Answer engines and data sources
Which engines Signal observes, how live observations are collected, and how sample data is labeled and kept separate.
Updated
Signal records an observation each time it asks an engine a prompt: the full answer text, the sources the engine cited, the model identifier, and the time. Metrics are calculated from these stored observations, and you can open any answer to read it in full.
Engines
| Engine | Vendor | Live collection |
|---|---|---|
| ChatGPT | OpenAI | Responses API with web search |
| Perplexity | Perplexity AI | Sonar API with citations |
| Gemini | Gemini API with Google Search grounding | |
| Claude | Anthropic | Messages API with web search |
| AI Overviews | Google Search | Sample data only |
| Copilot | Microsoft | Sample data only |
Live collection uses each vendor's official API with web search enabled, which closely approximates, but is not identical to, what a signed-in consumer sees in the vendor's app. Consumer answers can vary by account, location, conversation history, and product experiments. Signal records the model identifier on every observation so changes in the underlying model are visible.
Sample data
Every new workspace starts in sample mode. Sample answers are generated from your prompts, brand, and competitors so that charts, tables, and tasks have realistic structure. Sample data is labeled throughout the interface, is never presented as an observation of a real engine, and is excluded from live metrics. You can switch to live collection once your workspace has live engines available; sample data remains viewable until you clear it.
Variability
Answer engines are not deterministic. The same prompt can produce different answers minutes apart. Signal treats each answer as one sample from a distribution and reports rates with confidence intervals rather than single-answer "rankings."