System prompt is a set of instructions given to an LLM before any user interaction — invisible to the end user but foundational to how the model behaves. Every major AI engine (ChatGPT, Claude, Gemini, Perplexity, Copilot) uses a system prompt to configure the model’s persona, constraints, response style, and guidelines for handling specific topics. System prompts are one reason the same underlying model can behave very differently across different products and deployments.
What system prompts typically contain
- Persona and tone — “You are a helpful assistant. Respond concisely and accurately.”
- Topic constraints — “Do not provide financial advice. Do not discuss competitor products.”
- Citation behavior — “Always cite sources. Prefer peer-reviewed publications.”
- Refusal rules — categories of content the engine will decline to generate
- Brand treatment guidelines — some engines have policies about how they handle brand recommendations
Why system prompts matter for brand visibility
You cannot read or directly influence an engine’s system prompt. But system prompts shape:
Citation preferences: Some engines instruct the model to prefer official sources, government sites, or established publishers. If your domain category isn’t in the preferred tier, you may be cited less even if your content is highly relevant.
Recommendation neutrality: Some engines instruct the model not to make specific product recommendations or to present all options equally. Others are more opinionated. Knowing an engine’s disposition helps set realistic expectations for mention rates.
Competitor blocks: Enterprise deployments sometimes configure system prompts to exclude competitor mentions entirely. This is a B2B consideration for teams monitoring how their brand appears in third-party AI tools used by customers.
Format instructions: System prompts often specify response length and format. A system prompt requiring bullet-point answers creates a different brand mention context than one requiring prose — affecting which mention position types appear.
System prompts across different engines
| Engine | Known system prompt characteristics |
|---|---|
| ChatGPT | Safety-focused, generally neutral on brand recommendations |
| Perplexity | Citation-heavy by design; retrieval-first approach |
| Claude | Avoids definitive product recommendations in some contexts; high citation transparency |
| Copilot | Microsoft ecosystem-aware; Bing search-grounded |
| Gemini | Google ecosystem integration; may prefer Google-adjacent sources |
Why the same query behaves differently across engines
If your brand appears prominently in Perplexity but not ChatGPT for the same query, system prompt differences are a likely contributing factor — alongside training data and retrieval differences. This is expected and not a bug in your monitoring data.
LLM Metrix surfaces per-engine breakdown precisely because of this: a single average across engines would mask engine-specific patterns driven in part by system prompt behavior.