NEW:Turn AI visibility insights into ChatGPT ad creative.

Metrics and methodology

Definitions, scope and limitations for AI answers, website Health, crawler activity and imported Ads results.

On this page

Overview

This reference explains what the displayed numbers and evidence types measure. Always identify the site or workspace, observation dates, successful coverage and applicable filters before comparing values. A missing observation is not a measured zero.

All numerical examples below are illustrative arithmetic, not customer outcomes. Counts have no percentage denominator unless one is explicitly stated.

AI Visibility: selected collection

The following rows use one selected site/project and collection. Each successful provider response or repetition counts as one response. Providers are pooled without separate weighting. Failed and unfinished runs are excluded from answer metrics. These shared rules apply to every row in this table.

UI label or observationMeaning / numerator or countDenominatorTime windowSource evidenceExampleLimitation
VisibilitySuccessful target recommended or target mentioned responsesAll successful responsesSelected collectionrun status and target records3/4 = 75%, rounded to whole percentNeither unique-question coverage nor population-wide visibility; with no successes, no usable percentage exists
RecommendedSuccessful responses classified as target recommendedNone; countSelected collectiontarget recommendation records2 responses = 2Classification is recorded evidence, not a provider approval contract
MentionedSuccessful responses naming target without a recommendationNone; countSelected collectiontarget mention records; recommendation takes precedence1 mention-only response = 1Other views may use mention inclusively or count other subjects
CompetitorsDistinct non-target subject strings plus configured competitor names with an applicable explicit domain citationNone; countSelected collectionnon-target mention subjects and configured competitor domain matchesTwo collected names = 2Not guaranteed to be fully alias-normalized unique businesses; not share of voice
Sources citedDistinct explicit-citation URL stringsNone; countSelected collectionexplicit citation relationshipsA, B, B, C = 3Exact URL strings, not unique domains or citation-bearing responses
From your siteDistinct explicit URL strings marked ownedNone; subset countSelected collectionrelationship plus stored ownershipOne owned URL among three cited URLs = 1A third-party page about you remains third-party

Completion, history and prompt cells

UI label or observationMeaning / numeratorDenominatorScope and timeProvider treatmentIncomplete/failed evidenceSourceExampleLimitation
Responses completedSucceeded runsExpected runsOne collectionEach planned response/repetition is a positionFailed/pending remain outside numerator; show their states separatelyjob counts4/6 successfulDoes not identify both remaining runs as failed
Platform completionDistinct providers with at least one successNone; count; compare separately with expected platformsLatest Monitoring collectionA partial provider can count as completed at platform levelOne success does not erase its other failuresloaded run provider/status3 providers with a successPlatform completion is less precise than response completion
Expected platformsDistinct provider identifiers in loaded runsNone; countLatest Monitoring collectionOne per loaded provider IDMissing run rows may omit planned providersloaded run rosterFour provider IDs = 4Not an independent schedule roster
Prompt matrix resultStrongest recorded target outcome: recommended, then mentioned, then missedNone; categoricalExact prompt text/provider in selected collectionRepetitions collapse to the strongest outcomeVerify underlying run status; unavailable/failed cannot establish absencegrouped run resultOne mention and one recommendation show recommendationDoes not show the distribution across repetitions
Explicit citationURL cited in a recorded answerNone unless a separately defined report counts itIndividual answer and collection dateProvider-specific normalized evidenceNo usable answer cannot prove absence of citationsexplicit relationshipAnswer explicitly links URL ADistinct URL count differs from occurrence count
Returned sourceSource returned alongside an answerNoneIndividual response and dateDepends on provider payloadMissing source metadata may be a coverage limitreturned relationshipProvider returns A without citing it in final answerNot automatically an explicit citation
Searched sourceSource recorded as search evidenceNoneIndividual response/search record and dateDepends on provider evidenceAn unavailable search trace is not proof of no searchsearched relationshipSearch evidence contains BNot proof of final-answer citation
Owned sourceSource assigned to the site's ownership contextNoneSource record and relevant collectionOwnership is separate from provider/relationshipUnknown ownership should remain unknownownership fieldExplicit URL on approved owned domainOwnership alone does not make a returned source an explicit citation
Competitor appearanceApplicable non-target mention/recommendation or explicit configured-domain citationNone unless report defines a countInspect question/provider/date; competitor view may use retained historyKeep provider contextFailures are not negative competitor observationssubject/domain evidenceA competitor named in one successful answerTracked configuration and detected candidates are different sets
Monitoring history mentionsLoaded runs with a mention or recommendation record of any subjectNone; countOne listed historical collectionLoaded provider runs pooledMapper does not explicitly filter success; inspect underlying statuseshistory mappingA competitor-only answer may contributeBroader than target-only Overview; reconcile label before equating
Monitoring history sourcesDistinct URL strings across all loaded citation relationshipsNone; distinct countOne listed historical collectionProvider records pooledMapper does not explicitly filter success; coverage depends on loaded evidencehistory mappingReturned A and explicit A count as one URL; searched B adds anotherNot Overview's successful-response, explicit-only URL metric

