Ten short ChatGPT prompts cover a dealer's main follow-up work: a first reply to a web enquiry, a test-drive follow-up, a three-message ladder for quiet quotes, a best-price reply, three trade-in messages and three review jobs. Each uses only first names and the vehicle facts you paste, and ends with a check before you send.
These are drafts that you read and send under a salesperson's name. ChatGPT will happily invent a feature, a discount or a deadline, so every prompt tells it to use only the facts you paste and to write [CHECK: ...] where something is missing. The sample outputs below show the shape of a good result. They are illustrative, not captured runs, which is why the last section gives you a three-case test to run on your own enquiries before any prompt goes near a real customer.
The wider rules for what a dealer must never promise in writing sit in ChatGPT for car sales: follow-ups, listings and what never to promise, and the data side is in what customer data never to paste into ChatGPT. This set is the short, ready-to-use version.
Set the rules once, and keep the customer out of every prompt
Every prompt below starts from the same rules block. Put it in a ChatGPT Project's instructions, which apply only inside that project and override your global custom instructions, or paste it above each prompt. Add two or three real messages you are proud of as tone samples, with names removed, because "friendly" means different things in different showrooms.
DEALER RULES
You write short messages for a small independent car dealership. Plain English, friendly, no jargon, no exclamation marks.
Never: promise a price, discount, finance rate or approval; say a car is "perfect" or "problem-free"; mention other customers; use pressure lines ("only one left", "today only") unless I tell you they are true.
Always: use only the vehicle facts I give you; answer the question asked first; end with one clear next step; use the customer's first name only; sign off "[salesperson first name], [dealership name]".
If a fact you need is missing, write [CHECK: what is missing] instead of guessing.Then decide what never goes in. Remove the surname, phone number, email address, home address, driving licence details, date of birth, bank, finance and credit details, and anything about the customer's health or circumstances. Keep the registration plate and VIN out too: treat them as personal data, because they point to one person's car, and use year, make, model and trim instead. Where a customer's message contains any of these, cut them before you paste. On a personal plan, also switch off "Improve the model for everyone" under Data controls; ChatGPT Business does not train on business content by default.
One more rule belongs outside ChatGPT: message only people who asked to hear from you, honour any request to stop, and keep a record of it. The prompts draft words; they do not decide who you may contact.
The thread these examples follow
Prompts 1 to 7 follow one made-up customer, Customer A, through a fortnight. Customer A enquires about a 2019 compact SUV listed at $17,450, books a test drive, goes quiet, asks for the best price, and wants to trade in an old hatchback. The dealership's appraisal comes in at $6,200 against the $7,000 Customer A hoped for, so the balance on the SUV, before fees, taxes and finance, would be $17,450 minus $6,200, which is $11,250. Those numbers recur, so you can see how the drafts handle them. Prompts 8 to 10 cover reviews and stand on their own.
Prompt 1: First reply to a web enquiry about one car
Use this when an enquiry lands about a specific car. It targets the commonest weakness in a quick reply: a greeting, a thank-you, and then no answer to the question that was asked. You give it the facts of the car, the customer's message with the identifying details cut out, and your showroom hours.
Task: write the first reply to this web enquiry.
Vehicle facts: [year, make, model, trim, colour, price, mileage, fuel type, service history summary, anything the customer asked about]
Enquiry (first name only): "[the customer's message, with surname, phone, email and plate removed]"
Showroom hours: [days and hours]
Output: an email under 110 words and a text under 300 characters. Answer the question first. Offer one next step: a call, or a viewing at a specific time.Illustrative output for Customer A, who asked whether the SUV is still available and has a full service history:
Hi [first name], yes, the 2019 compact SUV is still here, and it has a full service history, last serviced at [mileage]. I can put the file on the desk for you to look through. Would a viewing on [day] at [time] suit you, or shall I call you after 5? [salesperson first name], [dealership name]
Check before sending: that the car really is unreserved, that "full service history" matches the paperwork in the file, and that the viewing slot is genuinely free. Delete any flourish the model adds, such as "it's a fantastic family car", because you gave it no fact to base that on. And if the customer did not ask for a call, do not ask for their phone number; the reply can happen in the channel they used.
