ChatGPT Ads Just Got Conversion Bidding: Here's the Test I'd Run

A week ago I wrote that UK brands can now buy ChatGPT ads and that it was worth a small, curious test.

Since then OpenAI has shipped eight changes to the Ads Manager in about ten days. Conversion bidding. Geo exclusions. Automatic advanced matching. A bulk API. App measurement. And a quiet but significant change to how daily budgets work that landed on Monday.

That's not a pilot tidying itself up. That's a media buying platform being built at pace. So here's what actually changed, and the test plan I'd hand a UK client tomorrow.

The headline: you can now bid for conversions

You can now pick Conversions as a campaign objective at setup and run conversion-optimised cost-per-click campaigns. There's a conversion-optimised bid cap in the ad group settings too.

The mechanic is worth understanding properly, because it's not quite Google's tCPA. Delivery is pushed towards the clicks most likely to produce your conversion event, but you're still billed on valid clicks. So it's optimisation towards conversions, priced as CPC. If you go in expecting to set a target CPA and walk away, you'll be disappointed. If you treat it as smarter CPC, you'll use it well.

This matters more than it sounds. Until now the honest answer to "what do I optimise for on ChatGPT Ads?" was traffic and a hope. Now there's a delivery signal pointed at the outcome, which is the difference between an experiment and a channel.

Daily budgets now work like Google's

From 27 July, daily budgets stopped being a fixed daily ceiling and became an average over seven days. Spend can run over on some days and under on others, and the platform paces to the seven-day total. Spend pacing was added at the same time, so budget distributes through the day rather than burning out by 10am.

No action needed, but do tell whoever checks the numbers. Someone is going to open the platform on a Wednesday, see 140% of the daily budget spent, and ring you in a panic. It's the same model Google moved to years ago and it's a good change. Just brief it in first.

The four that quietly matter more

Geo exclusions. You can now exclude locations. For a UK business that has been targeting the UK and quietly paying for clicks from parts of it that never convert, this is real money. Any client with a service area rather than a national footprint should use it in week one.

Automatic advanced matching. Hashed customer data used to match more website conversions back to ads. It's off by default. Go to Tools, then Conversions, then Data Source, and enable it for all web data sources. If you're running conversion campaigns without it, you're deliberately under-reporting your own results. As with any hashed matching, check it against your privacy notice and your lawful basis before switching it on. That's a five minute conversation, not a blocker.

Mobile measurement partners. AppsFlyer and Adjust integrations mean app installs and in-app events are attributable for the first time. If you have an app, ChatGPT Ads went from unmeasurable to measurable overnight.

The bulk API. Create and update campaigns, ad groups and ads asynchronously. It's the least glamorous item on the list and the clearest signal of intent. You don't build bulk tooling for a hundred pilot advertisers. You build it for agencies and for scale.

There's also a refreshed product card format for feed campaigns, now showing price and star ratings. If you're in ecommerce, that's your creative unit changing shape. Make sure your feed pricing and review data are clean, because they're now on the ad.

The test I'd actually run

Here's the plan, and the shape of it matters more than the budget.

Weeks one and two: get measurement right, run traffic. Pixel or Conversions API firing and verified. Automatic advanced matching enabled. OAI-AdsBot allowed through to your landing pages. Geo exclusions set. Run a CPC campaign against three or four tightly written context hints and gather baseline data. Don't touch the Conversions objective yet. It has nothing to learn from.

Weeks three to six: switch to Conversions. Once you have conversion events flowing, move to the Conversions objective with a bid cap you'd be happy paying. Keep the context hints stable. Change one thing at a time, because the auction is young and noisy and you will fool yourself otherwise.

Throughout: judge it on the right measure. Not CPC against Google. Not CTR against Meta. Judge it on cost per qualified enquiry or sale, against your blended target. And judge it over the seven-day budget window, not day by day, because that's how the platform now spends.

Budget: something in the low thousands a month for six weeks. Enough to learn, small enough that if it doesn't work you've bought a useful answer rather than a bad quarter.

The honest caveat

This is still a young platform. Ads only reach Free and Go users, inventory is deliberately light, and the third party performance data floating around is mostly vendors reporting on their own clients. I wouldn't move meaningful budget out of Google or Meta to fund this.

But the direction of travel over ten days is hard to ignore. Conversion bidding, better matching, bulk tooling and app measurement are the things you build when you expect professional buyers to turn up. They're turning up. The advertisers who learn this auction while it's cheap and unglamorous will have a year's head start on the ones who wait for a case study.

Same thing I said in June about being early, curious and measured. It just got a lot easier to be measured. 👊🏻

If you want a second opinion on whether a ChatGPT Ads test makes sense for your business, or you'd like us to run one, get in touch with the team >>

Clicky™