# One Prompt Across Four AI Services: What the Requests Tell Us

> Canonical source: [One Prompt Across Four AI Services: What the Requests Tell Us](https://www.answer.cloud/pages/1-prompt-4-llms-48-bots-hosted/)

Treat a small retrieval experiment as an observation, separating request counts from bot identity and citation evidence.

A past experiment compared one question across four AI services. Its original headline mentioned “48 bots”; that label should not be treated as a verified count of distinct bots. Requests, unique clients, and verified provider agents are different measures.

## What can be concluded

The observation suggests that retrieval behavior differed between the tested sessions. It does not establish a general prompt-to-request ratio or explain the providers’ motives. Cost is one possible hypothesis, not a proven cause.

An answer can cite a source without a new identifiable request reaching your origin during the test. Search indexes, caches, other retrieval paths, and classification limits can separate citations from your immediate logs.

## Make future tests reproducible

Record the date, product/model, search settings, exact prompt, observed time window, requests, identity checks, and answer citations. Repeat the test before generalizing. Keep unique clients, fetches, and citations as separate counts.

Request logs directly measure access; they do not provide a complete view of how your brand is represented inside models. Read the [metrics guide](https://www.answer.cloud/pages/llm-visibility-metrics-the-kpi-playbook-for-seo-and-marketing-leaders-hosted/).