1. What Was Added in September
The Ads Manager itself is young. It launched on 7 September 2026, with a minimum budget of €15 per day and a choice between clicks, reach and conversions as the objective. Nine days later came the package this article is about: the integrations into two systems in which sales and retail already work, plus two adjustments to billing and attribution.
The difference between these two dates is greater than the nine days suggest. On 7 September, ChatGPT was one more channel for which someone had to open a separate interface, build separate campaigns and collect separate figures. After 16 September, it can be docked onto the places where the work happens anyway. In practice, that decides whether a channel gets used more often than its reach does, because a separate workflow is the first thing to be dropped in day-to-day business.
Everything that follows reflects the status as of September 2026. The platform is a few weeks old, and experience shows that in this phase both the feature set and availability change faster than the documentation about them.
2. The HubSpot Connection
The HubSpot integration covers three things: creating ads, tracking performance and following up on leads — without leaving HubSpot to do so. The channel therefore sits in the same place as the other ad sources that HubSpot has been collecting for some time.
The practical gain does not lie in ad creation. An ad can just as well be built in a second interface; that costs time once. It lies in lead tracking: an enquiry that comes from ChatGPT carries a source in the CRM that reaches further than the first form page. Whether it turned into an order can thus be attributed to the same source as with the other channels.
This is, however, exactly where the caveat lies. A source entry in the CRM is only as reliable as what the platform passes on about it, and as the discipline with which the records are maintained afterwards. A new field in the CRM quickly produces a figure that turns into a budget decision, even though the basis is thin. If you already have clean channel attribution in HubSpot, you get one more row here. If you do not, you get one more row with unclear meaning.
3. The Shopify App
For Shopify there is an app, and it is free of charge. It launched in the US on 3 September 2026, and since 23 September it has been available internationally in all markets in which ChatGPT Ads runs. It handles the catalogue integration, installs the pixel automatically and allows campaigns to be managed from within the shop; according to the listing, product inventory is synchronised automatically between the shop and ChatGPT.
The automatic pixel installation is the least conspicuous and at the same time the most important feature in this list. A missing or incorrectly installed tracking pixel is the most common reason why campaigns in a new channel run without usable data, and the error only becomes apparent after weeks. The fact that the app takes over this step removes a source of error that is disproportionately expensive in the early phase of a channel.
The same applies to the automatic catalogue sync as to any feed: what is poorly maintained in the shop does not get better by being transferred. The product data that already flows into other channels is the basis here too.
4. Impression-Based Billing with a Conversion Goal
Conversion optimisation with impression-based billing is generally available and is listed by OpenAI as the recommended model for performance. Optimisation targets conversions from clicks and views; billing is by impressions.
This is the combination that deserves the most attention from a buying perspective. What is billed is an event that occurs without any further action by the user; what is optimised for is an event that usually occurs only later and somewhere else. Between the two lies the quality of the platform's model — and with a channel that has been open for a few weeks, that has naturally not yet been tested on many accounts.
That is not an argument against the model. It is an argument for running it with a budget whose loss would have no consequences, and for holding the results against your own count rather than against the platform's report alone. If you have the choice, start with clicks or conversions as the objective and switch only once you have enough data points of your own to judge the difference.
How impression-based billing can be introduced step by step and in a way you can verify.
5. Attribution Windows and Platform Breakdown
The click window can be set to 7, 14 or 30 days, the view window to 0 or 1 day. Both are settings that quietly determine the number of reported conversions without anything changing in the campaign.
Two recommendations follow from this. First: the window is set once, aligned with the length of your own decision path, and not touched again afterwards. Any later change renders the comparison with the previous month worthless. Second: setting the view window to 0 is the conservative choice. With 1 day, conversions are counted that would have occurred without the ad as well — with a new channel whose contribution nobody can yet assess, that is the more uncomfortable kind of imprecision.
The two attribution windows in the Ads Manager with their values and the recommendation from the text.
For targeting, five breakdowns are available: Android app, Android web, desktop web, iOS app and iOS web. The benefit lies less in excluding than in evaluating separately. App and web usage often differ markedly in session length and readiness to convert, and if you have this separation in your reports from the outset, you can see after four weeks where a budget is working and where it is merely running.
6. What This Means for the Buying Side
Taken together, the result is a channel that can be bought like a normal performance channel: a connection to the system in which the lead lands, an automatically installed tracking pixel in the shop, a billing model with a conversion goal and adjustable windows for attribution. That is the equipment a channel needs to get beyond a test phase.
What it is not: a reason to reallocate budget. The equipment says nothing about the results. It only says that the results can now be read in the same reports as those of the other channels — and that is the prerequisite for being able to judge at all whether the channel is worthwhile.
The sensible approach is therefore a small, cleanly measured set-up that runs for long enough. With a minimum daily budget of €15, the entry barrier is low enough that the question of budget is not the decisive one. The decisive question is whether somebody reads the figures at the end.
7. Checklist
- Check whether the HubSpot integration and the Shopify app are available in your own market
- Install the Shopify app and verify that the pixel has actually been set
- Check the catalogue data in the shop for titles, prices and availability before the first sync
- Define the click attribution window once — based on the length of your own decision path
- Set the view window to 0 days as long as the channel's contribution is unknown
- Evaluate the five platform breakdowns separately from the start
- Start with clicks or conversions as the objective; impression-based billing only once you have your own data points
- Choose a daily budget such that a total loss would have no consequences
- After four weeks, hold the platform figures against your own count in the CRM or shop
- Set a fixed date on which somebody looks at the figures and decides
8. Conclusion
The real news is not in the features but in the direction. OpenAI is expanding the Ads Manager at the points where advertisers do not have to rebuild their systems — CRM, shop, billing, attribution. This is the unspectacular work that takes a channel out of the test phase, and for an ad market that is three weeks old, it comes unusually early.
My assessment: the effort of getting started here now is low, and the insight gained after four weeks is real. What is missing is any reliable indication of what the channel delivers. There are no benchmarks, no experience from other accounts and no way of assessing the model quality of the conversion optimisation — anyone who chooses the recommended impression-based billing is relying, with a system that is a few weeks old, on an attribution they cannot verify. That is precisely why this entry belongs in a budget whose loss nobody would notice, and not in a reallocation from a channel that is already delivering results.
9. Frequently Asked Questions
Does the Shopify app cost anything?
No, the app itself is free of charge. You only pay for the ads that are run. Since 23 September 2026 it has been available internationally in all markets in which ChatGPT Ads runs.
Is the HubSpot integration worthwhile without existing channel attribution?
Only to a limited extent. The integration carries the source into the CRM, but the value only arises in comparison with other channels. If you do not yet maintain this attribution cleanly for the other sources, you get a figure without a reference point.
Which attribution window is the right one?
That depends on the length of your own decision path, not on a general recommendation. More important than the choice itself is that it is made once and not changed afterwards, because every change devalues the comparison with earlier periods.
Should impression-based billing be used from the start?
OpenAI lists it as the recommended model for performance. Even so, it only makes sense once you have data points of your own, because with billing by impressions the entire performance depends on the platform's optimisation. Starting with clicks or conversions is the more verifiable option.
What is the benefit of the breakdown by platform?
Above all, separate figures. App and web usage usually differ markedly, and if you evaluate the five breakdowns separately from the start, you can tell after a few weeks which part of the budget is working.
Is €15 per day enough for a test?
For a reliable statement on conversion performance, generally not; for an initial assessment of reach, click prices and enquiry quality, certainly. The more meaningful variable is not the daily budget but the duration: four weeks without intervention say more than two weeks with double the budget.