Give the AI only your spec sheet, inspection notes and history report, and forbid any claim that is not in them. For photos, allow cropping, light correction and a tidier background, but never remove, soften or recolour a dent, scratch, rust patch, wheel mark or interior wear. Keep every original photo on file.
Dealers rarely set out to mislead. AI makes the accident easy, because both its prose and its editing tools drift towards making a car look better than it is. A listing is a claim you will be held to when the buyer arrives, and a photo is part of that claim. The rules below are house rules I would adopt, not legal advice: check the advertising and consumer rules where you trade, and ask a solicitor or your trade body where a rule is unclear.
Write the honesty rules before you open any tool
A one-page policy does more than any tool setting, because it tells staff what "done properly" means. Seven rules cover most of the risk.
- Three sources only. Listing text comes from the spec sheet, the dated inspection notes and the history check. Nothing the AI remembers about the model in general.
- Unknown stays unknown. A blank field becomes "not confirmed: ask us", never a guess and never a typical figure.
- Opinions need an owner. "Drives well" is an opinion. It goes in only with the name of the person who test-drove the car and the date; otherwise it is deleted.
- Material facts always appear. Repaired damage, warning lights, missing service stamps, non-original parts, open recalls and anything that would change a buyer's decision belong in the listing, not behind a phone call.
- Numbers are copied, not retyped. Mileage, price, year, engine size, owner count and the size of each mark come straight from the documents.
- Photos show the car as it is today. Edits may tidy the setting. They may not tidy the vehicle.
- Originals are kept and edits are labelled. Every edited image has an original in a separate folder, and a replaced background is mentioned in the caption.
Print it and pin it where listings are written. Rules one to five govern the text, and rules six and seven govern the photos.
Get the spec sheet right first
An assistant can only be as honest as its documents. The commonest source of wrong facts is not the AI at all: it is a spec sheet filled in from a decoder or an old advert. Trim levels, option packs and engine variants decoded from a vehicle number can be wrong, and a previous advert may have carried the same mistake for months. So the person who handles the car fills the sheet while standing at the car, ticking each option they can see or switch on: the reversing camera, the heated seats, the parking sensors. Anything they cannot confirm stays blank, and blank becomes "not confirmed: ask us".
Keep the sheet to fields you can verify in a minute: year, engine, gearbox, body, colour as the manufacturer names it, mileage with a photo of the odometer, keepers, keys, service book status, equipment, warranty terms and price. Add the date and the name of whoever filled it in, because the sheet becomes evidence of what you knew when you wrote the advert. One more habit: when a car arrives as a part-exchange, copy nothing from the seller's own description. "Full history" from a seller is a claim to check, not a fact to paste.
From spec sheet to listing: a prompt that tags every sentence
The technique I would use is source tagging. Tell the AI to end every sentence with a tag naming the document it came from, and to delete any sentence it cannot tag. That turns checking from "does this sound right?" into "open the cited document and find the sentence", which takes seconds. Here is the prompt.
You write used-car listings for a dealer. Use ONLY the three documents below: SPEC SHEET, INSPECTION NOTES, HISTORY CHECK. Do not use anything you know about this model in general.
End EVERY sentence with one tag:
[SPEC] from the spec sheet
[INSPECTION date] from the inspection notes
[HISTORY date] from the history check
[OPINION name, date] only where the documents record a named person's opinion.
If you cannot tag a sentence, do not write it.
Rules:
- Copy numbers exactly: mileage, price, year, engine size, owners, sizes of marks. Never round.
- If a field is blank or unclear, write "Not confirmed: ask us." Never guess.
- Add a section MARKS AND CONDITION listing every mark, wear item or fault in the inspection notes, in plain words with sizes.
- Never use: mint, immaculate, perfect, flawless, as new, pristine, bargain, must see, no accidents, accident-free. Use "full service history" only if the documents say exactly that.
- Do not invent equipment, trim names or warranty terms.
SPEC SHEET:
[paste]
INSPECTION NOTES:
[paste]
HISTORY CHECK:
[paste]Now a worked example with an imaginary used hatchback, priced at $11,450. These are the three documents the dealer pastes in.
