If your AI chatbot promises a price, who is on the hook?

Who should approve a price before it leaves the dealership?

Pricing language follows state and federal rules, the franchise brand and the store's own rules. If none of those clearly allows it, it does not go out in writing. Interest rates, payments and credit scores are treated as confidential and stay out of written messages.

That is the whole policy. Everything below is how to hold to it on a busy Saturday.

How should a rep respond when a buyer pushes on price?

Here is a sample thread from the Hold the Send demo. The buyer is fictional. She writes:

Honestly the price is too high. $2,500 off and I'm there today.

The draft the rep sees does not name a number. It says:

Dana, it's still here and I can have it pulled up front. I can't make the price call myself, so I've already asked Morgan, our sales manager, to look at your number before you drive in. Can you do 4:30 today?

The buyer sounds angry and is ready to buy today. The draft does not argue and does not agree. It flags the ask for a manager, holds the visit, and leaves the price for a person to decide. A named rep reads it, edits it if they want, and presses Send.

That reply makes no price promise.

What should a dealership price-review policy include?

1. Name who approves any message with a price, a payment, a trade value or a hold in it. One person per shift, written down.
2. List the words that stop a draft until that person has looked: "guarantee", "final price", "out the door", "lowest price", "no fees".
3. Keep rates, payments and credit scores out of written messages. Move that conversation to a call or the store.
4. Keep one place where every sent message is logged, so you can show a named person approved it.
5. Review last week's flagged messages every Monday. Adjust the list.

Why can an automated draft overpromise?

• It is built to be agreeable, and a buyer's "$2,500 off and I'm there" invites a yes.
• It has no authority to give a discount, and it does not know that.
• It cannot tell a joke or a test from a real deal, and the buyer may screenshot either one.

A person has the same pull to say yes, so the fix is the same: the number waits for the person who owns it.

What happened when a dealership's chatbot named a number?

In June 2026, CBC News reported that a chatbot named Quinn at BMW Toronto offered a customer CA$27,162.79 to buy back his 2021 X3. That figure was the balance left on his loan, not the store's valuation. A salesperson then told him the offer was not valid and the store would pay about CA$20,000.

After CBC contacted the dealership, it reinstated the original offer. Its sales manager said a person had passed the bot information in a way that made it read the loan balance as the buyback price. The store said human employees would present buyback offers from then on.

No court ruled on it. The number went out before a person read it, and the store paid about CA$7,000 to keep its word. Reported by Carscoops and The Drive, citing CBC News.

What does FTC guidance say about prices in text messages?

The FTC's Automobile Industry Pricing Transparency FAQs (September 15, 2026) say an advertised price must be the price any consumer can walk in and pay, leaving out only charges the government requires the consumer to pay. They list text messages among the media covered. Earlier in 2026 the FTC sent warning letters to 97 auto groups about advertised prices.

Whoever sends the message, it goes out in your store's name. Ask your counsel how that applies to messages a tool sends for you.

Where does Hold the Send fit?

Hold the Send, the AI Messenger copilot for car dealership reps, drafts the reply from the facts your rep pastes about the vehicle and the store's rules. Your rep reads it, edits it and presses Send. Nothing goes out on its own.

• Copy turns off on a monthly payment, APR, approval promise, invented discount, trade number, out-the-door price or doc fee, holding the unit, and freight, prep or setup fees.
• Price, payment, hold, trade and finance asks are flagged for a manager, and the rep can mark that a manager OK'd the reply.
• An activity log records drafted, flagged, copied and marked-sent, with no message text. It lives in the rep's browser.
• Once the rep sends, the message belongs to the store's own communication history. That is your system, not ours.

The same desk works on text, Instagram DMs and Facebook Messenger. We name the channels we have run, so ask about the one you use.

Can a human-written message make the same mistake?

• A rep texts a monthly payment number they were not supposed to send. The customer screenshots it before the manager sees it.
• A rep offers an incentive off the record that the boss would not want to see in writing.

Both are human mistakes, and both are the reason the review step sits in front of the Send button, not after it.

What does Hold the Send not promise?

• Hold the Send is not approved or certified by any automaker.
• A flag asks a person to look. It does not guarantee compliance with any policy or law.
• The demo does not send through live Meta today.
• Your store's own pricing, consent and texting rules still apply.

Full limits are on the compare page and the trust page.

See it on your own leads

Book 15 minutes with Nathan. You see which of last week's leads waited more than an hour, and the draft your rep could have sent on each. Your rep still presses Send.

Book 15 minutes or try the click demo.

Sources

FTC, Automobile Industry Pricing Transparency FAQs and the September 15, 2026 release, linked above. BMW Toronto account from Carscoops (June 11, 2026) and The Drive (June 20, 2026), both citing CBC News. Product behavior from the compare and trust pages.