,

ChatGPT Ads Just Grew Up: What CPA Bidding, Audience Uploads, and Product Feeds Mean for Your Q4 Budget

,

Earlier this year, OpenAI launched ChatGPT self-service ads in the US, giving advertisers direct access to one of the world's fastest-growing AI platforms. The announcement made headlines, but the bigger story is what's happened since.

When the beta rolled out in May, we published a step-by-step guide to setting up campaigns, highlighting the features that were still missing. In the weeks since, the platform went live in four more countries and has rolled out new capabilities, including CPA bidding, product feeds, and custom audiences.

Let’s break down everything you need to know about these changes and what they mean for your Q4 planning.

CPA Bidding Is Live. Is It Ready for a Real Budget?

When ChatGPT ads first launched, clicks and reach were the only campaign objectives advertisers could set. As of June 5, that is no longer the case. Per Digiday and The Keyword, OpenAI activated conversion-optimized CPA bidding, meaning advertisers only pay for a completed action like a click-through, sign-up, or purchase, while OpenAI absorbs the cost of any impression or click that doesn't convert. This shift allows ChatGPT Ads to optimize the lower funnel, where it can tie campaigns directly to outcomes. 

Our hypothesis is that the current model will evolve as it scales. It's possible ChatGPT Ads eventually shifts toward something like Google's Target CPA bidding, where advertisers pay for every click, whether it leads to a conversion or not, but the system automatically adjusts bids to hit a target CPA.

The ultimate question is whether ChatGPT ads are worth the spend. So here's a straight answer: CPA bidding is real, but the measurement underneath it is still green. OpenAI's dashboard is currently the only source of truth on your CPA, because there is no independent, third-party verification yet. Digiday reports that OpenAI's global head of ads, David Dugan, called the arrival of third-party measurement "a natural evolution." This is not a reason to write off the feature but a reminder to treat the number on your dashboard as directional, not complete, and use a test budget against CPA objectives. 

A few things to note: CPA bidding wasn't formally announced. The update came through an email to pilot advertisers, which Digiday later reported on, revealing that only accounts with Pixel or Conversions API set up prior to June 1st would be granted early access on June 5th. Businesses that missed that window weren't part of the first wave. OpenAI hasn't given a broader rollout date yet, but we expect this gate to matter less as the platform matures. For now, the safest move is to get your tracking in place before the next wave, whenever it comes. 

Product Feeds are Here. What That Means for Ecommerce

OpenAI has officially shipped product feed ads in Ads Manager, allowing retailers to upload their current catalog directly into the platform and automatically generate ads from it, instead of building them one by one. Search Engine Land broke the news, sourced from an email JXT Group founder Menachem Ani received directly from OpenAI. The goal is to bring more inventory into the platform at scale and put the right products in front of users when intent is highest. For ecommerce businesses already running PMax, Google Shopping or Meta's catalog-based dynamic ads, the mechanics will feel familiar.

According to OpenAI, catalog ads are already outperforming other formats, but there are a few practical details worth knowing before you get started: feeds require a minimum of 1,000 products and can contain up to 2 million, but each ad account can only connect one feed at a time, so multi-brand or multi-region catalogs will need to be consolidated. Products in your feed won't show up in organic ChatGPT conversations, just in paid placements, though OpenAI has signaled that could change down the line. If you're ready to get started, OpenAI's Help Center walks through the full setup, from creating your feed to launching your first campaign.

Audience Uploads vs. Customer Match

The rumors are true, custom audiences are live. Custom audience uploads are ChatGPT's version of first-party targeting. It's the same idea behind Google Customer Match or Meta Custom Audiences: take a list of existing customers or prospects, upload it, and use it to tell the AI who gets served the ad.

The good news is that ChatGPTs identifier formatting rules are similar to Google and Meta, so using an email address or phone number from lists you've built for those platforms should upload here with little to no lift. Once that list is uploaded, you can put it to work in three ways: use it as an inclusion audience to limit a campaign to only people on that list, useful for reaching known customers or qualified prospects directly. Use it as an exclusion audience to keep a campaign from showing to people on the list, to avoid re-targeting existing customers or recent purchasers. Or apply it as a bid adjustment, raising or lowering how competitively you bid depending on the value of a customer.

What doesn't carry over automatically is your match rate. ChatGPT's identity graph is younger than Google's or Meta's. Some contacts may not use ChatGPT, others may have signed up with a different email or phone number. Expect your matched size to run lower than your uploaded list once duplicates and unmatched identifiers get filtered out. Craig Graham, CEO of Grayvault Consulting, told MediaPost that a modest match rate is perfectly normal.

ChatGPT requires a minimum of 25,000 matched users before an audience can go live, but recommends at least 100,000 for better measurement, so prioritize volume and quality when deciding which list to use first. There's also no lookalike modeling yet, so a small, high-quality list can't be used to find new, similar prospects the way it can on other platforms.

The Q4 Decision Framework

So where does that leave your Q4 budget planning? 

If your business has been quietly testing ChatGPT Ads since Q2 or Q3, this is your signal to formalize a Q4 line. Confirm your conversion event has been live long enough to qualify for CPA bidding, get your product feed submitted and approved, and set a real number against the channel. Do this now, and you'll be optimizing on live data by the time Q4 peak spend hits.

Adthena's research on AI-driven search found that brands cited within Google's AI Overviews saw a 91% higher paid CTR than brands that weren't cited (7.89% vs. 4.14%). That's not a ChatGPT-specific number, but it points to the broader theme of where conversational platforms seem to be heading: visibility inside the AI response itself is starting to shape how paid placements perform. Brands that pay attention to how they show up in AI-generated answers, may be the ones who see this channel pay off first.

If you haven't started at all, it's not too late to get in position, but the clock is ticking. Account verification alone has been running 5 to 15 business days, and feed review adds more time on top of that. If your Q4 planning locks in the next few weeks, get your account verified and your Pixel or Conversions API live now, even if your strategy isn’t fully baked in.

The pattern here is clear: this technology is evolving faster than most clients plan for. Brands with clean measurement and organized first-party data get to act the moment a new feature rolls out. Everyone else spends their first month catching up. As always, our team is watching every update as it lands and we'll keep you posted on what's worth acting on.

Sources