SPEC SHEET (copied from the dealer system, 12 October)
2019 compact hatchback, 1.0-litre turbo petrol, 6-speed manual, 5 doors, Slate Grey
Mileage 41,620 (odometer photo on file)
Previous keepers: 2. Keys: 1
Service book: stamped for 4 of 5 services; the 2023 stamp is missing
Equipment: reversing camera, heated seats, Bluetooth, parking sensors
Warranty: our 3-month warranty, terms sheet available
Price: $11,450
INSPECTION NOTES (our workshop, 11 October)
Brake pads 70%. Tyres 6 mm front, 4 mm rear. No warning lights. Air conditioning works.
Marks: front left alloy kerbed, about 6 cm. Small dent, rear right door, 2 cm, paint intact. Light scuffs, rear bumper corner. Driver's seat bolster worn. Scratch on tailgate trim.
HISTORY CHECK (run 10 October)
No recorded accident. No outstanding finance. Mileage records consistent. No theft marker.The AI's first draft looked like this.
2019 compact hatchback with a 1.0-litre turbo petrol engine and a six-speed manual gearbox, in Slate Grey. [SPEC]
41,620 miles, with the odometer photographed. [SPEC]
Two previous keepers. [SPEC]
Full service history. [SPEC]
Brakes and tyres were checked in our workshop on 11 October: pads at 70%, tyres 6 mm at the front and 4 mm at the rear. [INSPECTION 11 Oct]
New tyres all round. [INSPECTION 11 Oct]
Drives smoothly and quietly. [OPINION]
History check on 10 October: no recorded accident, no outstanding finance, mileage records consistent. [HISTORY 10 Oct]
Reversing camera, heated seats, Bluetooth and parking sensors. [SPEC]
Three-month dealer warranty; terms sheet available. [SPEC]
MARKS AND CONDITION
Kerb scuff on the front left alloy, about 6 cm. [INSPECTION 11 Oct]
Small dent in the rear right door, 2 cm, paint intact. [INSPECTION 11 Oct]
Light scuffs on the rear bumper corner. [INSPECTION 11 Oct]
Wear on the driver's seat bolster. [INSPECTION 11 Oct]
Scratch on the tailgate trim. [INSPECTION 11 Oct]
Price $11,450. [SPEC]Most of it is right, and three lines are wrong despite carrying tags. "Full service history" is tagged [SPEC], but the spec sheet says four of five services are stamped and the 2023 stamp is missing. "New tyres all round" is tagged [INSPECTION], but the inspection says 6 mm and 4 mm, which are used tyres, not new ones. "Drives smoothly and quietly" is tagged [OPINION] with no name or date, which the prompt forbids. The prompt ruled out the first and the third, and the model wrote them anyway: instructions reduce slips, they do not remove them. A tag is a claim, not proof, so the person publishing opens the cited document for every line. The fixes take two minutes: replace the first with "Service book stamped for 4 of 5 services; the 2023 stamp is missing", delete the second, and delete the third unless a test-drive record exists. Then strip the tags and publish.
The marks section is the part to protect. Left to itself, an assistant writing "sales copy" skips flaws, which is why the prompt makes them mandatory and why the listing is better for naming sizes. A buyer who reads "kerb scuff, about 6 cm" and then sees 6 cm has no reason to feel misled, and the price is easier to defend. Follow-up messages, and the promises to keep out of them, are covered in the tutorial on ChatGPT for car sales.
Words that need proof before they go in a listing
Some phrases arrive in AI copy because they are the genre's furniture. Each makes a claim a buyer can check, so each needs evidence or a safer rewrite.
| Phrase | What the buyer hears | Evidence needed | Safer wording |
|---|---|---|---|
| Full service history | Every service is documented | Stamps or invoices for each service | Service book stamped for 4 of 5 services |
| One owner | Never sold on | Keeper count on the history check | One previous keeper per the history check of [date] |
| No accidents | Nothing has ever happened to it | A history check only shows recorded incidents | No recorded accident on the history check of [date]; no repaired panels found at inspection |
| Immaculate, mint, as new | No marks at all | Any mark contradicts it | List the marks with sizes |
| Low mileage | Less than expected | A comparison you can defend | The figure and the age: 41,620 miles, first registered 2019 |
| Warranty included | Everything is covered | The terms sheet | Three-month dealer warranty; terms on request |
| Price drop | The car was dearer before | A real earlier price | Earlier price and date, or leave it out |
| Original paint | No bodywork | Paint-depth readings | Omit unless you have readings |
Notice how often the safer wording is more specific and sounds more trustworthy. Specifics are the cheapest honest persuasion you have, and a model that is told to copy them will write them. Here is the difference across a whole paragraph. The first version is the kind of copy an unconstrained assistant produces; the second describes the same hatchback from the documents.
