What this means for the dominant narrative

This kills the dominant take on ChatGPT Ads. Most coverage frames the platform as an "influence layer," a place to shape consideration before the search happens, to intercept demand early. Expedia, the most sophisticated travel advertiser on the platform, is doing the exact opposite. They're capturing near-purchase intent and deliberately ignoring everything broad. The smartest money in the room is treating this as bottom-funnel, not top.

And here's what should reassure every travel marketer reading this: there's no new playbook. Expedia isn't doing conversational marketing or AI-native influence or anything you need a new framework to understand. They're running the same high-intent, long-tail paid search discipline they've run for fifteen years (destination plus property type, matched at the moment of intent) on a new surface. The lesson isn't "learn a new channel." It's "the channel you already know how to win just opened a new front door."

Four things this tells us about how ChatGPT Ads work for travel

1. Specificity wins the auction, and breadth gets nothing. The relevance-weighted auction appears to reward tight alignment between the context hint and the query. Generic context hints ("travel," "vacation," "hotels") either don't trigger or don't convert. Expedia's media team has clearly concluded it's not worth bidding on them. The architecture to copy is a matrix of destination × property type, not a single broad "travel" campaign.

2. Your context hints should be written as destination plus property-type combinations. Not keyword stems. Not categories. Context hints are plain language at the ad-group level and don't function as exact-match keywords, but the granularity Expedia is hitting ("cabins in Lake Tahoe," "beach resorts in Florida") tells you the resolution at which this platform actually matches. Write hints at that level.

3. The conversion path is collapsed by design. Once you've matched a specific destination plus property-type query, the parameterized LP (dates pre-filled) and the "Book now" CTA carry the user from conversation to booking in a single click. The LP matching isn't the clever part. It's the necessary downstream of the clever part. The cleverness is upstream, in the decision to only show up for queries that are already that specific.

4. Your DSA feed is probably most of what you need. If you already run dynamic search ads on Google with destination × property-type × amenity coverage, the structural work to run ChatGPT Ads in the same shape is mostly already done. Adapt the context hints to read like conversational phrasing, route them at the parameterized LPs you already have, and you're 80% of the way there. The brands that don't have the DSA infrastructure are the ones with real catch-up work, and almost none of Expedia's competitors are even at the starting line.

How the ad-matching mechanism actually works

When I first observed the three-tier pattern, I hypothesized that the ad was being matched against the AI's response, not the user's question. The reasoning seemed clean: the AI generates an interpreted intent from the query, and it would be elegant for the ad system to match against that interpreted intent.

I tested it. The data doesn't support it.

The cleanest disproof is the "kid-friendly summer destinations" query. The AI's response recommends San Diego, Lake Tahoe, and Maui by name, with descriptions and rationale for each. If the matching were happening against the response, an Expedia ad for any of those three destinations should plausibly fire. None does. The AI's recommendations don't drive the auction. The user's typed query does.

The simpler explanation is the correct one. Ads match against the query the user typed, not the response the AI generated. The match depth scales with the specificity of the query itself. Destination plus property type triggers a fully parameterized ad. Destination only triggers a destination-tailored ad. A query containing neither triggers nothing, regardless of what the AI says in response.

This reinforces the "no new playbook" thesis rather than complicating it. The matching layer is functionally similar to keyword targeting in Google Ads, not some novel response-content matching system. Context hints work like keyword themes against the user's typed query. The exact mechanism that's run paid search for two decades is what's running here.

The practical implication for how to write context hints is now simpler. Write them to match how real users phrase their commercial intent: destination plus property type plus optional amenity, in plain language. Not how the AI describes things in responses. Not how SEO content is structured. How users actually type queries when they're close to a booking decision.

How the platform actually works (June 2026)

Here's a tour of what's actually inside Ads Manager Beta today.

Where ads run: Ads serve to users in the US, Canada, Australia, and New Zealand. Self-serve Ads Manager Beta access is currently US-only for advertisers. UK, Brazil, Japan, South Korea, and Mexico are rolling out.

Who sees ads: Free tier and Go ($8/mo) tier users only. Plus, Pro, Business, Enterprise, and Education subscribers are ad-free. That matters when you model addressable reach. Your customer base may or may not skew toward the ad-eligible tiers.

What you can target: Geo at country, state, DMA, or ZIP level. Plain-language "context hints" at the ad-group level. That's the entire targeting stack. No demographics. No behavioral targeting. No interest segments. No lookalikes. No audience-building pixel for remarketing. Targeting is intentionally narrow and contextual.

