OpenAI Publishes Its First ChatGPT Ads Benchmarks

OpenAI has circulated its first set of written performance benchmarks for ChatGPT Ads, a 48-slide "Insights, Performance and Measurement" pack dated late August 2026. It announces Conversion Optimization in general availability, puts numbers on the Conversions API and on product feeds for the first time, publishes a seasonality curve for shopping activity inside ChatGPT, and, in a footnote, reveals that more than 10,000 advertisers cleared $100 of spend in a single month.

The three benchmarks

Every one of them carries the same qualifier in OpenAI's own footnote: observational results, not causal lift.

Conversion Optimization is generally available, for services too

In August, conversion-optimised bidding was a beta limited to product feed campaigns, which meant a services business could not use it. The pack headlines "Conversion Optimization now in GA" and lists the events it optimises toward as leads, orders, registrations, trials, subscriptions and content views. Two of its six anonymous case studies are services advertisers: a financial services advertiser whose cost per lead fell 50%, and a jobs and recruiting advertiser whose cost per sign-up fell 95% while click-through rate rose 180%.

That second pair of numbers is the most useful thing in the pack. A click-through rate that jumps 180% without a creative change means the bidder started choosing different conversations to appear in. Conversion bidding on this platform is context selection automated against your own conversion data, and the requirement OpenAI applied to its own sample, at least five click-attributed conversions on the event, is the practical floor before switching. Our Conversion Optimization guide has the requirements, the six case studies and the sequence.

The six case studies

Each is labelled "observed and may reflect other campaign changes", and the travel case adds that the gain coincided with a change in bid mix and the conversion signal ramping up. Every advertiser in the table had the pixel, the Conversions API or both running before the switch.

The Conversions API has a number

Pixel plus Conversions API measured 57% more attributed conversions than the pixel alone, and a 28% lower measured CPA, across more than 1,000 advertisers. Most of the CPA drop is the same spend divided by more conversions counted, which OpenAI half-acknowledges by calling it "measured CPA". It still matters, because conversion bidding can only steer toward what it can see. The conversion-tracking guide now carries both figures.

The footnote that matters most

The product feed benchmark is a retail figure and does not apply to a services advertiser. Its sample does. To qualify, an advertiser needed at least $100 of spend, 100 clicks and three active days between 16 June and 15 July 2026, and the footnote reads n=10,000+. That is the first advertiser count OpenAI has put in writing, it is a floor because it is the eligible sample only, and it is a fair measure of how busy the United States auction already is. European advertisers, who received self-serve access on 31 August, are entering an auction that is a week old.

A seasonality curve, with caveats

Three slides chart 2025 US product-search activity inside ChatGPT against a matched-week baseline. Back-to-school built for five weeks from a late-July trough to a 26% average lift in August, led by personal care (+46%) and apparel (+42%). Labor Day week reached +24%. Cyber 5 averaged +47% with Black Friday at +94%, led by computers and appliances at +94% each, and activity was back to baseline by 3 December. It is US data, from 2025, measured as product-search activity rather than ad impressions, and the lifts are relative to the preceding four weeks. It is also the first seasonality curve anyone has for the channel, and a direct input to Q4 planning for any ecommerce advertiser.

Category volumes, and what they are not

The pack also lists average weekly ChatGPT user messages by category for July 2026: retail and ecommerce above 1.5 billion (up 74% since January), healthcare above 2.4 billion, professional services above 2 billion, B2B technology above 490 million, beauty and personal care above 450 million (up 44%), travel above 330 million. OpenAI's own footnote on every slide is the one to keep: "Volume is not ad inventory or verified purchase intent." Categories overlap, so they do not add up, and the growth figures compare July with January because a July 2025 comparison does not exist.

Who is on record

Update, 1 September 2026, 19:25: OpenAI confirmed to us in writing that the pack is public and may be shared in full, so the five named pilot advertisers can be quoted. None of them describes the channel as a traffic source. Each names a job it does.

Best Buy, Amy Adams, Vice President of Media:

What we're seeing from early results is encouraging. Click-through rates were higher than we anticipated, showing that customers are engaging with the ads in meaningful ways and reinforcing that this is a natural place for discovery, and that we're reaching people with real intent.

Best Buy also reports that the pilot uncovered demand across a broader mix of categories than expected, including unexpected areas of strong engagement.

VistaPrint, Garima Singh, Senior Manager of Paid AI Media:

The majority of traffic ChatGPT drove was new visitors, consistent with its discovery-oriented nature. This is an encouraging early signal that the platform has the potential to reach customers who aren't already in our ecosystem.

VistaPrint's other learning is the creative one: as it shifted from product-oriented messaging to use-case-driven creative, click-through rates improved.

HubSpot, Jonathan Tablada, Lead Marketer in Paid Advertising:

It's not often that marketers get the opportunity to test an entirely new advertising channel. Seeing ChatGPT ads drive real business outcomes has been encouraging, and it's given us another way to diversify our paid media strategy while reaching customers in a more conversational environment.

HubSpot reports competitive CPAs across a number of campaigns and downstream revenue signals beyond top-of-funnel metrics.

Figma, Reema Batta, VP of Growth Marketing:

The OpenAI ads platform helped us connect with people when they were actively exploring solutions like Figma. The early engagement and results we're seeing suggests this could become a meaningful new way to connect with users.

Figma says ChatGPT Ads has become a top performer for Figma Make campaigns, with higher click-through rates than traditional search and stronger-than-average onsite engagement.

Canva, Patricia Tan, Global Growth Marketing Technology and Innovation Lead:

There's no playbook for advertising inside AI yet, and we can see the platform improving as we go. We'd rather help write some of that playbook with OpenAI than watch it from the outside.

Read together, the five describe discovery, evaluation, new-visitor acquisition, a performance line next to search, and demand discovery across categories. That list is a better brief for a first campaign than any CTR benchmark.

What to do with it

The order of operations is the whole lesson. Install the pixel and the Conversions API together, pick one event and let it reach five click-attributed conversions, clone the click campaign into a Conversion Optimization campaign and run both for two weeks. The performance channel guide, which said in August that this page would be rewritten the day conversion bidding reached click campaigns, has been rewritten.