Troubleshooting

Something isn’t showing the score you expect? Find your symptom below. Most issues come down to one of four things: no property yet, a failed readiness check, an empty credit balance, or an engine that needs a search-provider key to run.

“I don’t have any data yet”

A workspace starts empty — a score only exists once you have a property (one tracked site) and it has been scanned. Enter your URL in onboarding: Lokrix creates the property, analyses the brand and industry, generates the prompt universe, and runs the first scan. Nothing to see usually means that first flow hasn’t completed.

  • Confirm you’re looking at the right property in the property switcher, not an empty one.
  • If onboarding stalled, re-run it from the property’s settings — it’s safe to repeat.
  • Background: see per-property onboarding and the property switcher.

The readiness pre-check failed

Before a scan spends credits, Lokrix runs a readiness gate: the URL must resolve, the site must be reachable and crawlable, and the brand/industry detection must succeed so prompts can be generated. If the gate fails, the scan won’t start — that’s by design, to avoid burning credits on a run that can’t produce a meaningful result.

Readiness pre-check: can AI crawlers reach you?An AI crawler reaches a gate that checks whether robots.txt and the sitemap allow access; the gate branches to a pass state (AI crawlers can reach you) or a fail state (blocked — fix robots or sitemap). A badge notes it costs no credits and needs no property.no credits · no property neededAI crawlerGPTBot, etc.robots +sitemap OK?Reachable — passBlocked — fix it
  • Check the URL is correct, public, and returns a 200 (no auth wall, no geo-block).
  • A robots.txt or firewall that blocks crawlers will fail readiness — allow Lokrix’s fetch or use a reachable page.
  • If brand detection came back thin, add context in the property settings and re-run, then read the readiness pre-check.

Analyze or scan errors

A scan orchestrates many engine calls, so a transient provider error can interrupt part of a run. Scans are resumable — retry before assuming the result is wrong.

  • Partial results: if some engines returned and others didn’t, the score still shows a confidence interval — a wider one, because fewer samples landed. Re-run to tighten it.
  • A provider is degraded: individual engine providers occasionally rate-limit or time out. Lokrix retries with backoff; if one engine stays down its component is marked unavailable rather than faked.
  • Unexpected result? Remember answers are non-deterministic — one run is a sample, not a verdict. Review the sampling & confidence intervals before concluding anything moved.

The scan won’t run — credits

Live engine runs are metered against your workspace credit balance — roughly one credit per engine response. Because cost is multiplicative (prompts × engines × K samples × cadence), a broad live scan can exhaust credits quickly, and Lokrix will block the run rather than overspend.

  • Check the workspace balance and any spend cap or budget that may have paused live runs.
  • Narrow the run — fewer prompts, fewer engines, or a lower sample count K — to fit the budget.
  • Don’t need fresh live data? The predictive model estimates visibility without spending credits. See credits & metering and budgets & spend caps.

An engine isn’t running (AI Overviews & AI Mode)

Five engines run live by default — ChatGPT, Claude, Perplexity, Grok and Gemini. Google AI Overviews and AI Mode are observed through the search results page, so they need a search-provider (SERP) key configured (e.g. Serper or SerpAPI). Without that key, those two engines simply won’t appear in a live run — that’s expected, not a bug.

  • Add a SERP provider key in the workspace integration settings to enable AIO and AI Mode.
  • Remember they’re scored separately: they cite the same URL only ~14% of the time, so a low result in one doesn’t predict the other — see AI Overviews and AI Mode.

“Is this number real?” — reading provenance

Every value carries a provenance label so you always know what you’re looking at. If a figure seems too good, too old, or too stable, check the label first.

Provenance labels: live, cached, predicted, mockFour provenance chips — live (emerald), cached (amber), predicted (indigo) and mock (violet) — each labelling how a figure was produced, so sampled or cached data is never shown as live.Livequeried just nowCachedrecent snapshotPredictedmodel estimateMocksample — never live
  • Live — a fresh real engine run (spent credits).
  • Cached — a recent live result reused to save cost; re-run for the current picture.
  • Mock (sample) — clearly-labelled fixture data shown in demos and dev. It is never presented as live; if you see it in production you’re in a sample view. More in data honesty.

Still stuck?

Work back through the pipeline in order — property exists → readiness passed → credits available → engine keys configured. The FAQ covers the conceptual questions, and scans overview explains each stage a run passes through.