Scans overview

A scan is the unit of measurement in Lokrix. It takes one property and its prompt universe, runs each prompt through the answer engines multiple times, detects where your brand is surfaced, and turns the results into a calibrated Lokrix AI Score with a 95% confidence interval — never a single-run yes/no.

The scan pipeline

Every scan moves through the same stages. Each stage is instrumented and metered, so you always know what a run cost and how confident its result is.

The Lokrix AI scan pipelineA six-stage flow: a URL is analyzed to detect the brand, which generates a prompt universe; each prompt runs across seven answer engines K times; the sampled presence is aggregated into a Monte-Carlo probability with a confidence interval and rolled up into the Lokrix AI Score.URLexample.comDetect brandindustry + NERPrompt universebuyer questionsEngines × K7 enginesMonte-Carlopresence + CILokrix Score0–100K samples
  • Readiness pre-check — the property is validated (URL reachable, brand detected, prompt universe present, credits available) before any engine call. See Readiness pre-check.
  • Engine runs — each prompt is dispatched to the selected engines. Runs are either live (real engine responses) or predictive (the calibrated model). See Live vs predictive.
  • Sampling — because engine answers are non-deterministic, each prompt is run K times across temperature, persona and locale to build a Monte-Carlo presence probability. See Sampling & confidence intervals.
  • Detection & scoring — mentions and citations are detected (URL match, brand NER, semantic match) and aggregated into per-engine presence, share-of-voice and the composite score.

Which engines run

Five engines run live by default — ChatGPT, Claude, Perplexity, Grok and Gemini. Google AI Overviews and AI Mode require a SERP provider key (Serper/SerpAPI) to observe, so they are off by default until you connect one. Read how each engine grounds its answers in The 7 engines.

Cost and honesty

Live runs are the dominant cost and are metered against your workspace credit balance — cost is multiplicative across prompts, engines, K samples and cadence. Lokrix defaults to the predictive model and escalates to live runs deliberately. See Credits & cost and Data honesty for how every result is labelled live, cached or mock (sample).