N-Sampling (N=5)
The per-query, per-engine sample size in a SOV run — each query asked N=5 times.
What is N-sampling?
N-sampling is the per-query, per-engine sample size in a SOV run, default N=5: each query is asked five times on each engine per run, so the platform computes a mean and a variance rather than one noisy reading.
N-sampling exists because AI answers are non-deterministic — ask once and you measure luck, not visibility.
The same prompt sent to the same engine twice can return different brands, different ordering and different citations. A single sample is therefore an unreliable measurement. N-sampling sends each query N=5 times to each engine in a run and aggregates the results, turning a stochastic answer into a stable statistic.
Sampling unlocks two things a single shot cannot: a mean, which smooths out the randomness, and a variance, which quantifies how unstable the brand’s presence is. That variance feeds L4 Weak signals of the SOV pyramid and underpins the A/B/C classification of how reliably a mention recurs.
N=5 is the default trade-off between statistical stability and the cost of probing four engines across many queries. It is documented in migration 051 and the SOV methodology v2 reflection, and is tuned per platform where an engine’s quota or behaviour demands it.