ChatGPT Ads are here, and they're worth testing .

The most interesting new demand channel in years opened to Shopify stores this month, and it takes ten minutes to connect. Here's how to test it so the result means something, and why it's half of how you show up inside ChatGPT.

Ads inside ChatGPT went from a US pilot in February to a self-serve Ads Manager with no minimum spend in May, and by the end of August they were running in around forty markets. On September 16 OpenAI named Shopify its first ecommerce partner and put a ChatGPT Ads app on the App Store: connect an ad account, sync your catalog, set up a pixel, build a campaign. Ten minutes, one install.

It’s the most interesting new demand channel in years, and the interesting part isn’t the novelty. It’s who’s on the other end of it.

The first two numbers are Adobe’s, from a study of more than a trillion visits to US retail sites. A year earlier, assistant-referred shoppers converted at roughly half the rate of everyone else. They now convert better than everyone else, which is what happens when a browsing tool turns into a buying one.

The ads themselves are built from your product feed, so the unit is a product rather than a banner, and the person seeing it is mid-conversation about what to buy. If you’ve spent two years watching AI traffic arrive with no way to influence it, this is the first lever you’ve been handed.

So yes, test it. Just go in knowing what you’re buying.

What you’re actually buying

A slice of ChatGPT, not all of it. Ads are shown to logged-in adults on the Free and Go tiers. Not Plus, not Pro. And it isn’t a free-versus-paying split, because Go is itself a paid tier at a few dollars a month. It’s the upper tiers, the ones people buy when they’re in ChatGPT all day for work, that no amount of budget will reach.

A category filter you don’t control. OpenAI keeps ads away from sensitive subjects, health and mental health and politics among them, and restricts categories like alcohol and gambling. Sell supplements and many of the conversations you’d most want to appear in are off limits before you bid.

A channel nobody can benchmark yet. There are no published performance figures, the case studies in circulation are vendor-written, and Perplexity quit ads altogether in February on the grounds that they cost user trust. This is a learning budget, not money moved off Google or Meta.

A click that lands on your own site. Since March OpenAI has been moving in-chat checkout into Apps, so the sale still happens in your checkout, on your terms. Good news commercially, and it means everything you already know about tracking still applies.

Where the one-click setup stops

OpenAI’s app is a genuinely good on-ramp. It connects or creates the ad account, syncs your Shopify catalog into Ads Manager, sets up a pixel and conversion events, and gets a campaign live in an afternoon. If you sell a few hundred tidy SKUs in one market, that may be all you ever need.

Catalog complexity is what changes the answer. Four things get harder as a store grows, and each one costs money quietly rather than loudly.

A catalog worth advertising. Whatever sits in your Shopify catalog is what goes out. On a small, well-kept store that’s fine. At a few thousand SKUs it means gift cards, samples, wholesale-only lines, discontinued products, near-duplicate listings and variants that haven’t sold since 2023, all arriving at a channel happy to charge you for clicks on them. Bigger catalogs carry per-channel work too: titles written for an ad rather than for a storefront, the right market and currency on each row, and a feed that refreshes before items expire after two weeks.

Both halves of the tracking. The browser pixel is the easy half, and the lossy one: ad blockers, consent banners and Safari’s seven-day cookie ceiling all eat into it. Server-side reporting is what fills the gap. The two need to share one ID per order, so a sale counts once instead of twice, and they need to agree with your catalog about which product sold. That last one is where most stores are quietly broken: we wrote about how that drift happens on Meta, and nothing changes here except the logo.

Enhanced match. Hashed customer details travelling with the events, so the platform can connect a sale back to the person who saw the ad. It’s the biggest single lever on how well any of this performs, and the first thing OpenAI’s grading looks at.

Somebody watching the signal. OpenAI grades each data source daily, as a score plus named warnings that say which part of your setup slipped, which only helps if someone reads them. For a preview of how you’d score, look at your Meta match quality today: same plumbing, years of history behind it.

One more, specific to this app. It reports conversions itself, and a second tool reporting the same sales only merges with it when both agree on the event name and the ID.

If you’ve run Meta beside Shopify’s native Facebook channel, or Pinterest beside its sales channel, you’ve seen this film. One reporter per event.

How to run the test

Keep it simple. One conversion goal, not three: a campaign optimizing toward something that happens twice a week learns nothing from it. A geo holdout, or clean on-and-off weeks, rather than trusting platform-reported ROAS on a channel with nothing to benchmark against.

Keep click-driven and view-driven conversions apart, because they’re credited over different windows and adding them together flatters the channel. And add ChatGPT to your “how did you hear about us” question, which on a channel this young will teach you more than the dashboard will.

The other half of showing up in ChatGPT

Ads are one way in. The other is being the product ChatGPT recommends when nobody paid for the slot, and that one reaches every tier, including the Plus and Pro subscribers the ads can’t.

It runs on the same catalog. ChatGPT’s own shopping answers draw on merchant product data rather than on ad spend, so the feed that would power your ads is also what makes you eligible to be recommended for free. Assistants read your storefront directly too, which is where the discovery files their crawlers look for, llms.txt and agents.md, complete product structured data with real offer details, machine-readable review ratings, and not quietly serving crawlers a worse page than humans get all start to matter. Google’s AI answers lean on the same Shopping data you already send to Merchant Center, so a tidy catalog pays there as well.

We covered what assistants actually consume in product feeds are the new SEO. If you want the state of your own storefront, our AI readiness audit reads it the way an assistant does and scores it out of 100.

Where Máximo fits

Both halves come off one catalog, which is why Máximo treats them as one product rather than two. ChatGPT Ads is a destination alongside Google, Meta, TikTok and Pinterest: the same catalog that goes everywhere else, carrying the same product IDs your sales reports use, with browser and server reporting matched so an order counts once, the ad click attached from the landing page through to the order, customer matching included, and the products Catalog Intelligence keeps out of your other feeds staying out of this one too. AI Discovery is the other half, built from the same catalog.

Planning a test? Talk to us. We’ll go through your catalog and your conversion tracking, tell you what’s worth fixing before you spend anything, and set up both halves, because the shoppers you can’t buy are still worth winning.