OA9 OppAlerts from Ben Wills
How AI Search Works: From Prompt to Response A guide for SEOs, AI search marketers, and marketing teams
Work with me
Work in progress Transcript and examples under review.

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.

Read the announcement →

How to use this guide

How to Read This Guide

Page layout, transcript labels, and the difference between commentary and recorded responses.

This guide includes editorial explanations and a recorded ChatGPT conversation. The page layout identifies which text belongs to each.

Page layout

Transcript pages begin with a context box and, where useful, a summary. A darker box contains the displayed prompt. A lighter box contains ChatGPT’s recorded response.

The displayed prompts are edited for readability. Their dictated originals remain in the page source. The response text is preserved, including contradictions, formatting problems, and corrections.

Read the labels before using an example

The color key distinguishes reported tool values, invented examples, documented structures, and conceptual representations. Some blocks are still unclassified.

A model can produce plausible JSON without copying a tool response. A repeated assertion that a block is real does not independently verify it. Use original tool logs and published documentation where available.

Suggested reading order

Scope of the evidence

This is a record of one session. It does not document every ChatGPT configuration or every AI search product. Statements about unexposed fields describe what the model reported seeing, not what the underlying search system stores.

Using the guide in a project

Use the prompt experiments to inspect differences in a recorded sample. Use the conversation to follow what ChatGPT said, what it corrected, and what it could not establish. Use provider documentation when implementing a supported API.

For example, OpenAI’s web-search API guide documents public source records and citation annotations. That documentation concerns the API. It does not turn the conversation’s reconstructed JSON into an API contract.

The optimization guide’s completed audit example shows how to use prompts, saved answers, source checks, and factual corrections in a marketing project.