For ChatGPT Ads conversion measurement, the strongest setup uses both methods OpenAI offers: the browser-based OpenAI Pixel and the server-side Conversions API. Each sends conversion events from a different place. Used together, they can give OpenAI more chances to receive the signals needed to connect an action on your site with an ad.
That doesn’t mean every order will appear in Ads Manager. A conversion must match an event configured for the campaign, fall within its attribution window, and have available signals that connect it to an eligible ad click. Set up the measurement before using the reported total to judge campaign performance.
Why one measurement method can miss part of the picture
The Pixel sends events from a visitor’s browser. The Conversions API sends events from your server, such as when your order system records a completed purchase. Those sources complement each other: browser tracking can supply signals tied to a visit, while server reporting can send the conversion recorded by your business systems.
Browser settings, consent choices, or storage limits can affect which signals the Pixel can collect. Server-side reporting adds another route for sending events, but it still needs useful matching information. Neither method can fix an event that’s misconfigured or outside the campaign’s attribution window.
What the OpenAI Pixel captures
The OpenAI Pixel setup documentation describes a browser tool you install on relevant site pages. When a visitor takes a measured action, such as making a purchase or submitting a lead form, the Pixel can send an event.
One important signal is oppref, OpenAI’s click reference. It’s added to the landing-page URL after an ad click. The Pixel captures it and stores it in a first-party cookie, so it can be associated with later conversion events. If a redirect or a change in the page URL drops that reference, you may lose a useful connection between the click and the conversion.
The Pixel relies on the browser, so conditions such as consent settings and storage restrictions can affect what it records. That’s one reason browser tracking alone may leave gaps.
What the Conversions API adds
The OpenAI Conversions API documentation describes a server-to-server way to send conversion events. Your order or lead system can report an action as part of its backend process. OpenAI calls the API a more reliable tracking source than the Pixel alone.
The API must be called from your server, not directly from page code. It also doesn’t capture oppref automatically. Your site or server needs to preserve the click reference through redirects and page navigation, then include it with the server event when available. Without it, the event may still be received, but OpenAI has one fewer signal to connect it to an ad click.
Use both, then deduplicate the same conversion
When your setup supports it, use the Pixel and the Conversions API together. The Pixel sends browser-side events; the API adds events from your business systems. OpenAI recommends the combined approach for more resilient measurement.
If both sources report the same purchase or lead, give that conversion the same event ID on each. OpenAI uses the Pixel ID, event name, and event ID to recognize matching events, then keeps the first matching event it receives and ignores the later duplicate. For custom events, the custom event name must match too.
For example, if an order is recorded as order_12345, use that same unique ID for the purchase event sent by the Pixel and the corresponding event sent by your server. Don’t send two unrelated IDs for one order and expect OpenAI to know they’re the same. Duplicate reporting is a small data-entry problem with an impressive ability to inflate a chart.
Set up the signals before judging campaign results
Start by creating a data source in Ads Manager. Choose conversion events that match the actions you want to measure, such as a completed order or a submitted lead form. Check that the event is configured for the campaign and that it occurs within the campaign’s applicable attribution window.
Next, preserve oppref through redirects and landing-page navigation. Pass it with server-side events when it’s available. Then confirm that a conversion sent through both methods uses the same event ID and event name, so OpenAI can deduplicate it.
Measurement also has to follow your privacy obligations. Explain what data your site or app collects, send conversion data only when permitted, and obtain any consent required by law. OpenAI’s conversion measurement guidance covers the signals and setup checks that can improve measurement quality.
Different dashboards may show different totals
Ads Manager, your analytics platform, and other ad platforms may not report identical conversion totals. They can apply different attribution methods or windows, use different time zones and date boundaries, handle deduplication differently, or have different campaign and event settings. Browser, consent, and storage conditions can also change which signals are available. Where modeled measurement is available, it can affect reported totals too.
A difference doesn’t automatically mean tracking is broken. First compare the event definitions, attribution windows, reporting dates, and deduplication rules. Then check whether the Pixel and server events are firing as intended and whether the click reference is reaching the server. You’re comparing systems with different rules, not necessarily two copies of the same report.
A practical measurement checklist
- Install the OpenAI Pixel on relevant pages and send events when the intended action happens.
- Send server-side events through the Conversions API, called from your server.
- Preserve
opprefthrough redirects and include it in API events when available. - Use the same event ID and event name for a conversion sent through both methods.
- Confirm the campaign’s conversion event and attribution window in Ads Manager.
- Review your data disclosures and obtain required consent before sending conversion data.
- Compare reporting rules before treating differences between dashboards as a tracking problem.