I-15, Nevada → California, 21 August 2026
AI chat ads (ChatGPT) · Travel, legal, auto · Airbnb, Gina Corena & Associates, Findlay Kia
The ad offered “Pay $0 upfront on select bookings,” on the drive from Las Vegas into California.
I-15 near Primm, Nevada, minutes before the state line, 9:18 am, Friday 21 August 2026. Photo: HaELM.
“The preserve protects large stretches of the surrounding desert from that kind of development,” ChatGPT answered when I asked about the solar towers we were driving past. Right under that answer was an ad, an Airbnb card showing an A-frame cabin in pine trees and the words “Joshua Tree National… Pay $0 upfront on select bookings.”
It was 9:46 am on Friday 21 August, the first day of our road trip. We were just past the state line between Nevada and California, surrounded by the Mojave Desert.
This road trip was very special for me, as it was our first trip to the West Coast together, my husband’s and mine, and we chose to drive from Las Vegas to San Francisco.
My plan was to keep chatting with ChatGPT about the local places we drove past. But this Airbnb ad for Joshua Tree bookings sent me looking for a Joshua tree. I had never seen one.
It wasn’t the only ad in that chat. In Nevada, the same chat had shown Las Vegas ads (Gina Corena & Associates, an injury law firm, and Findlay Kia, a local car dealer). After the state line, it showed a California stay. The ads changed as my place and my question changed. What decides which ad appears under an answer?
Fig. 1 · The Airbnb ad under ChatGPT’s answer about the solar towers near the Mojave National Preserve, seen on I-15 just inside California, 9:46 am, Friday 21 August 2026. Screenshot: HaELM.
10″ Decode
Mechanism: A message may feel pertinent when it fits what’s on someone’s mind.
A traveller might think: “Nothing to pay today. If it gets late, we’ll stay the night.”
Brand move: The brand paired a local cue with a reason to act.
Trade-off: Someone in a hurry can skip an ad under an answer.
Finding: Context can make an ad fit; the message must give people a reason to act.
The Brief
HaELM’s reconstruction of the brief, from the ads and public sources. Not the brands’ own.
Brand / Property: Airbnb (lead); Gina Corena & Associates; Findlay Kia
Category: Travel › Short stays; Legal › Personal injury; Auto › Car dealer
Market: US, I-15 from Nevada into California
Objective: Not public (inferred: bookings)
Mechanism: Contextual relevance (HaELM’s reading)
Spend model: Auction; the advertiser sets the budget and bid (OpenAI). Figures not public
Competitive set: None seen in the chat
Target audience: Logged-in adults on ChatGPT’s Free and Go plans (OpenAI); travellers asking about the desert (inferred)
Placement: A card labelled “Ad” under a ChatGPT answer
Spotted: Friday 21 August 2026, 9:29–9:46 am PDT, I-15
CTA: Airbnb: “Pay $0 upfront on select bookings”
Stakeholders: OpenAI (platform); the brands’ own teams or agencies (inferred)
Before the ads
OpenAI starts selling ads in ChatGPT
In February 2026, OpenAI began testing ads in ChatGPT in the US, on the Free and Go plans. The ads are displayed under the AI answers. They seemed to follow where I was and, later, what I asked. Each one was labelled “Ad”.
Airbnb, a marketplace for short stays and travel experiences, has to reach people who are planning a holiday and whose travel itinerary isn’t final, yet it’s hard to reach that exact person unless they tell ChatGPT that they’re looking to book a stay at Joshua Tree. In my case, I was driving through the Mojave Desert to my hotel in California, so the ad was likely aimed broadly at someone travelling through the area.
The drive
Three ads in seventeen minutes on I-15
At around 8:26 am PDT, I was asking ChatGPT to give me some background about the Seven Magic Mountains artwork as we drove past it and to share information about the places we would pass on our drive to LA. “Where Car Crashes Happen, see the latest Las Vegas crash data,” read the Gina Corena & Associates ad I saw at 9:29 am PDT. In the same stretch of chat at 9:44 am PDT, “Proudly Serving Las Vegas, We’re Here to Help You, Find Your Perfect Car,” read the Findlay Kia ad. Both ads named Las Vegas, and Gina Corena’s “Where Car Crashes Happen” seemed to speak to anyone driving. Neither had anything to do with the questions I had asked ChatGPT.
Fig. 2 · The Ivanpah solar towers, seen from I-15 near Primm, 9:19 am, Friday 21 August 2026. Photo: HaELM.
