What a Lokrix AI scan delivers

01Everything in every scan

Six outputs — concrete, documented, honest.

The list below covers every deliverable in the order Lokrix AI produces them. Where a figure is a prediction or an estimate, it is labelled as such.

01

Every answer engine — scored separately

Lokrix AI runs your prompts through ChatGPT (Bing-grounded), Google AI Overviews (Google-grounded), Google AI Mode (scored separately from AI Overviews — they cite the same URL only ~14% of the time), Perplexity (live Google + Bing search), Gemini (Google-grounded), Grok (X + live web; beta — live via a web-search fallback, not yet counted in headline scores), and Claude (ClaudeBot-crawled). Each engine produces its own presence probability, not a blended average. Microsoft Copilot (Bing-grounded) is tracked as a corroborating grounding signal rather than a separately scored engine.

Honest limitEngine availability and grounding behaviour may change. We model each engine on its currently-observed retrieval behaviour and update when behaviour changes.

02

An auto-generated prompt universe

Lokrix AI crawls your URL, detects your brand and industry, and generates the natural-language questions real buyers ask AI engines — category comparisons, expert recommendations, direct brand queries, and competitor contrasts. The universe is editable: add, remove, and weight prompts by funnel stage. All prompts are tracked over time.

Honest limitThe prompt universe is generated by a language model from your URL and publicly visible content. It is a starting point, not an exhaustive map of every possible buyer question.

03

Monte-Carlo presence probability — with a 95% CI per prompt

Each prompt is sampled K times per engine with browsing/grounding enabled, varying temperature, persona, and locale across runs. The fraction of runs where your site is mentioned or cited is resolved into a calibrated presence probability. A 95% confidence interval is always reported alongside the point estimate. The CI narrows as K increases.

Honest limitThis is a probabilistic measurement, not a guarantee. A 72% presence probability means 72% of sampled runs mentioned you — not that you will appear in 72% of real-world queries. Non-determinism in AI answers is a first-class fact, not a product limitation.

04

Feature-level reasoning — why the score

SHAP-based attribution shows which page and brand signals drove each engine's score: expert quotations, citation density, Bing freshness rank, Google top-20 position, schema markup depth, earned-media corroboration (Reddit, Wikipedia, G2), and content structure. You see which features helped, which hurt, and by how much — per engine, per prompt.

Honest limitFeature attribution is derived from the calibrated scoring model, not from direct access to internal engine ranking signals. The explanations reflect the model's learned patterns, which are calibrated against real engine outputs but are not the engines' own logic.

05

Competitor share-of-voice

Across the same prompt universe and the same engines, Lokrix AI measures the presence probability of your defined competitors. Share-of-voice (SOV) is the comparison of those probabilities — who is cited more, on which engines, and for which prompts. SOV is always shown with the same 95% CI honesty as your own score.

Honest limitCompetitor SOV depends on the same sampling variance as your own score. A narrow gap between you and a competitor may not be statistically significant — the CI band will show when it is.

06

A prioritized optimization plan with simulated uplift

The optimization engine ranks concrete, actionable moves by their predicted visibility uplift: adding expert quotations, embedding statistics, adding cited sources, improving earned-media corroboration, and improving schema markup. Each recommendation is simulated through the scoring model before you spend a sprint on it — so you can see the predicted lift, ranked, before acting.

Honest limitPredicted uplift is a model simulation, not a measured outcome. The Princeton GEO paper found +30–41% citation lift from quotations, statistics, and cited sources, on average. Your results will depend on your specific content, industry, and competitor baseline.