Using AI for Dealerships: Lead Replies, Listings and Follow-Up Texts

A small dealership can use AI to draft first replies to online leads and write vehicle listings from the spec sheet, as long as the facts come from your own stock sheet and a person approves anything about price, finance, trade-in or condition. Text follow-ups go only to customers who agreed to them, on a short fixed schedule.

The risk in this work is not bad writing. It is a confident sentence that promises something you cannot deliver: a price that has moved, a finance rate nobody quoted, or "no accidents" with no history report behind it. The fix is structural: every reply and listing is built only from your stock sheet, and every sentence about money, condition or a commitment is left as a marked gap that a person fills before anything is sent.

Follow me on Instagram@sagnikteaches

Sort every statement into green, amber and red

Before any prompt, agree one rule for the whole team: AI may draft anything, but only some of it may leave the building without a person looking. A three-colour split makes that easy to apply under pressure.

Connect on LinkedInSagnik Bhattacharya

The assistant's instruction block should name the red list and tell it to write "[PERSON TO CONFIRM: price]" in the reply wherever one appears. That keeps the gap visible instead of papering over it. For a view of how this plays out with chatbots, the rules to set before a dealership chatbot launches are worth reading alongside.

Subscribe on YouTube@sagnikteaches

Build the stock sheet the assistant will read

Everything else depends on this. One row per vehicle, kept up to date, with the fields a customer asks about. For an illustrative small used-car lot:

stock_id: C-0412
type: hatchback, 5-door
make_model: (make) (model) 1.2 petrol manual
year: 2019
mileage: 41,200 (as at 3 Nov, from the service book)
colour: grey
owners: 2 (per the registration document)
service_history: full, last service at 38,900
options: parking sensors, alloy wheels, Bluetooth
known_marks: small scuff on rear bumper, repaired kerb mark on nearside rim
price: $11,450 (staff only; changes need manager)
status: available / reserved / sold
test_drive: weekdays 10:00-17:00 by appointment

The "known marks" row is what protects you. If the sheet lists the scuff, the listing can mention it and the reply can answer a condition question truthfully. If it does not, the assistant has nothing to go on and must say so. A bike or motorcycle dealer adds fields such as engine size, number of keys, last chain and tyre change, and any accessories fitted, and the idea is unchanged.

Step one: reply to a lead in minutes, not hours

A lead is an enquiry from your website, a marketplace listing or a social message. The speed of the first reply matters, so the goal is a draft ready for a salesperson within a minute of the message arriving. The first reply automation tutorial covers wiring this into a CRM; here is the content work.

You draft first replies to online enquiries for a small used-car
dealership. Use ONLY the stock sheet provided. Rules:
- Write 60-90 words, warm and plain, signed "[salesperson name]".
- Answer the questions asked. Do not add upsell.
- For price, discounts, finance, trade-in value, condition beyond
  the "known_marks" field, history, warranty or delivery dates,
  write "[PERSON TO CONFIRM: ...]" instead of an answer.
- If the vehicle's status is not "available", say we will check
  and get back to them: [PERSON TO CONFIRM: availability].
- Offer two test-drive slots from the sheet and ask for a phone
  number only if they prefer to be called.
- End with one line starting "NOTES FOR STAFF:" listing anything to verify.

An enquiry and an illustrative draft:

Enquiry: "Is the grey 2019 hatchback still there? Any marks? Could
I get it for $10,500 cash? Can I come Saturday?"

Draft: Hello [first name], thank you for asking about the grey 2019
hatchback. [PERSON TO CONFIRM: availability]. It has two small
marks: a scuff on the rear bumper and a repaired kerb mark on the
nearside alloy, and I can send photos. [PERSON TO CONFIRM: price].
We are open for test drives weekdays by appointment; if you would
like a Saturday visit, tell me and I will check what we can offer.
[salesperson name]

NOTES FOR STAFF: confirm availability and price; check Saturday
opening; customer asked about a cash price.

