# ChatGPT Ads tracking study protocol

Status: Methodology only. No study results have been published.
Advarde · 11 September 2026

## Question

Can a controlled implementation failure be identified from the documented browser observations and diagnostic surfaces? This study does not estimate prevalence across advertisers.

## Environment record

Fixture source / revision:
Browser and version:
Consent implementation:
Pixel helper version:
Advarde version:
Network boundary: mocked / authorized test endpoint / other documented setup
Operator / date:

## Cases

| Case | Expected business actions | Deliberate condition | Expected observation | Observed calls / IDs | Receipt result | Pass / fail / unknown |
| --- | --- | --- | --- | --- | --- | --- |
| Baseline | 1 | Correct installation | One identified completed action | | | |
| Duplicate base | 1 | Direct plus tag-manager install | Duplication behavior documented | | | |
| Consent denied | 1 | No measurement consent | No permitted dispatch | | | |
| Consent withdrawal | 1 after withdrawal | Grant then withdraw | No later permitted dispatch | | | |
| Missing callback | 1 | Success trigger omitted | No completed-action call | | | |
| Wrong source | 1 | Different pixel ID | Request targets different source | | | |
| Confirmation refresh | Still 1 | Reload completed action | Stable identity / no unnecessary resend | | | |
| Server retry | Still 1 | Retry delivery | Same business action ID | | | |
| New action | 2 total | Another order or booking | New action ID | | | |

## Evidence per case

- Exact fixture and deliberate change.
- Successful business record using synthetic data.
- Event name, category, stable ID, and relevant timestamp.
- Browser call observation.
- Network or receipt result, explicitly labeled when mocked.
- Diagnostic sample scope and missing evidence.
- Screenshot captured from the actual interface.

## Publication gate

- [ ] Fixtures and procedure available for inspection.
- [ ] Another operator repeated the cases.
- [ ] Expected and observed results retained, including failures.
- [ ] Mocked responses are not described as platform acceptance.
- [ ] Receipt is not described as attribution.
- [ ] No customer data or credentials in the dataset or images.
- [ ] Claims do not exceed the tested scope.

Reference: https://advarde.com/research/chatgpt-ads-tracking-errors