On 16 September 2026, OpenAI announced a series of new features for ChatGPT Ads: a test of commercial agents called Sponsored Agents, AI tools in Ads Manager, and HubSpot and Shopify integrations. The same day, according to PPC Land, advertisers received a second batch focused on measurement: conversion optimisation with impression billing, adjustable attribution windows, targeting across five platforms and an event quality score. The package tightens measurement on existing formats and introduces a format for which OpenAI has disclosed neither the price nor the metrics.
The facts
- OpenAI's help page describes conversion-optimised campaigns with impression billing: delivery is optimised on conversions that follow a click or a view, and the invoice is calculated on impressions served. PPC Land reports that OpenAI presents this model to advertisers as the one it recommends for performance. Two limits are documented: only standard conversion events are accepted, and the billing model can no longer be changed once the campaign has been created.
- In Ads Manager, advertisers now choose an attribution window of 7, 14 or 30 days for clicks, and 0 or 1 day for views, applied to every conversion column displayed. OpenAI specifies that this setting only changes reporting: optimisation, bidding and billing stay the same, and attribution remains last-touch. Targeting now covers five separate platforms (Android app, Android web, desktop web, iOS app, iOS web), according to PPC Land.
- OpenAI is testing Sponsored Agents with select advertisers in the United States: after an ad, the user can open a "clearly labeled" conversation with the business-sponsored agent, distinct from ChatGPT's independent answers and separate from the original conversation, then follow a link to the website (Search Engine Roundtable). The help page for the format refers to a limited alpha test and states that OpenAI is not accepting early access requests.
How do you measure the performance of a sponsored conversation?
Most of this update follows a logic familiar to Google Ads and Meta Ads advertisers. Targeting is split into five segments, with performance reports segmented the same way. The event quality score, which PPC Land places in Ads Manager under Tools > Conversions with a rating from 1 to 10, assesses the Pixel or Conversions API setup and flags the fixes to make; OpenAI specifies that it is recalculated every day over seven complete days and that it does not predict campaign performance. On the tooling side, HubSpot has since 16 September let businesses create ads, track performance and follow up on leads without leaving the CRM. The Shopify app is reserved for US merchants at launch and will open on 23 September in the markets where ChatGPT Ads is available, Luxembourg included.
Sponsored Agents change that framework. A click opens a conversation with the brand's agent, before any visit to a landing page. OpenAI has disclosed neither the price of the format nor the success metrics planned, and does not say how that conversation will be counted: by the initial click, the number of exchanges, the link followed to the website or the sale. The company does not specify either what the advertiser will be able to see of the content exchanged. For a channel that has just gained attribution windows set to the day, it is the only lever in the update that does not yet fit the same reading grid.
Impression billing: the settings to fix before testing
With impression billing, the advertiser pays for every display, whether or not the user clicks. The model is judged on the cost per conversion obtained, compared with that of a click-billed campaign over the same period. Because the billing model is fixed at creation, that comparison requires two separate campaigns, each with a capped budget. The attribution window chosen should be the same as on the other platforms in the media plan: with different windows, conversion columns can no longer be compared from one platform to another.
AIxH's view
We apply the same rule to every new channel tested for a client: measure first what can be measured, and frame the rest within a limited budget. Impression billing, attribution windows and platform targeting fit right now into a scoped ChatGPT Ads test. Sponsored Agents will wait for a price and a success metric. If you are assessing ChatGPT Ads for the coming months, start with the building blocks that can already be measured.
