ChatGPT Ads Insights API: Report Non-Goal Conversions (attributed_events Guide, 2026)

Until this change, a conversion-optimised campaign could only tell you about the one event it was optimising toward. That forced a bad trade. Optimise for the high-volume event the auction can learn from — a lead, a sign-up, an add-to-cart — and you lost visibility of the revenue event your finance team actually cares about. Optimise for the revenue event and you starved the model of volume. Non-goal reporting removes the trade: optimise for what the auction needs, report on what the business needs.

Goal events vs non-goal events

Two rules frame everything else. First, goal reporting applies going forward from attachment — attaching a goal today does not backfill last week. Second, adding the non-goal expansion never changes the goal numbers. OpenAI states that adding or removing include "leaves the reporting clock, windows, and metric values for matching summary rows unchanged." It only adds detail.

The request

Non-goal reporting lives on the dedicated conversion endpoint, not on the general Insights endpoints.

curl -X POST "https://api.ads.openai.com/v1/conversions/insights" \
  -H "Authorization: Bearer ${OPENAI_ADS_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "aggregation_level": "campaign",
    "time_granularity": "none",
    "time_ranges": [
      "{\"type\":\"unix_range\",\"start\":\"1788235200\",\"end\":\"1788840000\"}"
    ],
    "entity_ids": ["cmpn_123"],
    "attribution_time_basis": "ad_event_time",
    "attribution_window_days": 30,
    "view_through_attribution_window_days": 1,
    "include": ["attributed_events"]
  }'

That example is OpenAI's own: one campaign, a single period total, Unix timestamps running from midnight on 1 September to midnight on 8 September in the account's time zone. Note that each time_ranges entry is a JSON-encoded string, not a nested object.

Reading the response

OpenAI's illustrative response shows a campaign with zero goal conversions that nonetheless drove two attributed purchases:

{
  "object": "list",
  "account_currency": "USD",
  "data": [
    {
      "entity_id": "cmpn_123",
      "conversions": 0,
      "click_through_conversions": 0,
      "view_through_conversions": 0,
      "order_created_attributed_sales": "100",
      "order_created_attributed_sales_currency": "USD",
      "attributed_events": [
        {
          "entity_id": "cmpn_123",
          "event_name": "order_created",
          "event_kind": "standard",
          "attributed_event_count": 2,
          "click_through_conversions": 2,
          "view_through_conversions": 0,
          "attributed_event_value_amount": 100,
          "attributed_event_value_count": 2,
          "attributed_event_value_currency": "USD",
          "conversion_event_setting_breakdowns": []
        }
      ]
    }
  ],
  "count": 1
}

Attribution windows and the time basis

The dedicated endpoint applies the windows you pass — independently of which events you select — and defaults to a 30-day click window, a 1-day view window, and ad-event time. Those defaults differ from how many older integrations assumed the endpoint worked. OpenAI's warning is direct: "Existing integrations that relied on click-only totals or conversion-date defaults should review their reporting assumptions. The dedicated endpoint now uses the selected click/view windows and defaults to ad-event time; it does not provide a legacy reporting mode."

The line that matters for this feature: "Conversion-time reporting has limited coverage of non-goal events; use ad-event time when reporting on those events." If your non-goal purchases look suspiciously low, check the time basis first.

To make API numbers match Ads Manager, OpenAI says to use the same date range, time basis, and click and view windows in both. Ads Manager exposes the same controls under Edit columns → Conversion Events: 7, 14 or 30-day click windows, a 0 or 1-day view window, and Conversions (by conv. time). More on reading those in the reporting guide.

Selecting specific events

Omit event_names to return every attributed event. To narrow the expansion, pass exact names:

{
  "include": ["attributed_events"],
  "event_names": ["order_created", "newsletter_signup"]
}

• Names must be recognised for your account. That means configured in conversion event settings — including archived settings — or present in published received-event history. They do not need to be a campaign goal or have activity in the period you request.
• Brand-new custom events are not selectable immediately. They become selectable after their received-event history is published, and publication delays and retention limits apply.
• Invalid selectors return HTTP 400.
• Selecting names filters the nested details only. It does not recalculate goal counts or summary sales. A retained row with no matching events shows attributed_events: [] — which OpenAI notes does not confirm that historical event processing is complete.

Empty rows, large reports and HTTP 413

include_zero_rows defaults to true. Set it to false with the expansion and a row survives only if its goal count, summary sales amount, or a selected event count is non-zero. OpenAI's example: a row with no goal conversions but USD 100 in sales stays in the report, with attributed_events: [], when you select an inactive event.

