Conversion Optimization is the bidding mode in which ChatGPT Ads buys conversations for you based on the conversion event you configure, instead of the click. It became generally available in September 2026 for standard click campaigns as well as product feed campaigns, and OpenAI's first benchmark for it is a median 44% lower cost per result than click bidding, measured within the same advertiser across 27 advertisers. It needs working conversion tracking and at least five click-attributed conversions on the event before it can do anything useful, and the right way to adopt it is a two-week side-by-side against your existing click campaign.
What it is, and what it optimises toward
Click bidding asks the platform to find people likely to tap the card. Conversion Optimization asks it to find conversations that lead to the event you care about, using the conversions your pixel and Conversions API report back. OpenAI lists the eligible events as leads, orders, registrations, trials, subscriptions and content views, which covers an ecommerce order as well as a demo request or a newsletter sign-up. In August 2026 the mode was a beta limited to product feed campaigns; since September it is in general availability and services advertisers can use it.
Mechanically the bidder is choosing where your ad appears. On this platform an ad sits under an answer to a question, so choosing conversations is choosing contexts, which is the same lever our context hints pull by hand. The difference is that the bidder works from your own conversion data and adjusts continuously.
What you need before you switch
OpenAI's eligibility filter for its own benchmark is the most useful checklist available, because it describes the minimum signal the bidder had in every published result:
- +Conversion tracking that works: the pixel, and the Conversions API alongside it. In OpenAI's data the pair measured 57% more attributed conversions than the pixel alone, and the bidder can only buy what it can see.
- +At least five click-attributed conversions on the event you want to optimise toward, at least $100 of spend and three active days on the existing strategy.
- +One event, chosen deliberately. Optimising toward a primary conversion that fires twice a month starves the bidder; optimising toward a leading signal that fires daily gives it something to learn from.
The third point is the one that catches services businesses. If nobody has booked a call yet, the booking is a reporting event and the step before it, the demo click or the form start, is the bidding event. Move the optimisation up the funnel once the primary event has the volume. The events themselves, and how to dedupe them across the pixel and the API, are in the conversion-tracking guide.
The published numbers
The headline is a median 44% lower cost per result than click bidding, within the same advertiser, across 27 advertisers who ran both strategies on overlapping events between 10 and 23 July 2026. Six case studies accompany it, each a 14 or 28 day comparison, each labelled observational by OpenAI.
| Advertiser type | Click-through rate | Cost per result | Spend on the new mode |
|---|---|---|---|
| Software and technology | +27% | Cost per order down 50% | 98% after adoption |
| Software and technology | +22% | Cost per order down 87% | 73% after adoption |
| Jobs and recruiting | +180% | Cost per sign-up down 95% | 94% after adoption |
| Retail and ecommerce | +7% | Cost per order down 65% | 40% of eligible spend, side by side |
| Travel and hospitality | +72% | Cost per booking down 80%, bookings up 123% | Not stated |
| Financial services | +4% | Cost per lead down 50% | 70% after adoption |
Four of the six used the Conversions API, one used the pixel alone, one used both. The travel footnote notes that the gain coincided with a change in bid mix and the conversion signal ramping up, which is an honest description of what a switch like this does to a young account.
The named pilot advertisers in the same pack describe the outcome in their own words. HubSpot reports competitive CPAs across a number of campaigns and downstream revenue signals. Figma says ChatGPT Ads has become a top performer for its Figma Make campaigns, with higher click-through rates than traditional search. Best Buy saw click-through rates higher than it anticipated and demand across a broader mix of categories. VistaPrint found the majority of its ChatGPT traffic were new visitors. Canva calls it a strong, high-intent channel and says performance is improving as the platform matures.
Why the click-through rate moves too
A 180% rise in click-through rate next to a 95% fall in cost per sign-up is the most instructive row in the table. The creative did not change. The bidder stopped showing the ad in conversations where people tapped it and then left, and started showing it where people tapped it and signed up. Click-through rate rose because relevance rose, and cost per result fell for the same reason. When you see this pattern on your own account, the earlier campaign was appearing in the wrong contexts, and the fix for a click campaign is the context hints; Conversion Optimization simply finds the same answer from the data.
How to switch without losing the comparison
- +Clone the existing click campaign into a Conversion Optimization campaign. Ads Manager prefills the conversion event when exactly one eligible event exists and asks you to pick when there are several.
- +Keep the click campaign running with a share of the budget for at least two weeks. The retail case in OpenAI's pack is a 40% split measured side by side, which is the cleanest comparison of the six.
- +Judge the result on cost per result and on volume together. A lower cost on a third of the conversions is a different outcome from a lower cost on more of them.
- +Expect the numbers to move in steps. The bidder needs conversions to learn from, so the first days after a switch tell you little; the second week is where the comparison starts to mean something.
The small print, kept honest
Everything above is observational and comes from the platform. The benchmark sample is 27 advertisers who chose to run both strategies, the case studies were selected by OpenAI, and the CAPI figure is partly a measurement effect: the same spend divided by more conversions counted. None of that makes the direction wrong, and the mechanism is the same one every other ad platform has followed. It does mean the number that counts is your own two-week comparison, run with tracking you trust. When you would rather have that test built, tracked and read for you, it is part of the managed service.
