Prompts and personas: building reliable visibility tracking

A tracked prompt is a template question, phrased the way a real user would ask a generative AI, that Vurto regularly sends to ChatGPT, Claude, Gemini and Perplexity to measure how your brand, your products and your categories show up in the answers. The quality of your tracking depends directly on the quality and diversity of these prompts.

The three sources of your prompts

Your prompts come from three sources. System prompts are suggested by Vurto based on common question patterns in your sector — comparisons, recommendations, best value for money. Onboarding prompts are the ones generated automatically during the initial analysis of your site, based on the detected industry, brands, categories and keywords. Custom prompts are the ones you write yourself, to cover a question specific to your business that no system prompt addresses — a particular warranty, a specific use case, a comparison naming a competitor. Solid tracking combines all three: system and onboarding prompts give a broad comparison base consistent with your sector, custom prompts cover your specific blind spots.

Dynamic variables

To avoid manually duplicating the same prompt for every tracked brand, product or keyword, Vurto uses variables: {{marque}}, {{mot_cle}}, {{categorie}}, {{nom_du_produit}}. A prompt written once with these variables automatically declines across all matching tracked elements, with no rewriting needed. This is what keeps tracking consistent even with a large catalog, instead of managing dozens of manual variants of the same prompt.

Organizing by intent

Vurto classifies prompts by intent: brand, category, keywords, products, price, competition, recommendations. This isn't just filing — it shapes what you can later read in your dashboards. An imbalance between intents skews the reading: if nearly all your prompts are about the brand and none about price or competition, your overall visibility score will reflect your brand awareness well but say nothing about where you stand in comparisons, even though that's often where purchase decisions are made. Regularly check that all seven intents are represented in reasonable proportion relative to your customers' actual questions.

The role of personas

A persona is a fictional user profile — for example a professional buyer focused on technical specifications, or an individual primarily price-sensitive — used to simulate different audiences facing the same topic. This diversity of viewpoints ties back to the goal of covering the widest possible AI Share of Voice: being visible to a single type of audience isn't enough if your real customers span several different profiles. The same prompt asked from two different personas' viewpoints can produce very different AI answers, with the competitors and sources shifting depending on the angle taken. Without personas, tracking only captures a single average viewpoint, which can hide very unequal positions depending on the actual target audience.

Example: a prompt template and its persona variations

A prompt template like "What's the best {{categorie}} brand for {{mot_cle}}?" automatically declines across every tracked category and keyword. Paired with a "professional buyer" persona, this same prompt can lean toward reliability and technical compliance; paired with a "first-time individual buyer" persona, it leans more toward ease of use and price. Across both personas, {{categorie}} and {{mot_cle}} stay identical, but the AI's answers — and therefore the competitors and sources that surface — can diverge sharply. That divergence is what makes the prompt × persona cross-analysis useful: it reveals competitive positions invisible in an average reading.

How many prompts and personas to track

There's no universal ideal number, but a few practical benchmarks help. Start by covering each intent with at least one system or onboarding prompt, add custom prompts only for topics where your customers actually ask specific questions, and limit active personas to those matching real, distinct customer segments for your business — multiplying personas without a real difference in buying behavior dilutes the reading without adding information. The number of tracked prompts is also bound by your plan (500 tracked prompts on Standard, up to unlimited prompts on Business), which should factor into your decisions too.

Managing your prompts over time

As your prompt library grows, managing prompts one by one becomes limiting. Vurto lets you export your full set of prompts as CSV to review offline, have another team validate them, or prepare a new wave in bulk, then re-import them. This is especially useful for periodic review: rather than editing each prompt one at a time in the interface, export them, flag the ones that have become outdated or redundant, adjust under-represented intents, and re-import the updated list.

Avoiding common writing pitfalls

A poorly worded system or custom prompt skews the reading before the first analysis even runs. A prompt too close to an SEO keyword ("best cheap {{categorie}} brand") doesn't match how a user actually talks to a conversational AI, and produces less representative answers than a real question would. A prompt that's too long, or that stacks several questions at once (brand, price and delivery in the same sentence), dilutes the answer and makes the result harder to interpret. Best practice is to keep each prompt centered on a single clear intent, phrased as a real question — creating several prompts if needed, rather than one overloaded one.

Evolving your tracking instead of freezing it

A prompt set is never final. A prompt that no longer produces useful information (because the AI answer has become stable and predictable) can be replaced with a prompt on an intent that's still under-covered. Conversely, a topic that becomes strategic — a new competitor, a new product line — deserves dedicated custom prompts. This review work naturally connects with reading your visibility dashboards and extends into analyzing your competitors in AI answers to understand why a given prompt tilts in their favor: the results you get are what tell you where your tracking needs more depth.