There is no pagination. The endpoint returns HTTP 413 when a report exceeds 2,000 summary rows, 2,000 event rows, or when an event's goal-setting breakdown exceeds 2,000 entries — with no partial report and no cursor. The fix is to split the request:

• Fewer entity_ids per call.
• Shorter, non-overlapping date ranges.
• Fewer event_names in the expansion.
• time_granularity: "none" instead of daily when you only need totals.

Revenue and ROAS from the general Insights endpoint

For spend and purchase value side by side, use the general Insights endpoints, which return order_created_attributed_sales, its currency, and order_created_roas:

curl -G "https://api.ads.openai.com/v1/ad_account/insights" \
  -H "Authorization: Bearer ${OPENAI_ADS_API_KEY}" \
  --data-urlencode 'aggregation_level=campaign' \
  --data-urlencode 'time_granularity=none' \
  --data-urlencode 'time_ranges[]={"type":"date_range","since":"2026-09-01","until":"2026-09-07","timezone":"America/New_York"}' \
  --data-urlencode 'fields[]=campaign_id' \
  --data-urlencode 'fields[]=campaign_name' \
  --data-urlencode 'fields[]=spend' \
  --data-urlencode 'fields[]=conversions' \
  --data-urlencode 'fields[]=order_created_attributed_sales' \
  --data-urlencode 'fields[]=order_created_attributed_sales_currency' \
  --data-urlencode 'fields[]=order_created_roas'

USD 1,000 in attributed sales against USD 250 in spend returns a ROAS of 4. These metrics only exist if your order_created events carry a value — the most common reason a ROAS column reads empty is a purchase event sent without amount and currency. See the conversion tracking guide.

Four ways to use non-goal reporting

These are our playbooks, built on the documented behaviour above.

1. Optimise for leads, prove revenue

A B2B or high-ticket campaign optimises toward lead_created because purchases are too rare for the model to learn from. Pull order_created as a non-goal event on the same campaigns, at the same windows, and you can finally report cost per customer rather than cost per lead — without moving the goal and resetting learning.

2. Choose the next goal with evidence

Before switching a campaign's optimisation event, pull every attributed event for the last 30 days. The event with enough volume and a stable relationship to revenue is your next goal. Because a campaign's conversion event cannot be changed after creation, this analysis decides what the new campaign should optimise for.

3. Measure the whole funnel from one campaign

Omit event_names and you get contents_viewed, items_added, checkout_started and order_created attributed to the same entity, date and breakdown. That is a per-campaign funnel without stitching three reports together.

4. Keep reporting honest when the goal is not revenue

If a finance team only trusts purchase numbers, report order_created_attributed_sales and order_created_roas for every campaign regardless of goal. The goal event becomes an internal optimisation detail rather than the headline.

A minimal Python example

A sketch that pulls non-goal purchases for a list of campaigns and computes cost per purchase. It assumes you have already fetched spend per campaign from the general Insights endpoint.

import json, os, requests

API = "https://api.ads.openai.com/v1/conversions/insights"
HEADERS = {"Authorization": f"Bearer {os.environ['OPENAI_ADS_API_KEY']}"}

def non_goal_purchases(campaign_ids, start_unix, end_unix):
    body = {
        "aggregation_level": "campaign",
        "time_granularity": "none",
        "time_ranges": [json.dumps({"type": "unix_range",
                                    "start": str(start_unix), "end": str(end_unix)})],
        "entity_ids": campaign_ids,
        "attribution_time_basis": "ad_event_time",   # required for non-goal coverage
        "attribution_window_days": 30,
        "view_through_attribution_window_days": 1,
        "include": ["attributed_events"],
        "event_names": ["order_created"],
    }
    r = requests.post(API, headers=HEADERS, json=body, timeout=60)
    if r.status_code == 413:
        raise RuntimeError("Report too large: split entity_ids or the date range")
    r.raise_for_status()
    out = {}
    for row in r.json()["data"]:
        purchases = sum(e["attributed_event_count"] for e in row.get("attributed_events", []))
        out[row["entity_id"]] = {
            "goal_conversions": row["conversions"],
            "purchases": purchases,
            "sales": row.get("order_created_attributed_sales"),  # may be None, never assume 0
        }
    return out

# cost_per_purchase = spend / purchases, where purchases > 0

Freshness and retention

Two consequences. Pull conversion data no earlier than the day after the period closes, and re-pull the last few days on each run because late events are still being attributed. And if you want more than 30 days of product or hourly history, export it on a schedule — the API will not hold it for you.

Sources and further reading

• OpenAI developer docs — Reporting & Insights, including Report goal and non-goal events
• OpenAI Help Center — Measure Results
• OpenAI Help Center — Conversion-optimized Campaigns
• GPT Ads AI — reporting guide, conversion campaigns, conversion tracking, event quality