What a prompt means
A prompt is the input a user gives an AI engine — the question, request, or instruction the model responds to. "Best CRM for a small nonprofit?", "compare Notion and Asana for a design team", "is Acme Analytics worth the price?" are all prompts. Technically the term covers everything fed to the model at inference time, including system instructions and retrieved context; in everyday and marketing usage it means the user's message.
In GEO, the word carries a second, more specific meaning: a prompt is the unit of measurement. AI visibility can't be observed directly — it's revealed one answer at a time, and every answer starts from a prompt. The prompts you choose to track define what "visibility" means for your brand.
Prompts are the new keywords
The cleanest way to understand prompts in GEO is by analogy to SEO keywords — with the differences noted honestly:
| Keywords (SEO) | Prompts (GEO) | |
|---|---|---|
| Form | Short phrases ("crm small business") | Natural-language questions |
| Result | Ranked links, same for everyone | A generated answer, varying per run |
| Success | Position on the page | Being named in the answer |
| Volume data | Public search-volume tools | None — panels are sampled, not looked up |
Like keywords, prompts are finite, repeatable queries you can measure against; unlike keywords, the "results page" is probabilistic, so a single run proves little and repeated sampling is essential. There's a further wrinkle: engines expand one user prompt into multiple background searches (query fan-out), so the surface being contested is wider than the prompt text itself. And in conversational search, prompts chain — a brand can enter at the first question and be eliminated by a follow-up.
What a good tracked-prompt panel looks like
A useful panel mirrors how real buyers actually ask, across intent types:
- Category prompts — "best X for Y" and "top alternatives to Z": the unbranded questions where recommendations are won or lost.
- Comparison prompts — "BrandA vs BrandB": where framing and caveats surface.
- Branded prompts — "what is BrandName?", "is BrandName legit?": where accuracy problems and hallucinations show up.
- Problem prompts — the pain-point phrasings ("how do I stop churn in a small SaaS") that buyers use before they know the category term.
Panel design choices — how prompts are generated, grouped, and displayed per engine — are covered in Brandflare's prompts and the prompt matrix.
From prompt to metric
Once a panel is fixed, measurement is mechanical: run every prompt against every engine on a schedule (prompt tracking), record which brands each answer names, and compute mention rate, share of voice, and trend. The prompt is where all of it starts — the question your buyer asks, turned into an instrument you can read.