What you fix: add the answer to the price question before sending, or reply that you will call; confirm Saturday. The draft did the right thing by not agreeing $10,500. Take note of how much was done correctly: it answered the marks question truthfully because the sheet listed them, and it flagged the rest.

Step two: listing text from the spec sheet

Listings are where a sheet and a template pay off most. Paste one stock row and ask for a short title, a description and a facts block. The point is accuracy and consistency, not poetry.

Write a listing for this vehicle using only the stock sheet row.
Output: (1) a title under 70 characters; (2) a 100-word description;
(3) a facts list in the order: year, mileage, gearbox, owners,
service history, options; (4) "Known marks" in plain words.
Do not use "immaculate", "perfect", "no accidents", "one careful
owner" or any phrase that is not supported by a field. If a field
is blank, leave it out.

An illustrative result: "2019 grey hatchback, 1.2 petrol manual, 41,200 on the clock. Two owners and full service history, last service at 38,900 miles. Parking sensors, alloy wheels and Bluetooth. Known marks: small scuff on the rear bumper and a repaired kerb mark on the nearside rim. Test drives weekdays by appointment." That reads flat but it is true, and a listing a buyer can trust is more useful than one that sparkles. Add one human line, such as why you like the car, at the end.

Before (hand-written): "Lovely little car, great on fuel, must see." After: the facts block above. The first earns a question about every fact; the second answers them in advance, so fewer messages reach you. The used car listings tutorial covers photos, and the tutorial on AI listing mistakes lists the claims that most often get a seller into trouble.

Phrases to ban from every listing prompt

• "No accidents", "never been in an accident" unless you hold a history report and have read it.
• "Mint", "immaculate" or "perfect" for any used vehicle.
• "Full service history" unless the stamps or invoices are in the folder.
• "One careful owner" when the registration document shows two.
• Any finance, monthly payment or "from" price the assistant works out on its own.

Step three: the follow-up cadence

Most leads that do not reply the first time have not said no; they are comparing options. A fixed, short cadence keeps you in view without nagging. This is an illustrative schedule for a small dealership, and you should shorten it if customers tell you they do not want contact.

Ask the assistant to write the Day 3 and Day 14 texts, each under 300 characters and in your voice, with no pressure language. An illustrative Day 3 text: "Hello [first name], it is [salesperson name] at the dealership. Would a short video of the grey hatchback help you decide? Reply YES and I will send it. Reply STOP to stop messages." Notice what is absent: a deadline, a hint that someone else is about to buy it, a price. Pressure phrases are a common mistake in drafts and are easy to ban in the instruction block.

Consent for texts is a record you keep, not a box you tick once

A person who fills in a web form with a phone number has given you the number. Whether they have agreed to receive marketing texts is a separate question, and the rules differ by place. Ask a qualified adviser what applies where you sell, and keep this practical minimum regardless:

1. A clear tick box or opt-in wording on the enquiry form, not pre-ticked, saying what you will send and how often.
2. A record of the exact wording, the date and time, and where it was collected, stored against the lead.
3. Every text carries an easy way to stop, and a stop request is honoured the same day across all your systems.
4. A quiet-hours rule: no texts early in the morning or late at night.
5. No texts bought from a list or collected for a different purpose.

The assistant can help draft the opt-in wording and a short privacy line, but the person responsible for compliance should read both. If you are not sure whether a particular message is allowed, leave it out until you have an answer.

The main example in numbers: a small lot for a month

Take an illustrative used-car lot with 25 vehicles on the forecourt and one salesperson plus the owner. Over a month it receives about 60 online enquiries and adds around 25 vehicles to its stock. Before AI the owner measured:

• Writing a first reply: about 6 minutes each, so 60 times 6 equals 360 minutes, 6 hours.
• Writing a listing: about 20 minutes each, so 25 times 20 equals 500 minutes, about 8.3 hours.
• Average first-reply time during opening hours: about 5 hours, because replies waited until between customers.

