What happened
On July 24, OpenAI shipped a batch of ChatGPT Ads upgrades, and the headline is conversion-optimized bidding. Pick the new Conversions objective and campaigns automatically optimize toward clicks that are more likely to convert, fed by conversion data you connect under Tools > Conversions > Data Source. Advanced matching uses hashed customer data to tighten website-conversion attribution, and AppsFlyer and Adjust integrations cover app installs and in-app events.
Budgeting changed too. An average daily budget model paces spend automatically across a rolling seven-day window, distributing it through the day rather than burning it in bursts — that piece starts rolling out the week after the announcement. Geographic exclusions arrived alongside it, so you can finally carve out regions you don't ship to.
For teams managing at scale, there's an asynchronous Bulk API for creating and updating campaigns, ad groups, and ads. And product-feed ad cards are being refreshed to show pricing and star ratings — a clear signal that shopping-style ad units are where OpenAI wants this to go.
Why this matters
Run down the checklist: conversion bidding, budget pacing, geo controls, app measurement, match-rate tooling, a bulk API. That's most of what separates an experiment from an ad network. Before this release, ChatGPT Ads was awareness spend with blunt controls. With oCPC and advanced matching, it becomes a direct-response channel you can actually measure — on paper, anyway.
Keep some skepticism handy. Conversion bidding needs conversion volume to learn, and nobody outside OpenAI knows what query mix or inventory depth the system has yet. There's no third-party measurement track record either. But the product cards showing price and star ratings tell you commerce intent is the target, and platforms usually reward advertisers who arrive before the auction gets crowded.
The buyers most likely to research in ChatGPT are the considered-purchase and B2B kind — people comparing specs, asking which option fits their situation. If that describes your customers, the cost of a small pilot is lower than the cost of learning this platform after your competitors have set the CPA benchmarks.
What to do about it
Run the qualification test first
Before spending a dollar, ask ChatGPT the five questions your buyers actually ask about your product category. If your products or competitors show up in the answers, your buyers are researching there and a pilot makes sense. If your category never surfaces, skip this round and revisit in a quarter.
Wire conversions before you launch
Set up your data source under Tools > Conversions > Data Source, enable advanced matching, and verify events fire on a test purchase. An oCPC campaign with no conversion signal is just expensive traffic — the bidding system learns from what you feed it.
Cap the pilot and judge it weekly
Set a modest average daily budget and remember the rolling seven-day pacing means daily spend will wobble by design. Don't evaluate until you have enough conversions to compare CPA against your Google and Meta baselines — a nervous day-three shutdown teaches you nothing.
Clean the product feed now
Pricing and star ratings render on the refreshed product-feed cards, so stale prices and missing review data are now visible in the ad itself. Verify feed pricing matches the site and that review counts are flowing before those cards start serving.