Coverage & freshness

A score is only as trustworthy as the data behind it. Coverage tells you how much of your prompt universe × engine grid was actually sampled, and freshness tells you how recently. Both qualify the Lokrix AI Score and its confidence interval — a high score built on thin, stale coverage is a weaker claim than the same score built on broad, live coverage.

Coverage: populated prompt × engine cellsA grid of prompts (rows) by answer engines (columns). Filled cells are populated and tinted by data freshness — green under 24 hours, amber under 7 days, red older — while unpopulated cells are shown as empty outlines.GPTAIOAIMPPXGEMGRKCLAPrompt 1Prompt 2Prompt 3Prompt 4Prompt 5Prompt 6< 24h< 7dolder72% populated

What coverage means

Coverage is the proportion of your prompt × engine grid that has been sampled, and how many samples each cell holds. Cells can be covered live, covered by the predictive model, or not yet covered at all. Two things commonly reduce coverage: engines that require a key you have not added — Google AI Overviews and AI Mode need a SERP provider (Serper / SerpAPI), so only five engines run live by default — and prompts you have chosen not to run live to conserve credits. Lower coverage widens the confidence interval on every score it feeds.

Why freshness matters

Answer engines change what they surface as the web and their indexes change, and several engines explicitly over-weight recency — Perplexity, for instance, favours freshness heavily. A cell sampled weeks ago may no longer reflect reality, so Lokrix tracks the age of every sample and flags stale coverage. Freshness is what turns the score into a living metric you track over time rather than a one-off snapshot.

Live, cached and modelled cells

  • Live — freshly sampled against the real engine, metered against your credit balance.
  • Cached — a recent live result reused to save credits, labelled with its age.
  • Mock (sample) — clearly-labelled fixture data, never presented as live.

Improving coverage

You raise coverage and freshness by running more live samples — but live runs cost credits multiplicatively (prompts × engines × K samples × cadence), so escalate deliberately rather than sampling everything. The predictive model covers the rest cheaply. See live vs. predictive, scan cost and credits to plan where to spend. Adding a SERP-provider key unlocks live coverage for AI Overviews and AI Mode — see AI Overviews.