AI-visibility numbers are easy to inflate and hard to check. This page is the antidote: exactly what we sample, how often, what counts as a mention, and what we refuse to claim.
We support 10 registered answer paths. Each uses an official grounded API, a consumer/search UI, or a raw model supplied with live retrieval. Those are not interchangeable: the exact transport and its limitations are stated on every row below.
OpenAI Search API
Grounded
Official provider API with web grounding enabled. API output can still differ from a personalized consumer application.
Gemini Grounding API
Grounded
Official provider API with web grounding enabled. API output can still differ from a personalized consumer application.
ChatGPT consumer UI
Grounded
Captured through a logged-in consumer browser session. Output and availability can vary with account, geography, session, product UI and DOM changes.
Perplexity consumer UI
Grounded
Captured through a logged-in consumer browser session. Output and availability can vary with account, geography, session, product UI and DOM changes.
Claude consumer UI
Grounded
Captured through a logged-in consumer browser session. Output and availability can vary with account, geography, session, product UI and DOM changes.
Grok consumer UI
Grounded
Captured through a logged-in consumer browser session. Output and availability can vary with account, geography, session, product UI and DOM changes.
Google AI Overviews search UI
Grounded
Captured from the Google search results interface through a browser session; availability varies by query, geography and UI state.
Kimi K2.6 via NVIDIA + Tavily
Grounded
A raw model hosted by NVIDIA answers over Tavily web-retrieval context. This is not the vendor’s named consumer application.
Mistral Medium via NVIDIA + Tavily
Grounded
A raw model hosted by NVIDIA answers over Tavily web-retrieval context. This is not the vendor’s named consumer application.
GPT-OSS 120B via NVIDIA + Tavily
Grounded
A raw model hosted by NVIDIA answers over Tavily web-retrieval context. This is not the vendor’s named consumer application.
Free is a one-time snapshot: reviewed tracked questions are sampled once on Gemini, with no ongoing tracking.
Paid subscriptions schedule their configured automatic paths on daily ticks; jobs may be staggered within the day, and browser-captured paths remain dependent on their worker sessions. The dashboard shows the last usable sample, attempted coverage, and the cadence assigned to each path.
Lifetime plans are scan-based: you trigger checks on demand from a monthly pool, plus an automatic weekly Gemini heartbeat attempt. Provider availability and the remaining pool still determine whether a usable sample is produced.
An automatic path can fail to produce a usable sample for several reasons. The dashboard calls it overdue only after its disclosed cadence window; on-demand and snapshot paths are not judged by a daily clock. A requested path that returns no valid answer is reported as incomplete rather than being converted into brand absence.
Each sampled answer gets one of three verdicts for your brand (and each tracked competitor):
The observed visibility score averages these over latest substantial prompt×path cells (recommended = 1, named = 0.5, absent = 0, scaled to 0–100). The same scale is used everywhere - dashboard, weekly email, receipts - so no surface can disagree with another.
Answer-derived verdicts, visibility scores, losses, and movements require substantial retained answer text with path and timestamp. Coverage, site-checklist, and prioritization figures identify their separate inputs. Tool chrome and unusable output produce no answer verdict.
Answers pass a sanitizer before they count. Boilerplate, truncated fragments, error and rate-limit notices, and non-answer text masquerading as answers are detected and discarded - they never become verdicts, alerts or competitor “wins.”
When a tracked answer changes after you ship a fix, you get the before/after receipt: both sanitized answer texts, with engine and dates. We say the answer changed. We never say your fix caused it - nobody can prove that, so we refuse to sell it.
An accuracy row requires an exact substantial sampled quote that conflicts with an approved fact. Independent model families can add a confidence label (low, medium, high, verified); low means a verifier disagreed. The receipt remains inspectable instead of being replaced by the confidence label.
Each path shows its assigned mode - daily, weekly, on-demand, snapshot, or historical - plus the last usable sample. Only overdue automatic paths receive a warning; on-demand and snapshot paths are not judged by a daily clock.
AI answers can vary from run to run. Change alerts ignore wording-only differences and compare substantial categorical endpoints on the same question and path (an entity entering/leaving or a new citation appearing). Each event is an observed transition, not universal ground truth, and duplicate transitions are suppressed.
That a fix you shipped caused an answer to change. We show the change; attribution in AI answers is not provable, by us or anyone.
That answers will change on any schedule. Provider retrieval and model behavior are outside our control; a later comparable sample may change immediately, much later, or never.
Coverage we don't have. If an automatic path is overdue, a requested path produced no usable answer, or a path is not included in your plan, the product says so instead of interpolating.
Precision the sample size can't support. Where evidence is thin, you'll see “inconclusive” - not a confident number.
Free one-time snapshot · No credit card