Then, a little ahead, I asked about the solar towers around the Mojave Desert near Primm. “Joshua Tree National… Pay $0 upfront on select bookings,” read the Airbnb card that appeared two minutes after the Findlay Kia ad. We were near the Mojave National Preserve, known for its Joshua tree forest at Cima Dome. Its $0 upfront offer may give a traveller who hadn’t planned a stay a reason to think about one now.
Fig. 3 · The three ads in the same ChatGPT chat, seen on I-15 between 9:29 and 9:46 am, Friday 21 August 2026. Screenshots: HaELM.
That Airbnb card made me curious about the trees themselves, so I started looking for one to photograph. I kept sending pictures of trees. ChatGPT said they were desert wash trees, palm trees, and low shrubs, not Joshua trees. “Keep looking, but you haven’t photographed one yet!” it said. At Barstow, it said, “I wouldn’t turn around for it.”
My itinerary was already planned, so I did not click the Airbnb ad to book a stay.
So why did ChatGPT show me a Joshua Tree stay?
The media
How ChatGPT picks the ad under an answer
The ad is a native card under ChatGPT’s answer, marked “Ad” and styled like the chat, with a square image, the advertiser’s name and logo, a short headline and one line of text. Ads show to logged-in adults.
On 21 October 2026, about eight months after the US test began, OpenAI said ChatGPT ads had reached 89 countries.
OpenAI says the ads are chosen based on “the context and intent of your current conversation” and “basic context such as general location or language”. If ad personalisation is switched on, ChatGPT can also use memories and recent chats, but I don’t know whether mine was on. The ads are sold in an auction where fit with the chat counts as well as price, which OpenAI calls “a relevance-weighted, second-price auction”. Each advertiser sets the most it will pay, and an ad that fits the chat can beat a bigger bid. The winner then pays less than its maximum.
Gina Corena & Associates likely wants to reach people driving a busy stretch of Nevada highway, so that if they have an accident on the road, its name may be the one they remember. The local Kia dealer’s ad seemed to target people on the highway and at roadside attractions, to make its name known. The Airbnb ad didn’t show me a stay in Los Angeles, where I was heading. ChatGPT chose an ad for a stay in the same desert region, though not on our route. I had told ChatGPT I was driving from Las Vegas to Los Angeles and was asking about local attractions, so it seemed to treat us as tourists who might take a day trip into the desert.
The brand lens
Where Airbnb’s $0 upfront offer came from
“Pay $0 upfront on select bookings”
It may move the pain of paying to later, so booking feels easier for someone whose plans aren’t fixed.
In August 2025, Airbnb launched “Reserve Now, Pay Later” in the US, which lets US guests book eligible stays in the US for $0 upfront and pay only near the last day they can cancel for free. In a survey Airbnb commissioned, 60% of US adults said flexible payment options matter when they book a holiday (Focaldata, July 2025).
Airbnb likely gets to reach last-minute, on-the-road travellers who are driving past or towards places where its stays are listed.
The trade-off
A driver in a hurry can skip it
A driver in a hurry who only wants an answer from ChatGPT might skip the ad under it. Despite that, buying this placement may make sense for Airbnb, because it gets to reach a traveller mid-route. A billboard on I-15 reaches anyone driving past, but can’t know who they are or what they are thinking about. The chat ad could use roughly where I was and what I had just asked about, the solar towers near the Mojave National Preserve. It couldn’t know I had already booked.
Test
Would the bookings have happened anyway?
1. The question
Did seeing the Airbnb ad lead to more confirmed bookings than there would have been without it?
2. The method
Take the people ChatGPT would show the ad to and pick some at random not to see it. This is the holdout group. Over the same period, compare the share of each group that booked.
3. How to read it
If the people who saw the ad (the exposed group) booked more often, by more than chance would explain, the gap is the estimate of bookings the ad added.
4. What it can’t tell us
The test can’t tell which part did the work. ChatGPT showed the card based on roughly where people were and what they had asked, and the card carried the “Pay $0 upfront” offer. The holdout group used ChatGPT as usual but never saw the card, so any extra bookings can only be credited to the card as a whole.
Fig. 5 · How a holdout test would be read. Illustrative numbers only; Airbnb and OpenAI have not published results for this ad.
Fig. 4 · Joshua trees behind a chain-link fence near Hesperia, California, Friday 21 August 2026. Photo: HaELM.
Closing thought
Near Hesperia, behind a chain-link fence
Near Hesperia, just before Cajon Pass, I saw the Joshua trees. They stood behind a chain-link fence, beside a faded sign covered in purple graffiti.
“This is the one you were looking for,” ChatGPT replied.
