AI Models
The AEO (Answer Engine Optimisation) pillar measures what AI models say about your domain when people ask them directly. To do that at scale and reproducibly, Paceghost queries models through their official APIs rather than typing into chat apps by hand.
Why we query models through APIs
Section titled “Why we query models through APIs”Every audit needs to be consistent, repeatable, and comparable over time. APIs give us a controlled way to send the same prompts to many models and record their raw responses. This lets us run the same benchmark for every domain, on demand, and track how results change between audits.
An API call is a simulation of a real user asking an AI assistant a question. It approximates the behaviour a person would see — it is not a byte-for-byte copy of any single user’s chat session.
Why we focus on mainstream models
Section titled “Why we focus on mainstream models”Paceghost queries the widely used, mainstream models — the ones most people actually reach for. Covering the popular models gives a realistic, broad approximation of how a typical user’s AI assistant would answer, rather than optimising for niche or experimental systems that few people use.
Broad coverage matters more than exhaustive coverage: the goal is to reflect the AI landscape your audience is most likely to encounter.
Models are black boxes
Section titled “Models are black boxes”AI models are not deterministic databases, and their public behaviour depends on many factors we cannot fully observe or control. When reading AEO results, keep in mind that a model’s answer can differ depending on:
- API vs app or website. The model served through an API can behave differently from the same model inside its consumer app or website, which may add its own retrieval, system prompts, and post-processing.
- Logged-in vs anonymous. A signed-in user can get different answers from an anonymous one, because accounts carry settings and context.
- User memory and history. Assistants that remember past conversations personalise their answers. Our simulated queries start fresh, without a personal history.
- Temporal drift. Models are updated, retrained, and re-tuned over time. The same prompt can return a different answer next month, which is exactly why point-in-time benchmarking is useful.
Because of these factors, AEO results are best read as a directional signal of your AI visibility — measured consistently the same way each time — rather than an absolute, universal truth.
Which models are queried
Section titled “Which models are queried”The specific models included in an audit depend on your tier. See AEO Endpoints for the current, live list of models by tier.