Attribution and customer journeys

Investigate duplicate ChatGPT conversion events across browser and server

Diagnose duplicate ChatGPT conversion events between pixel and server. Check shared identity, retries, source alignment and conflicting purchase values.

Start reading
Editorial illustration: Light from a glass pear travels through two prisms and meets in one pear-shaped silhouette.
Editorial illustrationTwo delivery routes can describe the same business action. Shared identity needs checking in both payloads.
The working guide

What you can work through.

Attribution and customer journeys
  • Trace one business action through both actual delivery payloads.
  • Keep retry identity stable while giving distinct actions distinct IDs.
  • Do not assume a duplicate resend corrects an earlier payload.

Installing server-side conversion delivery alongside a browser pixel creates two routes for the same business action. That can improve the resilience of collection, but it also creates a diagnostic question: are the two messages recognized as the same action, or are they being treated as separate events?

For a ChatGPT advertiser, start with one known action and trace both routes. Do not infer duplicate counting merely because two delivery attempts appear in logs. Multiple transmissions can be intentional. The issue is whether their shared identity is correct and whether the resulting interpretation matches the documented deduplication behavior.

Compare the identity, not the visible order label

OpenAI’s deduplication uses the same Pixel ID, event name and action identity across both routes. The browser’s options field event_id must match the server event’s id; the fields have different names but carry the same value. Custom events also need matching custom_event_name values. Compare this complete combination, not just one shared identifier.

An internal order number shown in an admin screen is not evidence that both payloads sent the same value. Capture the relevant fields from each actual request through your permitted diagnostic tools. Preserve the event source, name, identifier and send time while excluding unnecessary personal data.

In a hypothetical failure, the browser sends order-451 and the server sends a newly generated random value for the same purchase. Both messages can look valid individually while failing to express that they describe one action. The repair concerns shared identity, not deleting one route without understanding it.

Check repeat behavior on each route

Reloading a confirmation page can trigger another browser transmission. A delayed acknowledgment can trigger a server retry. If each attempt creates a fresh action identifier, the implementation turns a transport retry into a new apparent business event.

Reproduce the behavior in an appropriate test environment. Complete one action, inspect its identifier, repeat the relevant delivery attempt and verify that the identity remains stable. Then complete a genuinely different action and verify that it receives a different identity.

This second test matters. Reusing one identifier for every purchase may suppress legitimate events rather than inflate counts. A deduplication investigation needs to look for both accidental multiplication and accidental collapsing of distinct actions.

Workflow

One action, two delivery routes

  1. Browser and server

    The same action must share the documented source and event identity.

  2. Retry

    A delivery retry must not create a new business-event ID.

  3. New action

    A separate purchase needs separate identity so it is not suppressed.

Inspect actual payloads, not just the order label in administration.

Inspect source alignment

Browser and server messages that use different Pixel IDs do not describe the same deduplication scope. A copied production configuration, an old tag or a second integration can send one route elsewhere. Check the source values from the transmitted requests rather than relying on a deployment checklist that may be outdated.

Map the routes explicitly: website action, browser sender, server sender and intended source. Include the code or tag owner for each. When two teams own the routes, a shared debugging record prevents each from confirming only that its own request succeeded.

The event-source audit examines this mapping more fully. Here the key question is whether both deliveries of the same action use the same documented identity context.

Do not mistake deduplication for payload correction

OpenAI’s Conversions API documentation describes retaining the first received event for a matching key and ignoring later duplicates. Therefore, sending the same identity again with a corrected amount should not be assumed to update the earlier event. Validate the authoritative payload before delivery and consult supported correction behavior when a real error occurs.

If browser and server disagree about value or currency, deduplication alone is not a complete fix. Establish which system owns the business fact and why the routes differ. A correctly deduplicated event with the wrong value still produces unreliable purchase analysis.

Keep the diagnostic conclusion precise. “The two routes now share an identity in our test” is narrower than “all historical reporting is fixed”. Existing periods and unrelated event paths need their own evidence.

Close the investigation with a repeatable case

Save a test record showing one business action, both delivery identities, expected deduplication scope and the observed result available through supported diagnostics. Include a retry case and a second distinct action. These examples are more useful for future maintenance than a statement that duplicate tracking was fixed.

Monitor the relevant aggregate trend after the change while allowing for reporting timing and normal variation. Do not promise an exact reduction in conversion counts: the earlier problem may have affected only part of the traffic, and attributed reports apply additional rules beyond event receipt.

Use the event-ID contract to prevent recurrence and purchase-value unit checks for payload consistency. The platform rules are documented in OpenAI Conversions API and Measurement Pixel. This is a troubleshooting procedure for the advertiser’s implementation, not a claim that AthillyAds installs or validates either integration automatically.

Sources and scope

Diagnose duplicate delivery of the same business action by comparing source, event identity and retry behavior across existing integrations.

Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.

Your next chapter

See what campaign reporting covers.

Explore reporting in AthillyAds and how it fits alongside your own measurement of enquiries, purchases and other business outcomes.