Glossaire · GEO

Prompt

A prompt is the text (or multimodal) instruction that a user or system sends to a language model in order to get a response. In a conversation with ChatGPT, Perplexity, Gemini or Claude, the prompt is the entry point: it defines the task, the context, the tone and the expected format, after which the model generates an answer based on its training data and, where relevant, on web sources retrieved in real time. In GEO, the prompt is central because it determines which queries your content must satisfy in order to be cited. A single need can be expressed through dozens of prompt phrasings, whereas classic SEO focused on short keywords. Understanding the real prompts your audience uses means anticipating the answers AI engines will build and identifying the passages on your pages that are likely to be reused inside those generative responses.

The prompt has become the base unit of search in the AI era. Where people once typed "best GEO agency," they now ask ChatGPT, "Which agency should I choose to get cited by AI when I sell B2B software in France?" That difference changes everything about your visibility strategy.

How a prompt works

When a user sends a prompt to a language model, the model interprets it, identifies the intent and generates a response token by token. AI search engines often add a retrieval step: they turn the prompt into several queries, fetch web sources, then synthesize an answer that cites those sources. This decomposition mechanism is called query fan-out, and it explains why a single prompt can surface very different content.

The quality of a prompt — its precision, its context, its format constraints — directly shapes the relevance of the response. This is just as true for the end user as for a marketing team testing its brand visibility.

Why it matters in GEO

In GEO, you no longer target isolated keywords but families of prompts. The goal: your pages should contain the exact passages an AI will reuse to answer. This requires strong AI citability — direct, factual, structured and self-contained answers.

Key takeaway

A single need is phrased through dozens of prompts. Optimizing for prompts means anticipating every way your audience questions AI, not just the keywords they once typed into Google.

A concrete example

A SaaS brand that wants to be recommended in ChatGPT must map its prospects' prompts: comparisons, use cases, alternatives, pricing. For each family, it produces content that answers in a citable way. This is the core of our approach at LUWIZ.

FAQ

Questions fréquentes

A keyword is a short string of terms typed into a classic engine. A prompt is an often long, conversational and contextualized instruction sent to an AI. A single need can generate hundreds of different prompts, whereas SEO targeted just a few keywords.

Yes. By mapping your audience's real prompts and structuring clear, factual, self-contained answers, you increase the likelihood that your passages will be extracted and cited by generative AI engines.

Pour aller plus loin