How to Structure a Project in LLM Monitor for Actionable Results

Too often, users create a project by piling up random questions with no common thread. The result: scattered data, difficult to interpret, that doesn’t support decision-making. The right approach is different. It consists of structuring each project around a specific angle, then methodically building the personas, themes, and questions that follow from it.

One project = one analysis angle, not a catch-all

The central concept to understand is the analysis angle — or focus. A project in LLM Monitor is not meant to cover your entire brand all at once. It targets a specific aspect you want to explore in the responses of artificial intelligences.

This angle can take very different forms depending on your context:

• A key moment or seasonal trend — Summer holidays are approaching and you want to know whether AIs recommend your brand when users ask questions related to that period. Your themes and questions will then be oriented around your customers’ seasonal needs.
• A product range or area of expertise — You’re a toy brand and you want to measure your visibility on wooden toys. Not just the products you sell, but the entire topic: how AIs talk about this category, which brands they recommend, how they position your offering.
• A strategic challenge — A new market to conquer, a brand repositioning, competitive intelligence focused on a specific segment.

The idea is simple: the more precise your angle, the more actionable your results. A project that tries to cover everything produces noise. A well-targeted project produces actionable insights. It’s the same logic as for organic search: you don’t work on “SEO in general”, you work on a coherent cluster of queries. GEO works on the same principle.

Step 1: Identify your brand and its positioning

Every project starts with identifying your brand. Name, website, target country, language. This information may seem basic, but it is fundamental: it determines the context in which LLM Monitor will query the AIs.

From this data, the platform automatically generates:

• a description of your brand as AIs might perceive it;
• an analysis of your market and your competitive positioning.

These elements can be edited at any time. Take the time to review and refine them — they serve as the foundation for all subsequent analysis. If the generated description doesn’t accurately reflect your actual positioning, that’s actually an interesting signal: it means the information available online about your brand may not be conveying the right message to AI models.

Step 2: Define your personas — the strategic foundation

This is the most important step, and often the most underestimated. Personas represent the typical user profiles who ask AIs questions about your topics.

Why is this so crucial? Because a consumer and a professional don’t phrase the same question. A beginner discovering a topic and an expert comparing solutions don’t get the same answers from ChatGPT, Gemini, or Claude. AIs adapt their responses to the perceived user profile — and therefore the recommended brands change depending on the persona.

In practice, here’s the difference it makes:

Your brand can be highly visible for one type of persona and completely absent for another. This is why defining your personas is not a configuration exercise, but a strategic reflection.

Step 3: Choose your analysis themes

Once your personas are defined, the themes follow naturally. A theme represents a specific visibility angle that you want to measure in AI responses.

Themes emerge from the intersection of your analysis angle (the project) and your personas. Here are a few examples to illustrate:

• Summer seasonality focus → Possible themes: pre-holiday preparation, safety, equipment, seasonal maintenance.
• Product range focus → Possible themes: category comparisons, selection criteria, alternatives, specific use cases.
• New market focus → Possible themes: existing players, unmet needs, perception of your brand in that segment.

The challenge is to choose themes that are specific enough to produce readable results, but broad enough to cover the different ways users approach the topic with an AI. Each theme will then become a lens in your dashboard — you’ll be able to analyze your visibility theme by theme, identify your strengths and blind spots.

Step 4: Formulate and group your questions

Questions are the operational core of your analysis. They are the ones that will be sent to the various AI models — ChatGPT, Gemini, Claude, Perplexity — to observe how they respond and which brands they recommend.

LLM Monitor automatically generates questions from the intersection of your themes and personas. Each question corresponds to a query that a real user might ask a conversational AI. You can:

• select the most relevant questions from those suggested;
• add your own questions — ones you hear from your customers, ones you spot in your search data;
• remove those that don’t match your analysis angle.

The crucial point: your questions must form a coherent set within each theme. No isolated or off-topic questions — a group of queries that covers the theme comprehensively. It’s this coherence that will give you a reliable, actionable reading of your positioning in AIs.

The multi-angle approach: creating multiple projects

The onboarding guides you through creating your first project — a first angle to get started and understand how the platform works. But the real power of LLM Monitor emerges when you adopt a multi-angle approach.

In practice, you can create as many projects as needed, each targeting a different angle:

This approach allows you to progressively build a complete map of your visibility in AIs, without ever sacrificing the precision of each analysis. Each project remains readable and actionable, and together they give you a strategic 360° view.

Don’t try to cover everything in a single project. Start with the angle that has the most business impact for you, get your first results, then gradually expand with new projects.

The complete workflow, step by step

To sum up, here’s the workflow to keep in mind for every project you create:

It’s this structure — angle, brand, personas, themes, questions — that transforms LLM Monitor from a simple tracking tool into a true strategic lever. Without it, you get data. With it, you get decisions.

Common mistakes to avoid

After guiding many brands through their onboarding with the platform, here are the most common mistakes:

• Keeping the default personas without refining them. The pre-configured personas from onboarding are a starting point, not an end. Take the time to adapt them to your reality.
• Multiplying themes in a single project. Better to have 3 well-targeted themes than 10 vague ones. If you have too many themes, your angle probably isn’t precise enough — consider creating a second project.
• Asking questions that are too generic. “What is the best product?” will yield less relevant results than “Which [product] to choose for [specific context]?” Contextualize your questions the way your real customers would.
• Trying to analyze everything at once. Start with a priority angle, analyze the results, then create new projects. An iterative approach is always more effective than an exhaustive attempt right from the start.
• Ignoring the time dimension. AIs are constantly evolving — a project created to analyze a specific seasonal trend has a limited shelf life. Remember to renew your analysis angles according to your business calendar.

Guillaume