BEFORE: Stunning, immaculate hatchback in superb condition with full service history and new tyres. Drives like new! Won't last long, so call today!
AFTER: 2019 compact hatchback, 1.0-litre turbo petrol, six-speed manual, 41,620 miles. Two previous keepers. Service book stamped for 4 of 5 services (2023 stamp missing). Tyres 6 mm front, 4 mm rear. History check on 10 October: no recorded accident. Marks we found are listed below, with a close-up photo of each. Three-month dealer warranty, terms on request. $11,450.The second version has no adjectives and tells the buyer far more. It also gives you nothing to retract.
A repaired car: how disclosure reads in practice
The hatchback was easy because its history was clean. The harder case is a car with a repaired incident, where an unconstrained assistant tends to skip the awkward line. Take an imaginary seven-year-old SUV priced at $14,900 whose history check shows one recorded rear impact four years ago. The repair invoice is on file: rear bumper replaced, tailgate repaired and repainted. Asked to "write an attractive listing", a typical assistant produces something like "Well looked-after SUV in superb condition, ready for its next family". Nothing in that sentence can be checked, and the repair has vanished.
Under the three-source rule and the mandatory marks section, the same assistant writes what the documents say. These are the lines that matter.
History check, 10 October: one recorded incident, a rear impact four years ago. [HISTORY 10 Oct]
Repair invoice on file: rear bumper replaced, tailgate repaired and repainted. Invoice available to view. [SPEC]
Inspection, 11 October: tailgate and rear bumper are repainted; panel gaps even; no corrosion found. [INSPECTION 11 Oct]
Unedited close-ups of both repaired panels are in the gallery. [SPEC]Buyers read a disclosed, documented repair very differently from a discovered one. The first becomes a point to negotiate; the second becomes a reason to walk away and to tell other people. The photo rules bite hardest on this car. The repainted tailgate gets an unedited close-up, and the glare-removal tool stays off, because reflections are how buyers judge paint.
What a photo edit may and may not change
Photo tools are where the temptation lives, because the same button that removes a bin from the background will remove a dent from the door. Decide the boundary once, in a table every editor can follow.
| Edit | Allowed? | Rule |
|---|---|---|
| Crop, straighten, resize | Yes | Never crop out a damaged area to hide it |
| Brightness, contrast, white balance | Yes | Paint colour must still match the car in daylight |
| Remove clutter: bins, other cars, people | Yes | Keep the car's shadows and ground contact |
| Replace the background | Yes, labelled | Caption says "background replaced"; edges, sills and wheels must stay exactly as shot |
| Remove reflections or glare | With care | Reflections can hide or reveal dents; compare with the original |
| Automatic "enhance" or "beautify" | Test first | These can smooth surfaces; check at 100% zoom |
| Remove dirt, stains, scratches, dents, chips, rust, kerb marks | Never | This is condition |
| Change paint colour or wheel design, add accessories | Never | It shows a car that does not exist |
| Smooth interior wear, stains or tears | Never | Same reason |
| Generated or stock images of the model | Only as clearly labelled stock | Never as the photos of this car |
Two practical rules sit on top of the table. Every mark named in the inspection notes gets its own unedited close-up in the gallery, and the first image is the least edited, not the most polished. A buyer who sees honest close-ups of the flaws is more likely to trust the glossy hero shot.
Whichever tool does the work (a phone editor, Canva, Photoshop or a dealer-focused photo service), the same questions apply, and they are worth asking any vendor before you trust it with a lot of stock. Can it ever alter panels, wheels or glass? Can automatic retouching be switched off? Does it keep the original alongside the edit? Does it embed a record of the edit? And what does it do to reflections? Some editors can attach Content Credentials, which the Content Authenticity Initiative describes as verifiable metadata, based on C2PA standards, that can record who made an image, when and with which tools. Using them is optional, not every site displays them, and a missing credential proves nothing, so treat them as a bonus on top of your own edit log. The tutorial on choosing an AI platform for used car dealerships covers the wider buying questions.