Trend and manual reporting calculations

UI label or observationMeaning / numeratorDenominatorScope and timeProvider treatmentIncomplete/failed evidenceSourceExampleLimitation
Visible percentage in a trend bucketSum of target recommended plus target mentioned responsesSum of successful recommended, mentioned and missed responsesSelected project; UTC day buckets for 7/30 days, UTC months for 12 monthsPooled response counts, no equal-provider weightingJobs with no success absent; gap is not measured zerotrend counts and frontend buckets1/2 plus 9/10 = 10/12 = 83%A point may include several collections
Not visible percentageSuccessful target-missed responsesSame successful response denominatorSame trend scope/windowSame pooled countsFailed runs excludedRecorded evidence2/12 = 17%Only the observed successful responses
Range percentageAll visible responses in rangeAll successful responses in rangeAvailable history in selected rangePooled successesRetention and missing collections limit coverageRecorded evidence10/12 = 83%Selector does not guarantee that much history exists
Client-report owned-citation rate; manual field, not a claim about a PB cardSuccessful answers with at least one explicit owned citation, once per answerSuccessful answers in the report's defined comparable subsetManually specified site/questions/periodState included provider profilesDisclose exclusions; zero denominator is unavailableapproved answer evidence plus report method2/10 = 20%Different from the distinct owned-URL count
Educational share of voice; manual formula onlyWhatever eligible brand appearances the report explicitly definesDefined eligible appearances across its selected brand setExplicit manual scope and periodDisclose weighting and provider setState exclusions and coveragereviewed evidence and written formula3 of 10 defined appearances = 30%Do not label this a shipped PB metric without verifying the implemented formula

Target matching in the Overview uses the configured brand/aliases. Historical trends use stored target records and exclude rejected records. Changes in configuration or review decisions can affect comparisons. Retain the method used for each report.

Search & AI Visits and portfolio activity

The inspected page loads a 30-day aggregate per available site, up to 100 recent requests and 250 page rows per site, plus a separately requested 365-day trend. These are different scopes. Site selection applies across those records; a chart's time selector is not a global time filter for every card and table.

With all platforms and no specific path filter (including the default slash value), cards use the aggregate. Platform or specific-path filtering switches relevant values to the bounded loaded request sample. Providers here are user-agent classifications of recorded requests, not AI answer providers and not authenticated proof of a crawler operator's identity.

