columns, show the working, and spot-check two lines in each batch.
- The wrong tone on a bad day. The first sign is a cheerful draft to someone whose ride has been cancelled twice. Make a second cancellation, or any complaint, a phone call. Treat the HUMAN ONLY rule as a trigger for a call, not just for silence.
The same habit applies to anything that arrives through an automated receptionist or chatbot. If a wrong promise does go out, the tutorial on putting right a booking an AI receptionist got wrong covers the repair.
Four numbers that show it is working after a month
| Number | Where it comes from | What good looks like |
|---|---|---|
| Time to first reply | Inbox timestamps for ten availability enquiries; take the middle value | Falling, without the error count rising |
| Promises you had to withdraw | Double-booked slots, wrong prices or wrong inclusions you had to correct | Zero for the month. Any one means checking the snapshot routine |
| Riders unsigned at the counter | A tally on the counter sheet each week | Falling as the reminders settle in |
| Weather-day outcomes | Of the affected bookings: moved, took credit, or went quiet | You know the split. "Went quiet" is the group to ring |
Time saved is worth measuring honestly rather than assuming. On a busy day with 20 enquiries, 4 minutes each by hand comes to 80 minutes. If a checked draft takes 1.5 minutes, the same 20 take 30, a saving of 50. Those two timings are placeholders: time ten of your own replies with a stopwatch before and after, and use what you see.
Tool costs for a small rental shop
This works on free plans for a small shop, because the heavy lifting is the snapshot and the rules, not the model. Prices below are list prices as of October 2026; vendors revise them often, so confirm each on the vendor's own page.
| Item | List price | Notes |
|---|---|---|
| ChatGPT or Claude, free plan | $0 | Limits change. Enough to draft a day's replies. Look at the privacy setting that controls whether chats are used for training |
| ChatGPT Plus or Claude Pro | $20 a month (Claude Pro $200 a year) | Larger limits and more project files, for one user |
| ChatGPT Business or Claude Team | $25 per seat on monthly billing ($20 on annual), two seats minimum | Content isn't used for training by default. Poor value for a one-person shop |
| Snapshot and waiver sheet | $0 | Google Sheets or Excel |
| Waiver tool (optional) | Checkfront $99 a month plus 3% online fee; Bookeo waiver add-on from $9 a month, on top of a Bookeo plan from $14.95 | Worth it once manual chasing takes more than an hour a week |
Whichever plan you pick, the habit that protects customers is what you paste. Booking references, first names, counts and times are enough. Leave emails, phone numbers, payment details and signed forms out of every prompt.
Questions bike rental owners ask before trying this
Can the AI see my booking calendar?
Not unless you connect it, and I'd start without that. Pasting a dated snapshot takes ten minutes, shows you exactly what the AI was given, and lets you trace a wrong answer to a stale list instead of a hidden integration. Consider connecting a calendar only after a month of clean results, and keep that connection read-only.
Can AI write my rental waiver?
It can produce a plain-English first draft and a list of questions for a solicitor, but the wording, who must sign and whether it holds up all depend on where you trade and what you hire out. Have a qualified adviser approve the final text and keep the AI to tracking and reminders.
What if a rider arrives without a signed waiver?
Follow the rule printed in your confirmation: no bike until the form is done. Keep a tablet or a QR code at the counter so signing takes a couple of minutes, and let staff say no without apologising. Count these cases weekly; a rising number means your reminders are landing too late.
Should a chatbot on my website do this automatically?
A bot that answers fixed questions about hours, prices and what to bring is a different job from one quoting live availability or promising refunds. Until your booking system feeds it live counts, keep a person sending availability and rebooking replies. If you do run a public chatbot, tell visitors at the start that it is one.
Further reads
- How to Start a Bike Tour Business: Routes, Pricing and Bookings With AI — If you also run guided rides, this covers routes, pricing and booking flow.
- Repair Shop Software for Bike Shops: Service Bookings and Quotes With AI — The repair-counter side of a bike business, with service bookings and quotes.
- Campervan Rental Software: Handover Checklists and AI Enquiries — Another hire business with handover checks that matter at the counter.
- AI for a Car Rental Business: Enquiries, Availability and Damage Notes — Enquiry replies and damage notes for fleets where a wrong yes is costly.
- Can AI Handle Fishing Charter Bookings and Weather Call-Offs? — Weather call-offs and rebooking when a trip depends on conditions.
- AI for Golf Course Management: Can It Plan Staffing Around Weather? — Planning staff and slots around the forecast.
- Go-Kart Track Software: Race Bookings, Waivers and AI Enquiries — Group bookings and waivers for a venue that sells timed sessions.
- Ski Rental Software: Sizing Forms, Bookings and Returns With AI — Sizing forms and returns for another equipment hire business.
- Dive Shop Software: Courses, Gear Hire and Waivers Handled With AI — Map AI help to course sign-ups, kit hire logs and waiver handling in a small dive shop, and keep safety calls with a human instructor.
- Tour Booking Software Compared: What Small Operators Should Check — How to compare tour booking software on booking rules, payouts, OTA sync, waivers and fees, with a cost sum for a small operator.
- 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.
- Can AI Draft an Escape Room Waiver? What a Solicitor Must Check — Draft an escape room waiver with AI, with clauses for the emergency exit, health warnings and minors, then see what a solicitor must check before use.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Checkfront and Bookeo pricing pages, plus OpenAI and Anthropic help and pricing pages, as confirmed in late October 2026. The shop, rates, thresholds and messages in the examples are illustrative.