Prompt 2: Same-day test-drive follow-up that uses their words
A generic "thanks for coming in" is forgettable. A message that repeats the one worry the customer voiced, and offers a concrete way to settle it, gets read. This prompt needs you to jot the customer's own words in the minute after they leave, which is the only fiddly part.
Task: write a follow-up to a test drive that happened today.
Vehicle facts: [year, make, model, trim, price]
What the customer said, in their words: [for example: "liked the seat height", "worried the boot is small for a pushchair"]
Next step I can offer: [for example: "bring the pushchair in to try the boot", "look through the service file"]
Output: a text under 300 characters and an email under 100 words. Mention one thing they said. Do not push for a decision.Illustrative output:
Hi [first name], thanks for coming in today. You mentioned the boot might be tight for the pushchair, so bring it in any time and we'll try it properly. I'm in until 6 tomorrow. [salesperson first name], [dealership name]
Check before sending: the draft must not say the pushchair will fit, only that you will test it. If the customer raised no worry, leave that line blank and the prompt falls back to a thank-you and the next step; do not let the model invent a concern to sound attentive. And read the sign-off times against your real rota, since "until 6 tomorrow" is a promise someone has to keep.
Prompt 3: A three-message ladder for a quote that has gone quiet
Most silent quotes get either nothing or a string of "just checking in" messages that give the customer no reason to reply. The rule that fixes this is that each message must add something new and true. So the prompt asks you for one real thing per message before it writes anything.
Task: write three follow-ups for a customer who has not replied to my quote.
Vehicle facts: [year, make, model, price]
Last contact: [what happened and when]
New things I can honestly offer, one per message: [message 1: ..., message 2: ..., message 3: ...]
Output: three texts, each under 280 characters, for day 3, day 7 and day 14. Each must add something new. Never write "just checking in". Message 3 gives an easy way to say no and says I will stop messaging.Illustrative output for Customer A, who went quiet after the test drive:
Day 3: Hi [first name], I've added the full service record for the SUV to its listing, in case it helps you decide. Happy to answer anything. [name], [dealership]
Day 7: Hi [first name], I can send a short walk-round video of the SUV, including the tyres and the boot. Want me to send it? [name]
Day 14: Hi [first name], I'll assume you've gone another way, which is completely fine. If you'd like me to look out for something similar, reply "yes". Otherwise I'll stop messaging. [name]
Check before sending, and decide the stop rules in advance. Any reply, even "not yet", ends the ladder and starts a human conversation. "Bought elsewhere" or "please stop" ends it for good, and you log that. Only send the day-7 message if you really have the video; the prompt will write the offer whether or not you do. The gaps of 3, 7 and 14 days are a suggestion, not a rule, so stretch them if your customers take longer to decide. What matters is that no message exists only to remind them you are there.
Prompt 4: Replying to "what's your best price?" without conceding
This is the prompt where the model is most dangerous, because a helpful assistant likes to offer a small discount to keep the conversation friendly. Your policy has to come from you, so the prompt makes you state it, and forbids any number other than the listed price.
Task: reply to a customer asking for our best price.
Vehicle facts: [year, make, model, listed price, what the price includes, such as warranty or a service before handover]
My position (choose one and delete the other): [A: the price is firm] [B: I will pass a written offer to the manager and reply within 24 hours]
Output: a text under 300 characters and an email under 100 words. Do not mention any price, discount or figure other than the listed price. Explain what the price includes. End with one next step.Illustrative output for Customer A under position A:
Hi [first name], thanks for asking. The SUV is $17,450, and that includes [what the price includes]. I can't move on the price, but I can look at what we'd offer for your trade-in. Shall I book a quick appraisal this week? [salesperson first name], [dealership name]
Check before sending: no figure other than $17,450 appears, and anything the model says about what the price "includes" is on your sheet and not its guess. If it writes "after checking comparable cars", keep that only if you really did, because it is a claim about how you priced the car. Position B needs an actual manager and an actual 24-hour reply, or you are promising a process you do not run.
Prompt 5: Trade-in enquiry reply that asks before it answers
"What would you give me for my car?" arrives long before anyone has seen the car. The honest reply is a list of what you need, not a number. This prompt writes that list and forbids any figure, including a vague hint such as "probably around". The tutorial on trade-in enquiries before the appraisal covers the process around it; this is the message.