After four weeks of drafts approved by a person: first replies take about 2 minutes to read, fix and send, so 120 minutes, a saving of 4 hours. Listings take about 8 minutes including checking the sheet against the car, so 200 minutes, saving 300 minutes or 5 hours. Average first-reply time falls to about 20 minutes because drafts are ready when the notification arrives. That is about 9 hours back in the month, and, more importantly, a quicker answer for every enquiry. These are illustrative figures; measure your own week before starting, as the small-lot business development centre tutorial suggests.

The same red list at a motorcycle shop, with different fields

A small bike dealer has the same pattern with different fields. Enquiries ask about the licence category needed, seat height, whether the bike has had a chain and sprocket change, and whether a helmet or luggage is included. The stock sheet gets those columns, and the instruction block adds two rules: never say a bike suits a rider's size or experience, and never comment on insurance cost. The assistant can still write a sound reply: "Seat height is 810 mm, listed in the specification. I can't judge fit from here, but you are welcome to sit on it in the shop." The honesty of that answer is the selling point.

The finance promise that nearly went out

Take an illustrative case: a dealer tried a generic chat assistant without a stock sheet. A customer asked, "Can I get this on finance for under $200 a month?" and the draft replied, "Yes, we have finance options from around $180 a month." The figure came from nowhere. The team's rule said red-list items must be marked, but nobody had put that rule into the instruction block, so the draft read like any other sentence. A salesperson caught it only because she knew no finance quote had been run for that customer.

The fix was in the instruction block and the habit: finance, price and trade-in are always "[PERSON TO CONFIRM]", and the salesperson searches every draft for the square bracket before sending. The search takes two seconds. If your CRM can add a rule that blocks sending a message that contains a bracket, even better.

A harder enquiry: the trade-in and the cash offer

Not every lead is a simple availability question. Take an illustrative message: "I have a 2014 estate to trade in, around 90,000 on the clock, it needs a clutch. What would you give me, and could you hold the grey hatchback until Friday?" Four red items sit in that one paragraph: trade-in value, condition, holding the vehicle and a date commitment. The draft should do three things: acknowledge, ask for what a person needs, and promise nothing.

Draft: Hello [first name], thank you for the details. To give you a
proper figure for the estate, [PERSON TO CONFIRM: trade-in value]
we would need to see it, so I can offer a short appraisal slot
on [two slots from the sheet]. Photos of the dashboard and the
service book would help in the meantime. Holding the hatchback
until Friday is [PERSON TO CONFIRM: reservation terms]; I will
call you today to talk it through.
[salesperson name]

NOTES FOR STAFF: trade-in vehicle needs clutch work; customer
wants a hold; call back today.

The assistant did what a good junior would do: it took the clutch comment into the staff note but did not repeat it back as a fact in the customer-facing reply, and it never offered a number. You then call, which a person does better than any draft, and the draft has saved the thinking time.

Handling out-of-hours messages and sold cars

Two situations catch dealers out. The first is the late-evening enquiry: nobody is there to approve a draft, so the customer waits until morning. An honest automatic acknowledgement works better than silence: "Thank you for your message. We are closed until 09:00; a person will reply with details of the vehicle first thing." It makes no promises and costs nothing. Keep it plain, name your opening time, and do not let an assistant answer the substance of the question overnight unless every part of it comes from a green item.

The second is the sold or reserved vehicle. Listings stay online for hours or days after a sale because someone forgets to take them down, and the assistant, reading a stale sheet, will happily say the car is available. The remedy is a single status column that changes when the deal is done, plus the rule that any reply on a vehicle whose status is not "available" says "we will check and confirm". If your lot sells a car a day, tidy the sheet at opening and at closing, a two-minute job.