How you can bid: Two campaign objectives: Clicks (max CPC) or Reach (CPM). No Max Conversions, no target CPA, no target ROAS. CPA bidding is in development. OpenAI's help docs frame $3 to $5 max CPC as a "recommended starting bid," but in practice $3 is the working floor. Bids below that effectively don't win impressions. Plan POC budgets accordingly: you're committing to roughly $3 per click minimum, and you need at least a few hundred clicks before you have signal. CPM default is $60 max, with observed clearing rates closer to $25 in some categories.

How you measure: Conversions API and pixel-based attribution went live April 30, 2026. You can measure conversions from ChatGPT Ads. You just can't tell the platform to bid toward them yet. Reporting includes impressions, clicks, spend, CTR, average CPC, average CPM, and conversions, aggregated, with no individual conversation data flowing back to the advertiser.

One platform behavior worth flagging: Ads only show on new conversations and don't persist. Starting a new session eliminates ads from previous conversations immediately. Even reopening a past conversation and posting a new comment in it doesn't re-trigger any ad. The pattern held across every old conversation I checked.

Caveats: this is from a free account, on browser (not the app), in a California-based setup. Different combinations (app vs. desktop, account tier, geo) may produce different behavior. Worth verifying for your own context before you bake it into a forecast.

The implications are real. You can't rely on a user returning to the same chat to re-encounter an offer. There's no multi-touch journey across a single conversation thread. Attribution windows matter more, because the ad impression is genuinely a one-shot event. And the screenshot-the-ad-and-share-it behavior that drives some viral discovery in other channels won't work the same way here, because users would need to recreate the conversation to re-trigger the ad.

Eligibility for travel: Travel and experiences is in the auto-eligible category tier. You can apply for an account, get approved, and run ads without case-by-case review. This is the easy door, and it's why Expedia is already inside it. Other auto-eligible categories include household and consumer goods, local services, digital products, and education.

How travel brands should think about testing today

The practical version, if you're deciding whether to test now:

Before anything else: does the math even work for your brand?

Floor CPC is $3. Industry-benchmark paid search CVR for travel sits around 2% for well-optimized accounts (better for some segments, worse for others, but 2% is a defensible starting point). That math gives you a rough cost per booking of around $150.

If your margin per booking is $30, you're at 20% ROAS. You're buying revenue at a loss. This channel does not work for you, full stop.

If your margin per booking is $200, you're at roughly 133% ROAS. The channel pays for itself with margin to spare. This is where the math starts to work.

The 2% CVR assumption is one of the things my POC will refine. Conversational referral could convert higher than Google Search traffic (the user is one decision from booking, the AI just recommended the category to them) or lower (research-mode users might not convert as cleanly). The benchmark gives you a conservative starting gate. Real numbers will come from real campaigns over the next 30 to 60 days.

This is, incidentally, part of why Expedia is alone in lodging on the platform. Their effective margin per booking, especially on bundles and packages, clears that math comfortably. Most thin-margin OTAs, low-AOV property categories, and metasearch arbitrage businesses can't justify $150 CAC. The unit economics gate the entire opportunity. The brands that can play here are the ones with high-margin booking products: premium experiences, luxury and full-service hotels, multi-night packages, tour operators with margin per traveler, and OTAs with strong attach economics on insurance, car rentals, and extras.

It's also why the one travel advertiser I found outside lodging, Business Travel Group LLC, is selling business class specifically. Economy flight margins are notoriously thin, often a single-digit percentage of fare and sometimes negative once distribution costs are factored in. They cannot clear $150 CAC. Metasearch and travel arbitrage businesses like Kayak, Skyscanner, and Trivago run on per-click spreads of $1 to $3 between what they pay for traffic and what they earn from outbound referrals. The $3 floor CPC breaks that model on impact. Business class margins are large enough to clear the math, and that's exactly the segment the one flight advertiser is selling into.

A side note on this. The economic gate cleanly explains the flight category, where airlines and metasearch are absent because the math forces them out. It does not cleanly explain the lodging category. Booking and Airbnb have take rates and AOV that should plausibly clear $150 CAC. Their absence from lodging is something else: channel skepticism, operational lag, strategic choice not to chase Expedia into a managed-partnership channel, or something not visible from outside. Whatever it is, Expedia has the field to themselves while the math decides who can even step on it elsewhere.

If the math doesn't work, none of the operational guidance below matters yet. Wait until either your AOV/margin profile changes or the platform's effective CPC comes down.

If the math works, test today if you have:

• LP infrastructure that supports long-tail intent: destination-specific, lodging-specific, amenity-specific URLs with parameter handling

• Conversion tracking already wired into your booking funnel

• Budget for a real POC: $3K to $5K minimum to generate meaningful click volume at floor CPC

• A defined audience segment that maps to specific conversational query types ("cabins in Tahoe," "beach resorts in Florida," "family-friendly resorts in Cancun")

Wait 30 to 60 days if you have:

• Only a homepage LP or generic category pages

• No conversion tracking on the booking funnel yet

• A test budget under $2K. The auction floor will eat it before you learn anything meaningful.

