AI Car-Buying Agents Favor Carvana? Here’s How Dealers Win the Shortlist

Dealer AI Websites · October 1, 2026 · 6 min read

On September 30, TheStreet reported on a Morgan Stanley note about AI car-buying agents like Meta's Muse. The bank's conclusion: Carvana screens as the best-positioned auto retailer if agents take over vehicle shopping, and franchise dealers face the most pressure.

If you run a dealership, that headline deserves a serious answer, not a shrug. Morgan Stanley gets a lot right. But the note measures who owns trucks, reconditioning centers and national logistics. It underweights the thing an AI agent actually needs before any of that matters: dealer facts it can read, trust and act on. That gap is fixable, and it's fixable this quarter.

What Morgan Stanley gets right

Agents compress the worst part of car buying. Today a shopper bounces between listings, loan offers and trade-in quotes before negotiating anything. An agent can search across retailer websites and marketplaces, standardize the offers and, eventually, negotiate. Cars are expensive, bought rarely and hard to compare, which is exactly why agents could hit auto retail harder than most categories.

Brand recognition matters less. If the agent does the first search and the comparison, a store can't win just because the shopper knows its name, saw its ad first or started on its website. Price, available inventory and the ability to complete the transaction become the deciding factors.

Unit economics give room to compete on price. The note estimates franchise dealers' gross profit per retail unit will run about $1,106 below Carvana's in 2026, and CarMax about $740 below. Carvana sold 197,325 retail units in Q2, up 38%, with a 10.4% adjusted EBITDA margin. That's real scale, and it's still growing.

Agents still can't touch the car. An AI agent can compare vehicles. It can't inspect, recondition, transport or deliver one. Physical fulfillment still decides who completes the sale.

What the note underweights

Dealers already own the physical layer, locally

Morgan Stanley credits Carvana for inventory, inspection, reconditioning and delivery. A franchise store has every one of those: vehicles on the lot today, a service department that reconditions every unit, factory-trained technicians and a buyer who can drive off this afternoon. The difference isn't that dealers lack the physical pieces. It's that an agent can't see them.

When an agent asks "which nearby store has a certified three-row SUV under $42,000, available today?", a dealer with the right truck and an unreadable website loses to a national retailer with a cleaner answer. That isn't a logistics gap. It's a data gap.

The handoff model favors whoever's data is cleanest

The note highlights CarMax's app inside ChatGPT, which lets shoppers search more than 45,000 vehicles with plain-English prompts. Morgan Stanley's own read is that it mainly helps customers discover vehicles, then sends them back to CarMax to finish the deal. The bank's broader Muse analysis describes the same emerging model: an external agent helps with discovery, then hands the transaction back to the retailer, which keeps the customer data, checkout and fulfillment.

That model doesn't require a national footprint. It requires a dealer-controlled way for agents to search inventory, read a specific vehicle, check availability and get a straight price, then hand the shopper to a human. Any dealer can have that. You don't need 45,000 cars to be the best answer within 30 miles.

The FTC just made dealer prices comparable

In September, FTC staff guidance said a dealer's advertised price must be the price any consumer can actually pay, apart from government-required charges, with dealer-required fees included. Agents comparing "all-in" offers is exactly what that guidance enables.

For a dealer, that's a threat only if your price data is messy. A store that publishes clean, itemized, machine-readable pricing becomes easier for an agent to recommend, not harder. Separate the advertised price, required charges and conditional rebates, mark anything unknown as unknown, and you look trustworthy next to a rival whose page an agent can't parse.

Margin isn't the only lever an agent weighs

A $1,106 gross-profit gap is a reason not to fight a national retailer on price alone. It isn't a reason to lose. An agent working for a real person also weighs distance, how soon the car can be in the driveway, whether a test drive is possible this weekend, whether the trade can be appraised in person and who will service the vehicle for the next six years. Those are dealer advantages. They just have to be stated in a form software can use.

The real gap: legibility, not logistics

Carvana's lead isn't that it has cars and dealers don't. It's that its data is built for software and most dealer data is built for a 2014 web page. That's the part a dealer can close.

What to do this quarter

1. Ask the agents. Put your most common shopper questions to ChatGPT, Gemini, Claude and Meta AI, the way a buyer in your market would ask them. Write down who shows up: you, a marketplace or Carvana.
2. Fix your price data before an agent does it for you. Make every advertised price an all-in price under the FTC guidance, with dealer charges included and conditional offers labeled. Have counsel review it.
3. Make inventory machine-readable. Publish complete structured data on every vehicle page, keep it synced with your feed, and expose a direct inventory interface agents can call instead of scraping your pages.
4. Publish your physical advantages. Pickup today, delivery radius, test-drive availability, in-person trade appraisal, service and warranty: make them explicit facts, not marketing copy.
5. Keep the handoff human. Let agents discover and compare. Then route the shopper, with their vehicle, questions and consent, into your CRM, so your team picks up where the agent left off.

How Dealer AI Websites closes the gap

Every Dealer AI Website ships with the layer this note assumes dealers don't have. The Dealer Agent Gateway gives AI agents typed tools to search your inventory, read a vehicle, verify availability and get an itemized price disclosure, without touching your DMS or CRM. It follows the open Dealer Agent Protocol, so any agent that speaks it gets the same trustworthy answers from your store. The DealershipMCP visibility check shows whether AI recommends you today.

On the human side, shoppers get fast pages, plain-English search and an AI assistant grounded in each car's real data, and every lead lands in your CRM as standard ADF. Pre-Owned is $499 a month, Franchise is $999, the first month is free and there's no contract.

We covered the agents themselves in Meta Muse Is Here and Agentic Automotive Retail: Why OpenAI Dots Changes the Fight Against Carvana.

The bottom line

Morgan Stanley is right that AI agents will reward price, inventory and the ability to close. It's wrong to treat franchise dealers as passengers in that shift. Dealers already have the cars, the shops, the service relationship and the local trust. What most don't have yet is a way for an agent to see it.

The question isn't whether you can out-Carvana Carvana. It's whether, when a buyer's agent builds the shortlist, your store is readable enough to be on it.

Sources

• TheStreet: Carvana, CarMax jockey as agentic AI tools look to fix the worst parts of car buying
• FTC: Automobile industry pricing transparency FAQs
• Meta: Introducing Muse, the personal AI agent