Our methodology

For every audit we ask a series of buyer questions (“who is the best X in Y?”, “who would you recommend for…?”) in your language and market - to ChatGPT and Gemini on the free audit, with Perplexity added on subscription. Nothing is estimated: every analyzed answer comes from a real query - yours, or the same question asked by an audit within the past 48 hours; we never keep an answer longer than that. Paid monitoring always asks fresh.

We add a recognition probe (“what do you know about this business?”) to distinguish two very different situations: an AI that doesn't know you, and an AI that knows you but doesn't recommend you. This question never counts toward the score.

Your presence: whether your brand is cited, and at which position in the recommendation list.

Your competitors: which businesses get recommended instead of you, how often, at which position and by which AI assistants - that's your share of voice.

The tone: sentences where your brand appears are classified positive, neutral or negative. This analysis relies on the answer's vocabulary: it is indicative, and labeled as such.

The sources: Perplexity exposes the web pages it consults to answer - we aggregate them to show you precisely where to get listed. Gemini, when it searches, reports its searches but its API does not provide a usable list of the pages consulted; ChatGPT exposes nothing. We only display what is actually exposed - we do not invent sources.

Not all AI assistants answer the same way. Perplexity reads the web on every question. Gemini runs a Google search at answer time whenever our monthly search quota allows it - free audits included. ChatGPT, queried via API, answers from memory: it does not consult the web, while the consumer app may.

That difference changes what “not cited” means. An engine that searched and did not find you measures a real absence; a memory-based engine that does not cite you is only saying “I don't know them”. Since 9 August 2026, whenever at least one assistant consulted the web, the score counts only those assistants - and every report states which engines searched and which answered from memory.

We query the models' official APIs. They approximate - without exactly replicating - the consumer apps' answers: those apps may add geolocation or conversation history, which the APIs do not have. As for web search, every report states precisely which engines searched and which answered from memory - we do not leave that difference in the dark.

AI answers vary between conversations. An audit is a snapshot, not a to-the-decimal truth. Monitoring repeats the measurement at a fixed interval: each report gives that day's measurement and its gap to the previous one, and it is the series of points that shows a trend - we do not smooth the number, we give you the series.

Our questions are typed templates for your sector and market. They represent what real buyers ask, but can't cover every possible phrasing.