Task: reply to a trade-in enquiry. Do not give or hint at any value.
Customer's message (first name only, plate and VIN removed): "[message]"
What I need before an appraisal: [year, make, model and trim; mileage; condition including dents, scratches and warning lights; service history; number of keys; whether any finance is still owed; photos]
Output: an email under 130 words with the list as short bullets. Say the figure is confirmed after an inspection. Offer one next step: send photos, or book a 20-minute appraisal.Illustrative output for Customer A, who asked about a 2013 hatchback:
Hi [first name], happy to help. I can't give a figure without seeing the car, but these details let me start the appraisal: year, make, model and trim; mileage; any dents, scratches or warning lights; service history and number of keys; whether any finance is still owed; and photos of the front, back, both sides, the dashboard showing the mileage, and the interior. The final figure is confirmed after an inspection. Send what you have, or I can book a 20-minute appraisal. [salesperson first name], [dealership name]
Check before sending: scan the draft for any number that is not a time or a count. The reply that comes back from the customer will contain the sensitive items, especially anything about finance owed and perhaps the plate, so keep that reply out of ChatGPT and out of later prompts. If you need to draft the next message, summarise the car in your own words: "2013 hatchback, mid mileage, service history partly missing".
Prompt 6: Explaining a trade-in figure lower than the customer hoped
This is the hardest message in the set, and the one where a made-up reason does the most damage. So the prompt may use only the reasons in the appraiser's notes, with their costs, and must show whether they add up to the gap.
Task: explain a trade-in figure to a customer who expected more.
Appraised figure: [$6,200]. Customer's expectation, from their message: [$7,000].
Reasons from the appraiser's notes, with costs (use only these): [for example: "two tyres worn, workshop estimate $320"; "rear bumper scratch needing repair, estimate $280"; "service history gap, we will service before sale, allowance $200"]
Options I can offer: [for example: "re-inspect if you have the missing service invoices"; "you may prefer to sell privately"]
Output: an email under 150 words. Acknowledge the gap. Give each reason in plain words with its cost. Say whether the reasons add up to the gap. Do not blame the customer or criticise the car. Offer the options.Illustrative output:
Hi [first name], I know $6,200 is below the $7,000 you were hoping for, so here is exactly how we got there. Two tyres are worn, and our workshop estimates $320 to replace them. The rear bumper has a scratch that needs repair, estimated at $280. The service history has a gap, so we would service the car before sale, and we have allowed $200 for that. Together that is $800, which is the difference between your figure and ours. If you have the missing service invoices, I'm happy to re-inspect with them in front of me. And if you would rather sell privately, we understand. [salesperson first name], [dealership name]
Check before sending: the sum. $320 + $280 + $200 is $800, and $7,000 minus $6,200 is $800, so it closes. If it did not, the prompt should say so, and you should read that as a gap in the appraiser's notes rather than something to smooth over. Send this after a phone call, not instead of one, so the written version confirms what the customer has already heard. Delete the "sell privately" line if you would rather not say it. And never let the model add a "the market is soft" style reason unless the appraiser's notes contain it.
Prompt 7: A trade-in offer reminder with a real deadline
Appraisal figures should expire, because cars and markets change. A reminder is fair when the deadline is real and stated plainly, and manipulative when the model dresses it up. This prompt bans urgency words and states what happens afterwards.
Task: remind a customer that their trade-in appraisal is about to expire.
Facts: figure [$6,200]; valid until [date]; condition of the figure [the car matches the description and photos]; what happens after [we re-appraise, and the figure may go up or down]
Output: a text under 280 characters and an email under 90 words. State the deadline plainly. No urgency words. Offer one next step.Illustrative output:
Hi [first name], a note that the appraisal of $6,200 for your hatchback is valid until [date], provided the car matches the description. After that we would re-appraise, and the figure could move either way. Happy to talk it through. [salesperson first name], [dealership name]
Check before sending: the deadline appears in the terms you gave the customer in writing, not just in your head. "Either way" stays only if you would genuinely re-appraise upwards when the car and market justified it. Strike any "don't miss out" or "last chance" the model adds despite the instruction. And put a reminder in your own diary to do the re-appraisal, because a reminder message that promises one and then does not deliver is worse than none.