Things to test specifically:

1. Granular vs. broad context hints. Build one ad group with destination plus property-type hints ("cabins in Lake Tahoe," "beach resorts in Florida") and one with broad hints ("vacation rentals," "hotel deals"). My strong prior, based on what Expedia is doing: the granular group delivers and converts, the broad group barely serves. Prove it for your own destinations.

2. Whether hints written to mirror real user phrasing ("best cabins in Lake Tahoe for a family of four") outperform keyword-style hints ("Lake Tahoe cabin rental")

3. Whether parameterized LPs outperform category LPs on the traffic you do capture

4. Whether "Book now" CTAs outperform "Learn more" for hotel and tour operator brands

5. How conversational referral conversion rates compare to Google Search referral conversion rates, holding LP constant

What's not there yet (and what it means)

Honest list of what's missing today:

• No Max Conversions, tCPA, or tROAS bidding

• No audience layering (no demographics, behavior, interests, lookalikes)

• No remarketing

• No query-level reporting. You can't see which conversational queries triggered your impressions.

• Self-serve Ads Manager Beta still US-only for account creation

• Reporting is aggregated; no individual conversation context flows back

That gap will close. CPA bidding is in development. Reporting is getting deeper with each release. International self-serve is rolling out.

The practical implication for travel brands: the platform is mature enough today to run a real long-tail intent capture test with parameterized LPs. It is not mature enough today to be a primary acquisition channel, and anyone telling you it is hasn't been inside it.

What I'm testing first

I'm starting a first ChatGPT Ads campaign with one client this month and evaluating fit across the rest of my existing roster based on eligibility and the unit economics above. Most brands that pass eligibility won't pass the math, and that's fine. The point of running a test isn't to confirm the platform works in general. It's to confirm it works for your margin profile, at the CVR you actually deliver, on the LP infrastructure you actually have.

There's a useful outside signal worth noting. Neil Patel published an Adthena-sourced index of 580 observed ChatGPT ads in May 2026. The top five advertisers by volume: Lowe's (25 placements), Nordstrom (22), Best Buy (20), Liberty Mutual (12), and Insurify (10). Retail and insurance brands, every one of them with mature paid-search infrastructure and a long history of long-tail intent capture on Google. The top 10 advertisers account for 32% of all observed placements; another 23% sit in the top 11 to 25; 36% are one-time testers who showed up once and never returned.

Patel frames this as the market being "wide open," early opportunity to claim territory before the auction tightens. That framing is fine, but it understates the more interesting pattern. The story isn't how empty the market is. The story is who's winning. The early winners are exactly the brands you'd predict if this is high-intent paid search discipline ported to a new surface: established performance advertisers who already know how to run granular query-to-LP campaigns at scale. The 36% of one-time testers are most likely the brands that approached this as something new, didn't apply their existing paid search discipline to it, and got nothing back.

Worth flagging directly: Expedia isn't in Patel's top five. That's consistent with the architecture observation above, not in tension with it. Surgical destination plus property-type targeting produces fewer total ad observations than broad retail or insurance campaigns. Adthena's 580-ad index is a sample, not exhaustive, and travel-vertical queries are almost certainly under-represented in it. Different strategies, same underlying discipline.

The two questions I'm working to answer through the upcoming campaign:

6. How does context-hint granularity affect delivery? Does writing hints at the destination plus amenity combination level (the Expedia approach) significantly outperform broader category-level hints? My strong prior is yes.

7. How does conversational referral traffic convert relative to Google Search referral traffic, holding the LP constant? This is the question every travel CFO will ask once you propose a budget for this channel. It's also the assumption underneath the economic gate above. Real numbers replace the 2% benchmark in 30 to 60 days.

I'll publish what I find, including the things that don't work. That post lands in roughly 30 days.

The bottom line for travel

ChatGPT Ads aren't a future channel. They're a current channel with a specific shape: conversational query, ad creative match, parameterized LP. Expedia has already figured this out and is running the play. The discipline is one travel performance teams already have.

If you have the LP infrastructure to take advantage of long-tail intent matching, the time to learn this channel is now, before CPA bidding lands, before reporting deepens, before the auction tightens with new advertisers, and before the cost of being late shows up in your acquisition numbers. Not because everyone else is asleep, but because the platform is genuinely early and the people who learn its quirks now will run it better when it matures.

The question isn't whether to pay attention. It's whether you'd rather be early or late.