A four-minute check before any photo goes live
Open the original and the edited version side by side and zoom to 100% on the six places edits tend to hide trouble: wheel faces, lower door edges and sills, bumper corners, panel edges on the bonnet and roof, glass (chips and cracks) and the seat bolsters and steering wheel. If a mark in the original is not visible in the edit, the edit fails and the original is used. For the first week, have a second person do the same check. Then log it. A filled-in log for the hatchback looks like this:
| Photo | Edit made | Checked by | Result |
|---|---|---|---|
| 01 front three-quarter | Background replaced, brightness up a little | [initials] | Kerb scuff still visible at 100% |
| 07 front left wheel | None (damage close-up) | n/a | Unedited by rule |
| 09 rear right door | Straightened only | [initials] | Dent visible, matches the inspection note |
| 12 driver's seat | Exposure only | [initials] | Bolster wear visible |
The log takes a minute per car and pays for itself the first time a buyer disputes a photo, because you can show the original and say exactly what was changed.
What goes wrong when the tool is trusted blindly
Consider a composite example, not a real case. A dealer starts using a photo service with an automatic enhance setting switched on. On one hatchback the setting flattens the dent in the rear door, because it reads the shadow as an imperfection. Nobody compares the originals. A buyer drives two hours to see an $11,450 car, finds the dent in the first minute, leaves, and posts a review saying the photos do not match the car. The dealer has lost a sale and gained a review that every future buyer will read.
No one lied, and nobody intended to. The failure was a default setting and an absent check, which is exactly what the policy, the decision table and the log are designed to catch. The cost of the check is four minutes a car. The cost of skipping it is a sale and a public complaint.
Canned replies that stay honest
Listings generate the same six questions again and again. Draft each answer once, with AI doing the wording and the dealer approving the facts, then store them as templates. The rule for every reply is the same: state only what the documents show, say when it was checked, and offer verification rather than reassurance.
| Question | The reply must contain | The reply must never say |
|---|---|---|
| Is it still available? | A time-stamped yes or no, and how you will tell them if it sells | "It will go quickly" or invented interest from others |
| Any accidents? | What the history check shows and its date; what your inspection found | "No accidents" or "never been in one" |
| Can you do a better price? | The advertised price and what it already reflects | A figure you have not agreed with the manager |
| Will you take my car in part-exchange? | The details you need to give a fair estimate | A firm value before inspection |
| Can I see more photos? | Unedited photos of the panels they name | Edited versions presented as originals |
| Is the service history complete? | Exactly which services are stamped or invoiced | "Fully serviced" when anything is missing |
Here are two of them written out, using the hatchback. Placeholders stay in square brackets for the dealer to fill in.
REPLY TO "ANY ACCIDENTS?"
Hi [first name], thanks for asking. The history check we ran on [date] shows no recorded accident, and our workshop inspection on [date] found the marks listed in the advert: a 6 cm kerb scuff on the front left alloy and a 2 cm dent in the rear right door. A report cannot show everything about a car's past, so you are welcome to have it inspected independently, and I can send unedited photos of any panel you choose. Would you like me to do that?
REPLY TO "CAN YOU DO A BETTER PRICE?"
Hi [first name], thanks for your interest. The advertised price of $11,450 already reflects the marks listed in the advert. I would rather talk about price once you have seen the car, so you can judge it for yourself. When could you come in?Store them where staff already work: your CRM's template feature, or quick replies in the free WhatsApp Business app, where typing a slash calls up a saved reply. These replies are drafts for a person to send, not for a bot to send alone, and the buyer's personal details stay out of the prompt: the tutorial on what customer data never to paste into ChatGPT lists what to leave out. Review responses deserve the same honesty, as the review-reply samples for dealerships show.
