How Bubbling measures
AI assistants never answer the same way twice. Our methodology turns that variability into a stable, neutral and reproducible measurement.
How Bubbling handles the variability of AI answers
Language models are stochastic: the same question gets different answers. A single test therefore proves nothing. Bubbling treats this variability as something to measure, not to ignore.
At least 10 runs per query
Each query is replayed at least 10 times. Individual answers vary; how often a brand appears across ten runs does not.
Aggregation into visibility rates
Repeated runs are aggregated into rates: visibility rate, recommendation rate, conversational share of voice. That aggregate is what gets tracked over time — not an isolated answer.
Personas and real scenarios
Queries are not generic prompts: they are carried by personas built on real intents and decision triggers, reproducing credible buying scenarios.
Multi-turn conversations
The recommendation is rarely settled in the first answer. Bubbling simulates the full conversation, objections and comparisons included, to identify the inflection points where the recommendation shifts.
Measurement without conversation history
Every run starts from a fresh conversation, with no history. No answer contaminates the next one: each measurement is independent.
Reproducibility
Same personas, same scenarios, same protocol: the measurement can be replayed a month later and compared. That is what makes monthly tracking in the platform actionable.
The measurement chain
From the initial brief to monthly tracking: each step feeds the next, and continuous measurement closes the loop by re-evaluating the prioritized actions.
Covered AI engines
The methodology is engine-agnostic. Bubbling currently covers ChatGPT, Gemini, Perplexity, Claude, Mistral (Le Chat), Copilot, Grok. Adding a new model or market extends coverage without changing the method.
Each query is replayed at least 10 times and results are aggregated into visibility rates. Measurement runs without conversation history and is then tracked monthly in the platform.