An afternoon's check of AI answers against your dealership's lot

What to have open before the first question

The check compares what four assistants say with what is true at your dealership on one afternoon, so write down what is true first. Export the stock list from your dealer management system (DMS) that morning, or save your site's vehicle search results, and note the time. Cars sell during the afternoon too, and the time on the export is what separates a wrong answer from one that is only a few hours late.

• The stock export: stock number, VIN, year, model, trim, mileage, advertised price and status (on the lot, on hold, in transit, sold).
• The cars sold in the last two weeks, with the day each one went.
• Your own vehicle, financing, trade-in and service pages, so every term in an answer is compared with the page rather than with memory.
• Each department's number and hours, exactly as the site states them.
• For a dealer group, the addresses, numbers and hours of the sister stores.
• An empty document for the answers, with the export time at the top.

Whoever runs it has to know the lot well enough to notice a car that is not there: the owner, the general manager or the manager of the BDC (the business development center, the team that answers internet leads and phone inquiries).

Eight questions across sales, finance and service

Fill the brackets from the export and your own details, then ask each question exactly as written. They are numbered because the section on wrong answers refers back to them.

1. "[year] [model] [trim] for sale near [place]", using a car that is on your lot today.
2. "used [model] for sale near [place]", using a model whose last unit you sold in the past week or two.
3. "certified pre-owned [make] dealer in [place]".
4. "does [dealer] arrange financing for buyers with [credit situation]".
5. "does [dealer] take trade-ins that still have a loan on them".
6. "what is the phone number for [dealer] service".
7. "what time does [dealer] service open on saturday".
8. "what do people say about buying a car at [dealer]".

Three of the eight name only a place, and five name you. The balance leans toward your name on purpose, because a question that names you is where a wrong fact about you shows itself. For questions 1 to 3, also note whether you were named at all and whether the answer recommended you or only listed you: that is this afternoon's answer to the question in the title, for one day. Whether you are named when a buyer does not know you yet, month after month, is the monthly panel's question.

Thirty-two answers in one sitting

Ask each question once in each of ChatGPT, Claude, Gemini and Perplexity, for thirty-two answers. Use a new chat for every question, with web search on where it is offered and memory off, or signed out, where that is allowed. Paste each answer whole into the document with the assistant and the time, and copy every link it cites; the cited page is the first place to look for anything it got wrong.

Run all thirty-two on the same afternoon as the export. Spread over a week, they would be checked against a lot that had changed under them.

Holding each answer against the lot and the site

Read the answers one at a time for facts, which is fact-checking with the stock list as the reference. Identify a vehicle by its stock number or VIN if the answer gives one, and otherwise by year, model, trim and mileage together, then write down its status in the export. Compare a price with the vehicle page, not with memory; a number or an hour with the department's own line on your site; a financing or trade-in statement with the sentence on its page.

Then write down where each fact came from. The link behind it, the source citation, is your own page, the manufacturer's site or dealer locator, a listing site, your group's site, or nothing. The source matters even when the fact is right: a correct price quoted from a listing site sends the buyer there, to your car beside other dealers' cars.

Five wrong answers, and the page each one comes from

Each kind of wrong answer points at a different page. When AI gets facts about your business wrong covers the two ways a wrong fact arises: a published source that still states it, or a gap the model filled, a hallucination. The five below are a dealer's versions of the first.

A vehicle you sold last week

Question 2 is built to find this one. Open the page the answer cites. If it is your own vehicle page, the page outlived the sale: it still offers the car in its text or its markup, or it now shows a different car at the same address. If it is a listing site's page, your inventory feed did not send the sale. Change your own page first, to a plain sold notice with similar cars linked, as the used-car page as a record describes, then check what the feed sends on the day a car sells.

A price from an old listing

Questions 1 and 2 are where prices turn up. Search your site for the exact figure the answer gave, then the listing sites that carry your cars. An old price survives on a copy nobody edits: the same car under an earlier stock number, an expired promotion page, a listing that missed the last change. Keep one current price per vehicle, identical on the page, in its markup and in the feed, with the date it was set. How a price may be advertised is for your compliance officer or counsel, since the rules differ by jurisdiction.

The wrong department's number

If the answer to question 6 is the showroom's number, look at how your site states the service number. It needs its own labeled line on the service page and the contact page; a footer with a main number and a tracking number per campaign gives a reader several numbers and no way to tell which one rings the service drive. In our audit of 9,483 small business websites (40 trades, car dealerships among them, late August to early September 2026), 36.9% published more than one phone number across their own three main pages; we publish no dealership cut. The article on dealer service pages covers the department's own number and hours.

The manufacturer's locator or a listing site cited instead of your page

When the facts are right but the cited page is the brand's dealer locator or a listing site, the answer found you on someone else's page, so check that yours can be reached. The crawler access check reads your robots.txt the way each AI crawler token does, and OpenAI's crawler documentation says a site that disallows OAI-SearchBot is not shown in ChatGPT search answers, though it can still appear as a navigational link. Then open a vehicle page with scripts turned off and look for the car in the HTML.

Bot protection needs its own test, because robots.txt does not reveal it: request a vehicle page with a crawler's user agent and look at what comes back. In the audit, 7.2% of the sites served a challenge page instead of the site, and 3.0% rendered their content only in JavaScript. A wrong fact on the locator itself is corrected through the manufacturer's channel for its dealers, not on your own site.

A sister store's hours given as yours

If question 7 comes back with a sister store's hours, find where the two stores share text: a group page listing every rooftop above one set of hours, a shared contact page, or markup that describes the whole group as one business. Each rooftop needs its own page with its own address, number and hours, and its own Business Profile; Google's guidelines ask for one profile per location, and for a dealership's profile to carry its sales hours. In the markup each rooftop is its own node, as building an entity explains.

What one afternoon tells you, and what it does not

It tells you what four assistants said to eight questions on one day, once each. That is enough to find wrong facts worth fixing tomorrow morning. It is not enough to say whether you are recommended more or less than last month, because the same question can name different dealers on its next run. One afternoon's answers are a snapshot, not a trend.

Change is read with a monthly panel: fixed lines, three runs on each assistant, and the stock lines read beside that day's inventory, as AI visibility for car dealerships lays out. Until then, fix what you found in order of what each costs a buyer: the sold car and the wrong price first, since either can mean a wasted trip, then the wrong hours and the wrong number. The overview for car dealerships lists what each page should state. Next month, ask the same eight questions and compare the two afternoons answer by answer.