Twelve minutes a car, plus a monthly listing audit
Illustrative numbers for a lot that lists 20 cars a month. Writing a listing and tidying a dozen photos by hand takes around 25 minutes. With the spec sheet pasted in, the tags checked against the documents (about five minutes) and the photo check (four minutes), the whole job takes around 12 minutes, so the saving is roughly 13 minutes a car, or about four hours a month. The assistant costs about $20 a month on a paid plan (ChatGPT Plus and Claude Pro both list at that price), and the photo tools range from free to subscription, so ask for a trial on your own stock before paying.
Once a month, audit five live listings. For each, take every sentence and find it in the documents; open the photo log and check one edited image against its original; and read the last three canned replies sent. Count the disputes: buyers who arrive and say the car is not as described. The count should be zero, and any non-zero month is a prompt for one rule change, not for blame. Re-run the audit after any tool update, because photo and text tools change their defaults without announcing it. When you price a car, the sibling tutorial on pricing each car against local listings uses the same sheet of facts, which is one more reason to keep it accurate.
Questions dealers ask about AI listings and photos
Is it acceptable to replace a car photo's background with AI?
Replacing the setting is common and reasonable if the vehicle itself is untouched, nothing about its condition changes and you say the background was replaced. Show every damaged area unedited, and keep the originals. The test is whether a buyer would feel misled on arriving to see the car.
Can AI write 'no accidents' in a listing?
Only if the evidence supports it, and the evidence is narrower than the phrase. A history check shows recorded incidents, not unreported ones. Safer wording states what the check shows and the date it was run, plus what your own inspection found. Invite the buyer to inspect the car, and say so in the advert.
Should I tell buyers a photo or description was AI-assisted?
Disclose edits that change what an image shows, such as a replaced background, in the caption. Text that you have checked and approved is your advert, and you are responsible for every claim in it. If a buyer asks how a photo was edited, answer plainly and share the original.
What if a buyer says a listing or photo was misleading?
Compare the advert, the photos and your unedited originals with the car before you reply. If the advert was wrong, say so, correct it at once and offer what your complaints or returns process provides. If it was right, show the buyer the dated documents and originals. Either way, log the case and change the rule or check that let it through, and take advice on your obligations if money has changed hands.
Do marketplaces allow AI-edited vehicle photos?
Rules differ by site and change often, so read each marketplace's photo policy before uploading edited images. Some ad platforms have rules for AI-generated images specifically: Google Merchant Center, for example, asks that they keep their IPTC TrainedAlgorithmicMedia marker. Never assume an edit is permitted because the tool offered it.
Further reads
- AI Chatbot for Car Dealerships: Rules to Set Before Launch — Rules to set before a chatbot talks to buyers for you.
- AI Automation for Car Dealerships: Answer a Web Lead in Minutes — Answer a web lead in minutes without over-promising.
- How to Use AI in Car Sales: Trade-In Enquiries Before the Appraisal — Handle trade-in enquiries before the appraisal.
- How Can AI Help Car Dealerships? A 30-Day Plan for a Small Lot — A 30-day plan for a small lot starting with AI.
- AI for Auto Dealerships: Where a Small Independent Lot Should Start — Where a small independent lot should start.
- ChatGPT Prompts for Car Dealers: 10 for Follow-Ups and Trade-Ins — Ten prompts for follow-ups and trade-ins.
- Brand Guidelines for AI Images: Colours, Style and Limits — Set limits for AI images across all your marketing.
- AI Marketing for Car Dealerships: A Four-Week Plan for One Lot — A four-week AI marketing plan for one small car lot: honest listings, social posts, past-buyer emails and compliant review requests.
- AI-Powered Marketing Tools for a Motorcycle Dealer: Spring Campaigns — A ten-week spring plan for a motorcycle dealer: which AI tools to use for used-bike posts, service reminders and demo days, with list prices.
- AI BDC for Car Dealerships: Does a Small Lot Need One? — A break-even sum on 140 leads a month, a twelve-enquiry test and three lot sizes show whether a small independent lot should buy an AI BDC.
- AI-Generated Marketing Photos: A Cafe Test of What Looks Honest — One brunch dish photographed three ways, scored by customers and checked against platform and advertising rules, to show where AI photos stay honest.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Content Authenticity Initiative, how Content Credentials work; Google Merchant Center policy on AI-generated image metadata; WhatsApp Help Center (quick replies); vendor plan and privacy pages for ChatGPT and Claude.