Stage 2 continued: How do you check whether ChatGPT recommends your company?

Test a recorded experience, not an ambiguous engine name. For example, “ChatGPT with Search selected in the consumer interface” is a clearer specification than “OpenAI”.

Select surfaces using customer evidence and the decision in your charter. A possible starting panel is ChatGPT Search, Google AI Mode and Perplexity. If Google AI Overviews or Microsoft Copilot matter more to your buyers, substitute them—or budget explicitly for adding them.

What should you record before collecting answers?

Create a collection specification containing:

• Product and interface: consumer website, app, API or supplier-collected result.
• Model label where visible; otherwise record not disclosed.
• Account tier and signed-in state.
• Search mode and any evidence that web search occurred.
• Country, language, device and location settings you can actually establish.
• Memory, personalisation, custom instructions, connected apps and uploaded-file state.
• Fresh conversation versus a predefined multi-turn sequence.
• Date, time and time zone.
• Prompt version and repetition number.
• How answers, links and screenshots will be retained.

OpenAI documents that memory can affect search-query rewriting. Its current Temporary Chat guidance also distinguishes personalised and unpersonalised modes. Therefore, do not assume that merely opening a new or temporary conversation removes all personalisation; inspect and record the settings available in the account you use. (help.openai.com)

What is the manual collection procedure?

1. Confirm the approved account and settings.
2. Open a fresh conversation or search session.
3. Submit the exact prompt without adding your preferred answer.
4. Wait for completion and open the source panel where necessary.
5. Save the answer, citation URLs and a screenshot showing the relevant interface.
6. Record whether the answer included web sources, an AI result appeared, or collection failed.
7. Score the result using the rules below.
8. Repeat according to the schedule, not until you obtain a flattering answer.
9. Retain the raw result separately from your interpretation.

Randomise or rotate prompt order between collection days as a design safeguard. Do not put every competitor query into the same continuing conversation.

Can you automate the collection?

Only through an approved access route—and label what you collected. OpenAI’s consumer Terms of Use prohibit automatic or programmatic extraction of data or output. Do not turn this guide into an unreviewed browser-scraping project. (openai.com)

A monitoring supplier should explain its collection route and contractual basis. An API-based experiment should retain the endpoint, model, search configuration and other relevant settings. This guide treats API observations as a separate dataset unless comparability with the target consumer experience has been demonstrated.

For a DIY API build, budget for configuration, scheduling, retries, response storage, citation extraction, scoring QA and maintenance—not merely generation charges. If nobody owns those tasks, start manually or evaluate a monitoring product.