ChatGPT Ads: What to Configure on Day One

As of September 16, conversion-optimized campaigns with impression billing in ChatGPT Ads are generally available. The platform charges for every impression served, but delivery optimizes toward conversions that can come from either a click on the ad or a view with no click, and OpenAI describes it, verbatim, as "our recommended model for maximizing performance." For a US advertiser already running self-serve ChatGPT Ads, this isn't an exotic option anymore: it's the setup the platform is pushing from day one.

The model OpenAI recommends: you pay per impression, delivery optimizes to click and view

The impression-billed campaign type charges for every impression served, not per click, while delivery optimizes toward a conversion goal that counts both ad clicks and ad views. Until a couple of weeks ago this was the feature OpenAI had flagged as coming; now it's live, and the platform isn't presenting it as one option among several: it's naming it the model it recommends for maximizing performance. That changes the first day-one decision: optimizing to click conversions only is no longer what the platform recommends.

Which window to look at, now that you can choose

When customizing columns in Ads Manager, advertisers can now pick a 7-, 14-, or 30-day click window and a 0- or 1-day view window for conversion reporting, and that preference applies to every conversion column shown. It's worth picking the window based on the product's decision cycle instead of leaving the default unexamined: a 7-day click window suits fast purchase decisions, 30 days suits longer consideration, and the view window decides whether an impression with no click earns credit at all: at 0 days it doesn't, at 1 day it does if the conversion lands within that day. This is a decision about what shows up in the report, not about what the account gets billed for: according to specialized coverage (aixh.com), this window setting "only changes reporting: optimisation, bidding and billing stay the same, and attribution remains last-touch." As of September 3, help.openai.com described view-through conversions separately as "supplemental reporting" that "do not affect CPA, post-click conversion rate, bidding, billing, or conversion optimization." We could not confirm whether that page changed with the launch.

The same meter that bills you per impression is the one counting the views it sold you

The impression-billed campaign is run, measured and billed by the same platform: OpenAI decides which impression counts as a view-conversion, credits it as the result of its own optimization, and bills for that same impression. That isn't an accusation of bad faith, it's the shape of the incentive: when the party selling the inventory is also the party certifying whether it worked, that number is worth checking rather than taking at face value just because it comes from the account dashboard. Before shifting real budget into this model, it's worth testing it against a measurement that doesn't depend on the same platform, for example a holdout test or a geo test that isolates the actual effect of the views. For brand presence inside ChatGPT there are already third-party tools like Profound Ads Studio, but for conversions the bar is still a test you run yourself.

Event Quality Score grades whether the setup is built right, not whether the campaign is working

Event Quality Score, now available in Ads Manager under Tools > Conversions, gives each eligible Pixel or Conversions API data source a 1-10 setup quality score, assesses user data coverage, and provides actionable recommendations to improve conversion event implementation. It's a setup grade, not a performance grade: it tells you whether the pixel or Conversions API is sending well-formed data, not whether the campaign is converting better or worse. The logic tracks what Google rolled out recently with the Data Strength Uplift Metric, which also scores the quality of an account's first-party data setup, though there the number is a percentage of recovered conversions and here it's a 1-10 score. In both cases the platform grades itself: the score is calculated by the same company running the campaign.

Who rewrites the ad copy: here it's something you turn on, on Google it already comes turned on

Text customization in ChatGPT Ads is opt-in: it adapts existing headlines and descriptions to fit the context of the conversation and can translate copy into a user's preferred language, but the advertiser decides whether to enable it. That's the opposite direction from what Google did with AI Max: there, when a Search campaign gets auto-migrated and it originated from Automatically Created Assets, Text Customization starts turned on by default and it's Google rewriting the copy without the advertiser asking for it. Same feature name, opposite default: here the copy stays exactly as uploaded until someone decides otherwise.

The rest of day one: splitting by platform, and what's still a US test

Ads Manager and the Advertiser API now let advertisers target Android app, Android web, desktop web, iOS app and iOS web separately, and performance insights can be segmented the same way. For a US account, that matters at the campaign-structure level: splitting mobile app from mobile web before there's enough volume to optimize each segment can fragment learning, so it's worth deciding how granular to go on day one instead of segmenting everything by default.

What follows isn't generally available yet, and a US buyer already sits inside the geography these tests are scoped to:

OpenAI is testing Sponsored Agents with select US advertisers: after tapping an ad in ChatGPT, a user may be able to start a clearly labeled conversation with the brand's agent to ask follow-up questions and explore the offer. It's worth asking OpenAI directly whether an account qualifies for the test.
The Shopify integration is live now for US-based merchants, with international availability planned for later this month, and the HubSpot integration lets advertisers create ads and follow up on leads without leaving either tool.
AI-assisted ad creation, suggested copy and imagery pulled from the landing page, is still in testing everywhere, with broader availability expected in the coming weeks.

OpenAI hasn't published how this campaign type's performance compares, in practice, to optimizing on click conversions only. Until an independent benchmark exists, the only confirmed fact is that the platform has already decided which model it recommends.

This content was developed with AI assistance and reviewed by the Zenda team. Any bad ideas are entirely our own.