A configured search tool does not, by itself, establish that a particular answer used web search. Record what was available, what executed and what the answer cited as separate facts. This is an API analysis task; it does not reproduce a signed-in ChatGPT conversation.
Preserve the execution evidence
OpenAI documents web search tool-call output and source metadata. For a diagnostic run, retain the response item types and search-call status alongside the generated text. Do not infer execution merely from an answer that sounds current.
Use three fields in your review: search configured, search execution observed and citations returned. Each can be true, false or unknown when the available record is incomplete.
Read the response in layers
- Inspect the submitted tool configuration.
- Find any returned web search call and its status.
- Inspect the generated answer and citation annotations.
- Associate sources with the relevant call or answer segment.
- Record missing evidence without filling it from assumptions.
Illustrative example: an integration enables search, but the saved record contains only answer text. You can establish that search was configured if the request was retained. You cannot establish execution from that record alone.
A search call can retrieve information that is not cited in the final prose. Conversely, a URL typed in answer text should not automatically be treated as a structured provider citation.
Diagnose the measurement difference
If two runs differ in search execution, inspect that difference before interpreting a change in mentions. A brand's website may be unchanged while the observation's retrieval context differs.
Do not strip away failed or incomplete tool calls when preparing an evidence appendix. They may explain why an answer lacks sources. Keep transport status, tool status and answer validity distinct.
The provider schema is the authority for the fields your integration uses. Recheck the current documentation when updating an adapter; do not copy field assumptions from a different API or older implementation.
Keep the product claim bounded
This procedure describes evidence to retain in an API diagnostic. It does not claim that every raw provider field is exposed as a report control or export in MonitorMyGEO.
The ChatGPT discovery guide explains crawler eligibility separately. Next, review MonitorMyGEO's measured surfaces, then inspect one observation with its available execution evidence before describing it as a web-grounded answer.
Fact-checked: 17 October 2026. Examples are illustrative; no customer results are reported.