Core concepts

Lokrix answers one question: when a buyer asks an AI engine, how likely is your brand to be surfaced? Before you run a scan, it helps to share the same vocabulary — what an answer engine is, what “surfaced” means precisely, why every result is a probability with a confidence interval rather than a yes/no, and when Lokrix pays for a live run versus predicts one.

Answer engines

An answer engine is an AI system that responds to a natural-language question with a synthesised answer, often citing or linking sources. Lokrix tracks seven of them, and treats each as distinct because each grounds its answer on a different index:

  • ChatGPT retrieves via the Bing index — roughly 87% of its citations match Bing’s top-10.
  • Google AI Overviews and AI Mode both synthesise from Google’s own index, yet cite the same URL only ~14% of the time — so Lokrix scores them as two separate engines.
  • Perplexity runs live search across Google and Bing, and over-weights Reddit (~47% of top citations) and freshness.
  • Gemini grounds on Google Search; Copilot grounds on Bing.
  • Claude and Grok (xAI) both use live web search.
The answer engines Lokrix AI measuresA horizontal strip of engine cards — ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, Grok and Claude — each marked as a measured surface, above a caption noting each is sampled K times.ChatGPTAI OverviewsAI ModePerplexityGeminiGrokClaude7 answer engines · each sampled K times

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 opt-in. Read more in Answer engines.

Surfaced probability

To be surfaced means the engine cited, linked, recommended or named your brand in its answer. Because engine answers are non-deterministic — the same prompt can yield different answers minutes apart — a single run tells you almost nothing. Lokrix therefore reports a presence probability: the estimated share of answers, for a given prompt and engine, in which you are surfaced.

Statistical honesty and confidence intervals

Non-determinism is treated as a first-class concept, not a bug. Each prompt is run K times across varied temperature, persona and locale. Those repeated observations produce a Monte-Carlo presence probability together with a 95% confidence interval — the range within which the true probability is likely to sit given how many samples were drawn.

The Lokrix AI Score is a calibrated 0–100 composite, also reported with a 95% CI. Per engine you additionally get a component score, presence, share-of-voice (SoV) and the underlying citations. Lokrix never presents a single-run result as a probability, and always discloses the locale, persona and time context behind a number. See the Lokrix AI Score for the full breakdown.

Live vs. predictive

Lokrix runs two cooperating subsystems. A live answer-engine simulator queries the real engines with browsing/grounding enabled and detects mentions and citations of your brand. A predictive scoring model — a calibrated ML model — estimates citation probability from page, brand and web features without paying for a live run, and powers cheap what-if simulations for the optimization engine.

Live runs are the dominant cost: it is multiplicative — prompts × engines × K samples × cadence — and each engine response is metered against your workspace credit balance (1 credit ≈ 1 engine response). Lokrix defaults to the predictive model and escalates to live runs deliberately. Every result is labelled by provenance — live, cached, or mock (sample) — and mock data is never shown as live. Learn more in Live vs. predictive scans and Credits.

Next steps

Ready to try it? The Quickstart walks you from sign-in to your first score in about five minutes. To understand how projects and teams are organised, see the workspace model.