ChatGPT Ads' September Update: View-Through Billing, Platform-Level Targeting, and Localized Copy

What does "view-through billing" actually change about cost per result?

Under click billing, you pay when someone clicks; a conversion that happens after a view with no click doesn't cost you anything directly, and whether you see it at all depends on your attribution window and model. Under the new conversion-optimized, impression-billed campaigns, you pay per impression, and the system optimizes toward conversions from both clicks and views, meaning a conversion can now be attributed and billed for without a click ever happening.

The immediate consequence: your "cost per conversion" benchmark on ChatGPT Ads is no longer directly comparable to a click-billed campaign on the same platform, let alone to a click-attributed campaign on Google or Meta. Running both models side by side without accounting for the billing basis is how a team ends up comparing two different kinds of numbers and drawing a conclusion neither one supports on its own.

The edge case worth planning for: a low-click, high-impression campaign (top-of-funnel awareness copy, for instance) will look structurally different under impression billing than it did under click billing, and that difference is mechanical, not a signal that the creative suddenly got better or worse.

How does platform-specific targeting change how you'd structure a campaign?

Before this update, ChatGPT Ads treated its surfaces (mobile app, mobile web, desktop) as one undifferentiated environment for targeting and reporting purposes. Now you can target Android app, Android web, desktop web, iOS app, and iOS web independently and see performance broken out by each.

That's a meaningful jump in granularity, and it raises the same question every platform's first pass at surface-level targeting raises: is the performance difference between iOS app and iOS web real user-behavior signal, or an artifact of a newly-split, thinner data sample per surface? Until a platform's per-surface reporting has run long enough to be statistically stable, the temptation to over-optimize toward whichever surface shows the best early number is worth resisting, the same caution that applies to any newly-introduced segmentation on any platform.

Does automatic ad-copy localization change how you should write for this platform?

The new text customization adapts your headlines and descriptions to conversation context and translates them into the user's language. Practically, that means the copy a user sees may not be the exact string you wrote. It's a variant generated from your input, tailored to what the person was asking ChatGPT about and what language they're using.

This is a meaningfully different creative model than platform ad copy has generally worked under. On Google and Meta, your written headline is (with some Advantage+/PMax-style asset-shuffling exceptions) largely what gets shown. Here, the baseline assumption shifts toward your copy being a set of inputs an AI system adapts per conversation, not a fixed asset. The practical takeaway: write source copy that's accurate and on-brand at the level of claims and tone, since that's what needs to survive adaptation. A headline built on a very specific wording flourish is more fragile in this model than one built on a clear, simple claim.

What do Event Quality Scoring and the Shopify/HubSpot integrations add?

Event Quality Scoring is the measurement counterpart to the billing change: a way of assessing how reliable a given conversion event is, which matters more once the platform is billing and optimizing toward view-based conversions rather than the more directly verifiable click-then-convert path.

The Shopify and HubSpot integrations matter for a more practical reason: they connect commerce and CRM data directly, which is the same first-party-data-quality problem every platform is racing to solve right now. See also Google's own Data Strength Uplift and Meridian upgrades, shipped roughly a week earlier. The pattern across platforms in September 2026 is consistent: better first-party data pipes, framed as a performance upgrade, delivered by the platform that also grades its own performance against that data.

What should a team running ChatGPT Ads through an execution layer check first?

Three things, before letting the new settings run unattended.

Re-baseline cost-per-result, don't carry over the old number. If you're moving a campaign to the impression-billed, view-through-inclusive model, your existing cost-per-conversion benchmark from click billing isn't the right comparison point. Treat the first two weeks as a new baseline, not a regression.

Decide whether surface-level targeting needs its own hypothesis before you split budget by it. New granularity is genuinely useful, but only once there's a real hypothesis about why iOS app and iOS web should be treated differently for your specific offer. Splitting budget by surface just because the option now exists isn't a strategy.

Confirm what "the ad" means for QA purposes now that copy adapts per conversation. If your compliance or brand review process assumes a fixed, approved headline goes live unchanged, that assumption needs revisiting under a system that generates contextual variants from your source copy.