Prompt 8: Replying to a five-star review without repeating yourself
Happy reviews deserve a reply, and the trap is sameness: forty replies that all open "Thank you for your kind words". The fix is to paste your last three public replies so the model can avoid them, and to make it use one detail from the review.
Task: reply to this five-star review.
Review text (reviewer's surname removed): "[text]"
Staff mentioned in the review, first names only: [names, or "none"]
My last three public replies, so you do not repeat them: [paste]
Output: one reply of 35 to 50 words. Mention one specific detail from the review. Thank any staff member named in the review. No sales pitch, no request for another review, no "we look forward to seeing you again".Illustrative output for a review that said the buyer's handover was smooth and the salesperson did not rush them:
Thank you, [first name]. I'm glad the handover was smooth and that [salesperson first name] took the time to explain everything without rushing you. Enjoy the hatchback, and do get in touch if anything needs a look. [dealership name]
Check before posting: that the staff name appears in the review itself, because a thank-you to someone the reviewer never mentioned is invented. Make sure the detail it picked is in the review and not an inference. Then remember how replies work: Google reviews them before they appear, which usually takes up to 10 minutes but can take up to 30 days, so do not post the same reply twice because it has not shown up yet.
Prompt 9: Replying to a one-star review without arguing
The urge is to defend the dealership. The better reply is short, calm and moves the conversation off the public page. This prompt works from your own two-sentence summary of the facts, so the model has nothing to embellish, and it forbids any offer of compensation.
Task: draft a public reply to this one-star review.
Review text (surname removed): "[text]"
What my records show, in two neutral sentences: [facts only, no opinions]
Who will follow up: [first name and role]
Output: one reply of 60 to 90 words. Say sorry for the experience described, not for anything we have not confirmed. Do not argue, do not say or imply the customer is lying, and do not reveal any price, finance or personal detail. Invite them to contact [dealership phone number or email address] and ask for the named person. Offer no compensation and do not ask them to change or remove the review.Illustrative output for a review complaining that the trade-in figure was lower than the online estimate and the explanation felt rushed:
Hi [first name], I'm sorry the trade-in figure came in lower than you hoped and that the explanation felt rushed. That isn't how we want anyone to feel. [manager first name], our [role], would like to go through the appraisal with you properly. Please call [dealership phone number] or email [dealership email address] and ask for [manager first name]. [dealership name]
Check before posting: the reply confirms nothing you have not verified, and it contains no detail a stranger could use to identify the sale. Google's policy bars offering incentives such as discounts or free services in exchange for reviews, and says this extends to revising or removing a negative review, so a "we'll make it up to you if you update your review" line is not just unwise but prohibited. If the review alleges something serious, such as a safety fault, a misdescribed car or discrimination, do not draft with ChatGPT at all. Take advice first.
Prompt 10: Asking for a review after delivery, with no gating
Asking is fine; Google's own help page says you can share a review request link or QR code. The rules are about how. Its policy bars offering incentives such as payment, discounts or free goods, selectively soliciting positive reviews, discouraging negative ones, and asking staff to collect a set number of reviews. So the prompt writes one neutral message that goes to every buyer.
Task: write a review request to send to every customer seven days after delivery.
Facts: first name; the car [year, make, model]; review link [paste your Google review link]
Output: a text under 260 characters and an email under 70 words. Ask for an honest review of their experience. Offer nothing in return. Do not mention star ratings. Do not ask only happy customers. Do not ask them to mention a salesperson, the car or any other detail. Do not ask them to change or remove any review.Illustrative output for a buyer a week after collecting their car:
Hi [first name], hope the car is settling in. If you have a minute, we'd value an honest review of your experience: [review link]. Whatever you tell us helps us improve. [salesperson first name], [dealership name]
Check before sending: the message goes to every buyer, not a list you have filtered by who seemed pleased, and it appears alongside no gift, no discount and no hint at a rating. Do not give the sales team a target number of reviews, because soliciting a set number is itself in Google's list of prohibited requests. The same list covers asking for reviews that include specific content, such as a staff member's name, so a draft that adds "and do mention [salesperson first name] if they helped" gets that line cut. Test the link on a phone before the first send. And look at the samples and rules for responding to dealership reviews for the other side of this: what to do when the reviews arrive.
