The recorded ChatGPT responses are preserved. The context boxes and summaries explain how to read them.
Some code blocks remain unclassified. The labels describe the evidence shown in this session; they do not verify private system internals.
AI Search Optimization
How AI Search Works, From Prompt to Response
Prompt experiments and a recorded ChatGPT session, with the evidence and limitations identified.
This guide examines how a prompt can lead to searches, source checks, and a recommendation. It combines prompt experiments with a recorded ChatGPT conversation.
The goal is to identify what can be observed and what remains an inference.
What the guide contains
The prompt experiments examine how wording and search settings affect recommendations. The section on designing prompts explains how to make those comparisons useful.
The transcript records questions about result cards, tool requests, returned text, and generated JSON. It includes corrections where ChatGPT acknowledges that an earlier example was illustrative or reformatted.
Five worked examples cover a hotel, a blender, CRM software, a local HVAC company, and a healthcare provider. The reflection prompt provides instructions for trying a similar exercise with another question.
What the transcript can establish
The transcript preserves what ChatGPT said and displayed during one session. A model’s explanation of its own process is not an independent record of the software that produced the answer.
Some examples are reported tool requests. Others are transcriptions, conversions, or invented sample data. Read the color key before using an example as evidence.
Three distinctions to keep clear
Result order and ranking scores
ChatGPT reports seeing ordered search results without numeric ranking scores. This does not reveal the search provider’s scoring formula.
Page markup and returned content
The described result cards do not contain the page’s raw structured markup. That does not show that the search provider ignored the markup or that another retrieval method could not expose it.
Returned text and generated JSON
Many JSON objects in the transcript were created to organize information after a tool call. Use the labels to distinguish those objects from the request arguments and reported responses.
How to read this guide explains the page layout, transcript labels, and limits of the evidence.