A prompt is the instruction a user sends to an LLM. In ChatGPT or Claude, the prompt is what you type in the input. Internally the model interprets it as context to generate the answer.
Why it matters in SEO/GEO
Traditional SEO optimises for Google queries (2-6 word strings typically). Prompts to LLMs are different:
- Longer: users write full sentences or instructions.
- More contextual: “I need X for Y specific case”, not just “X”.
- Multi-turn: the conversation refines in steps, each prompt affects the next.
- More varied: the same intent can be expressed in 20 different formulations, each selecting different sources.
Auditing GEO with prompts
Our SEOCOM methodology includes manual prompt audits:
- List 20-30 realistic formulations of prompts for each target topic.
- Run them on ChatGPT, Claude, Perplexity and Gemini.
- Record which sources each cites for each prompt.
- Detect what % of prompts mention the client’s brand.
- Detect which competitors are cited consistently and why.
The audit is slow (does not automate well) but produces insight no dashboard tool offers cleanly yet.
How to optimise
No trick: create content that answers well the prompts users actually write. Not 3-word queries, but complete questions with context. Topical authority + clear structure + freshness are the three that rule.