In SEO you tracked keywords. In generative engine optimization you track prompts, because nobody types "best crm software" into ChatGPT. They ask, "I run a 10-person marketing agency, what CRM handles client leads and project deadlines best?" That shift from two-word queries to full conversations breaks keyword tracking, and it is why you need a prompt map: a structured set of questions covering every way AI models might encounter and describe your brand. This guide shows you how to design one that gives you real monitoring coverage instead of a random list of questions.
A good prompt map is a grid, not a list. It maps buyer intent against how the model talks about your category, so you monitor the whole journey, not just when someone types your brand name.
Key takeaways
- Track prompts, not keywords: AI queries are long, conversational, and intent-rich. - Start from seed entities (brand, products, category, problems) and expand across five intent layers. - Cover the funnel from unbranded discovery to branded comparison, so you catch demand before it knows your name. - A balanced map is mostly unbranded, because that is where you are winning or losing new buyers. - Keep the map to 30 to 100 prompts, tagged by intent and refreshed as the category shifts.
Step 1: Define your seed entities
Start with the core entities tied to your business. These are the anchors everything else expands from.
- Brand name, for example "GEOly." - Product names, for example "GEO Audit," "Agent Team." - Category names, for example "GEO platform," "AI SEO tool." - Key problems you solve, for example "monitor AI visibility," "optimize for ChatGPT."
Write these down plainly. They are your seeds, and every prompt in the map should trace back to one of them.
Step 2: Expand across the five intent layers
AI buyers move through a conversation. Your map needs a band of prompts for each of five intent layers, so you see visibility across the whole journey rather than at a single moment.
- Navigational and definitional: "What is [brand]?", "What does [brand] do?" This is where hallucinations show up first. - Category and unbranded: "best AI visibility tools," "how to track brand mentions in ChatGPT." High volume, and usually where you are weakest. - Comparative: "[brand] vs [competitor]," "alternatives to [competitor]." These decide deals. - Problem and solution: "how do I know if ChatGPT recommends my brand?" Buyers who do not know you yet. - Transactional and long-tail persona: the full scenario query, like the 10-person agency example. Lower volume, very high intent.
Aim for a spread. If 80% of your prompts contain your brand name, you are only measuring people who already found you.
Step 3: Weight toward unbranded discovery
The instinct is to track your own name, but that only shows how you look to people already searching for you. The growth is in unbranded, category-level prompts, where a new buyer meets the category and the model decides who to recommend. Make unbranded prompts the majority of your map. That is where Share of Model is contested and where a competitor can quietly take mindshare.
