Frequently asked questions

Short, honest answers to the questions that come up most often — what Lokrix measures, how the engines are covered, what live runs cost, and how to move your score. For step-by-step fixes, see troubleshooting.

The basics

Lokrix answers one question across seven engines: how likely is it that an AI answer surfaces your brand? The map below shows which engines run live by default and which need a search-provider key.

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

Questions & answers

What does Lokrix actually measure?
The calibrated probability that a major AI answer engine cites, links, recommends or names your site for the questions your buyers really ask. Each result is a 0–100 Lokrix AI Score with a 95% confidence interval, not a single yes/no.
Which answer engines are covered?
Seven: ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, Grok (xAI) and Claude. Five run live by default (ChatGPT, Claude, Perplexity, Grok, Gemini); AI Overviews and AI Mode need a search-provider (SERP) key to observe.
Why are AI Overviews and AI Mode counted as two separate engines?
They both synthesise from Google’s own index but cite the same URL only about 14% of the time, so their behaviour and your visibility in each differ enough to score them independently.
Why do I sometimes get a different answer than Lokrix reports?
Answer engines are non-deterministic. Lokrix runs each prompt K times across temperature, persona and locale and reports a Monte-Carlo presence probability with a confidence interval — a distribution, never one lucky or unlucky run.
Is the data live or predicted?
Both, and every value is labelled. A live run queries the real engines and costs credits; the predictive model estimates visibility cheaply and powers what-if simulation. Provenance is always shown as live, cached or mock (sample) — mock is never presented as live.
What is a credit and how many do I need?
One credit is roughly one engine response. Cost is multiplicative — prompts × engines × K samples × cadence — so a broad, frequent live scan consumes far more than a narrow one. Lokrix defaults to the predictive model and escalates to live runs deliberately.
How do I actually improve my score?
The strongest content levers from the Princeton GEO research are adding quotations, statistics and cited sources (roughly +30–41% visibility; citations lifted low-ranked pages ~115%) plus earned-media corroboration on Reddit, Wikipedia and G2. Lokrix turns these into prioritised, simulated recommendations.
Does llms.txt help?
There is no measured effect on AI visibility. Lokrix keeps it optional and will never sell it to you as a lever.
Can I track more than one website?
Yes. A workspace holds many properties (each property is one site). Use the property switcher to change the active one; entering a new URL in onboarding creates a property and kicks off its first analysis and scan.
How is my data kept separate from other tenants?
Everything is scoped to Organization → Workspace → Property, and every query is tenant-isolated. No tenant’s data crosses into another’s.
Is there an API or MCP server?
Yes. The full engine is exposed through a REST API and an integrated MCP server, both authenticated with scoped API keys.
What if a scan won’t run or the score won’t update?
Check the readiness pre-check, your credit balance, and whether the engine you expect needs a SERP key (AI Overviews and AI Mode do). The troubleshooting guide walks through each case.

Learn more