# The engines: six ways a buyer asks, and what each answer records

Six engines, reached the way a buyer reaches them where that is possible, and every answer stored whole so a later reading can be checked against it.

## The six

| Engine | How it is asked | Plans |
| --- | --- | --- |
| ChatGPT | The ChatGPT app for a prompt with no place, asked from India; the OpenAI API for a prompt pinned to a district, told the city | All paid plans |
| Gemini | The Gemini app for a prompt with no place; the Gemini API with Google Search for a district prompt | All plans; Free asks Gemini only |
| Perplexity | Its API with search, told the country and the district | All paid plans |
| Google AI Overviews | A real Google results page, read for the overview or the map pack | All paid plans |
| Google AI Mode | A real AI Mode answer | All paid plans |
| Claude | Asked with web search, told the city for a district prompt | Growth and Agency |

## Why the app and the API are different instruments

The ChatGPT app searches before it answers a buying question and reads a different index from the developer API. The two name different brands. A session records which one answered, as its own surface, and never pools an app answer with an API answer. The first session on a new instrument is a new baseline.

## What every answer records

- The engine, the surface and the exact model build that answered, where the reply names one.

- Whether it searched, the queries it ran and up to six pages it cited.

- Where it was asked from and in which language.

- The full answer, the names it recommended, every name it mentioned, and the reasons it gave for each recommendation.

- How it treated your brand: recommended, mentioned with a caveat, named among others, cautioned against, or absent, with the sentence that decided it.

- Whether any fact it stated about you disagrees with the facts on your brief.

## Claude, and why it is unticked

Claude costs several times any other answer to ask, so it is on Growth and Agency, unticked by default in the run dialog, and a Claude answer draws five from the month’s answers where any other draws one. Tick it when a client asks for it. A Claude answer that did not search is recorded as a failure, not a loss, because an answer from memory is not the instrument the session is on.

All documentation: https://oftheard.in/docs