Test the drafts on old enquiries before a customer sees one

You can try the whole set-up with no risk by running it on enquiries you already answered. Take ten real messages from last month, remove the customer's name and number, paste each into the assistant with your instruction block, and compare each draft with what you actually sent. Score each draft on three questions: did it state anything not on the sheet, did it miss a question the customer asked, and would you be comfortable sending it with one small edit? Aim for eight of ten passing before you use drafts live. Record the two that failed and add a rule to the instruction block for each: for example, "if the customer asks about a warranty, write [PERSON TO CONFIRM: warranty]". The shadow mode tutorial explains the method in more detail.

The same test works for listings. Take three vehicles you listed recently, generate listing text from the stock row, and compare it with what is live. Check whether any claim in the live version is not in the sheet. Often it is, and the check turns up an old habit such as "FSH" for full service history on a car whose stamps are patchy. Finding that in your own listings is cheaper than finding it from a buyer.

Keeping customer details out of the chat window

An enquiry carries a name, a phone number and perhaps a trade-in registration. The assistant does not need them to draft a reply. Replace the name with "[first name]" and leave the number out; add them when you paste the final text into your email or CRM. If a tool is going to read leads directly from your inbox or CRM, check what it stores and for how long, and whether it uses your messages to improve its models. Business plans for ChatGPT and Claude do not train on business content by default, though each needs two seats; individual plans have a privacy setting you can check yourself.

Working the approval into the day

Drafts only help if someone reads them promptly. A workable pattern for a team of two or three: the draft arrives in the shared inbox or CRM; whoever is free reads it, resolves the brackets, edits and sends; a daily five-minute review looks at three sent replies for tone and accuracy. Keep a short log of any draft that had to be corrected and why. After a month, the log tells you which rules to add to the instruction block.

Which tool? A general chat assistant with a saved Project works for a small lot: ChatGPT Plus lists at $20 a month, Claude Pro at $17 a month billed annually or $20 monthly. Your CRM may include its own drafting feature, and your dealer management system vendor may offer an add-on. Compare platforms only after your own drafting has shown what you actually need. Phone calls and voice are a different job.

Questions dealers ask before they switch AI on

Can AI answer a lead without a person checking?

For simple facts taken from your stock sheet, such as the colour, the gearbox type and the opening hours, a system can reply on its own once it has proved reliable. Anything about price, finance, trade-in value, condition or availability should wait for a person. Start with every draft approved, then relax only on the safest question types.

Should the reply say that it was written by AI?

Customers mind a wrong or evasive answer more than the tool, but honesty is the safer course. If a chatbot replies without a person, say so plainly and offer a way to reach one. For drafts that a salesperson reviews and sends under their own name, you are the author. Check any disclosure rules that apply where you sell.

Where does the AI get the vehicle facts?

From a stock sheet you maintain: one row per vehicle with make, model, year, mileage, colour, options, price, condition notes and status. Keep it in a spreadsheet or your dealer management system export. If a fact is not on that sheet, the assistant must say it does not know rather than guess.

How soon should a first reply go out?

As fast as you can do it accurately. Minutes beat hours for web enquiries, but a quick wrong answer is worse than a slightly slower right one. A drafted reply that a person approves within 15 to 20 minutes during opening hours is a realistic target for a small lot.

Further reads

• Responding to Dealership Reviews With AI: Samples and Rules — Sample replies to reviews, good and bad.
• AI Marketing for Car Dealerships: A Four-Week Plan for One Lot — A month of marketing for one lot.
• Can an AI Voice Agent Book Car Dealership Service Appointments? — Whether AI can book service appointments by phone.
• Mia AI for Car Dealerships: Questions to Ask Before You Sign — Questions to ask before signing for a dealer AI product.
• Car Dealership AI Receptionist: Monthly Cost and Hard Limits — Costs and limits of an AI phone receptionist.
• What to Do When AI Gets Something Wrong With a Customer — What to do when a reply goes out wrong.
• What to Check in an AI Tool's Privacy Policy and Terms — What to read before a tool touches customer data.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: openai.com ChatGPT and Business pricing, claude.com/pricing (checked 5 November 2026). Messaging-consent rules differ by place and are not quoted.