Run each prompt on three past cases before you trust it
The sample outputs above are illustrations, not captured runs, so treat the set as untested until you have tried it on your own enquiries. For each prompt, pick three past cases with the details removed: an easy one, an awkward one, and one with a fact missing. Score each output one point per check.
- It answers the question the customer actually asked, first.
- Every fact in it comes from your vehicle sheet or notes.
- It makes no promise you would refuse to put in writing.
- It ends with exactly one next step.
- It sounds like you.
The pass mark is four out of five on all three cases, and any failure on the third check fails the prompt outright. Ten minutes per prompt is enough, so the whole set is an afternoon spread across a week. A typical thing the test turns up: the best-price prompt handles the easy case, but in the awkward one the customer quotes a rival's price and the model, being agreeable, writes "we'd be happy to look at matching that". That sentence is a promise. The fix goes into the rules block as a new line: "If the customer mentions another dealer's price, do not respond to it. Return to what our price includes."
Keep a one-line log: the date, the model name shown in ChatGPT's picker, and pass or fail per prompt. Re-run the set whenever the model name changes or you edit the rules block. Every edit you make by hand more than twice, such as deleting "I hope this finds you well", is a candidate for a new line in the rules.
Finance, complaints and auto-send: what this set leaves out
Four things are deliberately missing. Finance quotes, rates and approvals, warranty terms and consumer-rights wording need text approved by your finance provider or adviser, so draft around them with [CHECK] placeholders. Complaint handling beyond the first public reply belongs with a person. Anything that touches a customer's circumstances, such as bereavement, redundancy or illness, should never be turned into marketing. And sending: these are drafts a person reads. If you want replies to go out automatically, read answering a web lead in minutes with AI first, because the risk changes when nobody reads the message before it leaves.
For the first week, start with prompts 1, 2 and 8. They carry the lowest risk, because the facts are simple, nothing involves money, and you see the result quickly. In week two add 3 and 10. Leave 4, 6 and 9 until your rules block has settled and you have run the three-case test on them, because they are the ones where a wrong sentence costs real money or reputation. By then you will have a log of edits, and that log is the best guide to what your own rules block should say. The messages that follow a test drive deserve the same care in call notes, which AI tools for car sales: test-drive follow-ups, quotes and call notes covers.
Questions dealers ask about these prompts
Can ChatGPT write finance quotes or approvals?
I would not let it. Finance rates, approvals, warranty terms and consumer-rights wording need text that your finance provider or adviser has approved. Have the prompt write around them and leave [CHECK: finance wording] where they belong, then paste in the approved sentence yourself.
Is it safe to paste a customer's message into ChatGPT?
Only after removing the surname, phone number, email address, home address, registration plate, VIN and any finance details. On a personal plan, switch off 'Improve the model for everyone' in Data controls; business plans do not train on business content by default. First name plus the question asked is enough for every prompt here.
Will customers be able to tell a message came from ChatGPT?
Unedited output can read as generic, with stock warmth and invented enthusiasm. Tone samples in your rules block and a read-through before sending fix most of that. The rule that matters more is that every fact in the message, from the service history to the deadline, is one you have checked.
How often should I re-test the prompts?
Whenever the model name shown in ChatGPT's picker changes, and whenever you edit a prompt or the rules block. Keep a one-line log with the date, the model and a pass or fail. Models change often, and a prompt that behaved last month can drift without any warning.
Further reads
- How Can AI Help Car Dealerships? A 30-Day Plan for a Small Lot — A 30-day plan for where a small lot should use AI first.
- AI for Used Car Dealerships: Honest Listings and Photo Clean-Up — Honest listings and photo clean-up to pair with these replies.
- AI BDC for Car Dealerships: Does a Small Lot Need One? — Whether a small lot needs an AI business development centre.
- AI Chatbot for Car Dealerships: Rules to Set Before Launch — Rules to set before a chatbot talks to customers for you.
- AI for Used Car Sales: Price Each Car Against Local Listings — Price each car against comparable listings before you quote.
- AI Marketing for Car Dealerships: A Four-Week Plan for One Lot — A four-week marketing plan for one lot.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google's Maps user-generated content policy and Google Business Profile help on replying to reviews and sharing a review request link; OpenAI help pages on Projects; ChatGPT plan details from OpenAI's help pages (October 2026). All vehicles, prices and customer messages are invented illustrations.