UI label or observationMeaning / numeratorDenominatorScope and timeIncomplete/failed evidenceSourceExampleLimitation
Visits / RequestsRecognized crawler activity log rowsNone; countSelected available sites; 30-day aggregate or filtered recent sampleFailed requests remain included; unavailable site fetches can be omittedqualifying logs120 matching logs = 120 visitsNot human visits, unique users or AI answer appearances
Pages visitedAggregate: distinct stored full URL strings per site, summed across sites; filtered sample: distinct site/path pairsNone; countSame aggregate/sample scopeMissing coverage is not zero pagesURL/site/pathThree qualifying URLs = 3 in aggregateAggregate and sampled path definitions can differ
Cache hits cardLogs with HIT cache statusAggregate: HIT, MISS, STALE or COALESCED logs; sampled UI: loaded rows with a displayed cache valueSame card scope; rounded percentageStatus filtering is not identical to successful delivery; empty denominator has no usable rateRecorded evidence80 hits / 100 eligible aggregate cache responses = 80%Sample denominator may include Bypass; reconcile before comparison
Fresh renders cardMISS logs with HTTP status 200–399None; countSame card scopeMissing status and failed status excludedRecorded evidenceFive successful MISS rows = 5Includes redirects; does not by itself establish expected page content or billable usage
Failed requestsLogs with null status, status below 200 or at least 400None; countSame card scopeNull status counts as unsuccessful classificationRecorded evidenceTwo 403s plus one unknown = 3Inspect cause; not all failures are crawler blocking
Portfolio hits/cache/last crawlPer-site activity count, cache-hit ratio and most recent activity timeCache ratio only: eligible cache responsesLoaded 30-day rollup, with recent-row fallback where usedNo available data cannot establish inactivitysite augmentation80/100 = 80% cacheDo not equate it with a billing-cycle allowance
Chart Cache hitsHIT log countsNone; countSelected sites; 7/30-day or 12-month view of loaded historyMissing site retrieval can reduce coverageUTC-day source trendTwo HIT rows in bucket = 2A displayed zero in a filled bucket does not prove complete observation
Chart FreshAggregate trend: all MISS logs, without card's successful-status condition; sampled fallback: non-Hit rowsNone; countSame chart view; filtered views can use recent sampleAggregate includes failed MISS; fallback can include other cache statesRecorded evidenceOne failed MISS may increase chart Fresh but not Fresh renders cardKnown definition mismatch; not a substitute for fresh-render billing usage
Visiting platformsUser-agent-classified Google, Bing and AI request countsNone; countSame chart window/site or filtered sampleRequest failure does not imply no recorded requestRecorded evidenceThree classified request rows = 3 observationsClassification does not prove training, indexing, citation or downstream use
Delivered / Failed rowDelivered when HTTP status is 200–399; otherwise FailedNone; categoricalOne request timestampNull code is unsuccessful; rendering provenance may remain unverifiedstatus and delivery fields302 can display DeliveredFollow redirect and inspect delivery/content; Delivered is not verified semantic success
Fresh / Stale rowSuccessful non-STALE row is labeled Fresh; successful STALE is StaleNone; categoricalOne request timestampFailed status displays Failedcache/status mappingA successful cache HIT may be FreshFresh label does not prove the content reflects the latest CMS change
Rendered content access / DeliveryStored rendered-content flag and delivery sourceNone; categoricalOne request timestampUnknown flag remains unverifiedprovenance fieldsCached rendered content versus origin responseA response can succeed without establishing rendered-content delivery

Health, Readiness and page monitoring

These observations concern recorded page evidence, not AI answer coverage. AI monitoring provider weighting is not applicable. Evidence can include different check sources, so a report timestamp does not establish that every external system was queried at that moment.

UI label or observationMeaning / numeratorDenominatorScope and timeIncomplete/failed evidenceSourceExampleLimitation
Checks by categoryChecks with Clear status in the categoryAvailable checks in categorySelected site's latest reportUnavailable comparisons may be Warning; inspect detailsreport checks3/4 clearFour checks, not every page
Platform access summaryClear Access and Rendering checksAvailable checks in those groupsSame Readiness reportMissing evidence is not positive verificationRecorded evidence3/5 clearNot an indexing or traffic score
Content understanding summaryClear Entities and Content checksAvailable checks in those groupsSame Readiness reportInspect unavailable evidence and source datesRecorded evidence4/6 clearNot measured human or AI comprehension
Critical / WarningChecks with that severityNone; countSame reportWarning can represent uncertaintyRecorded evidenceTwo warnings = 2Triage classification, not probability of harm
Open issuesNon-Clear checksNone; countSame reportDistinguish actionable problem from unavailable comparisonRecorded evidenceOne critical plus two warnings = 3Do not claim every open check is a verified defect
Monitoring cadence / configured URLsSchedule and configured page set, not a performance metricNoneCurrent configuration; actual run dates separateA schedule is not proof each check ran successfullyconfiguration and run/incident evidenceWeekly schedule, two configured URLsScheduled monitoring is not continuous uptime coverage

AI Ads and allowance terms

Ads results summarize imported provider-account performance within the workspace. The inspected implementation selects snapshots whose window end is at or after the requested rolling cutoff (default 30 days, allowed 1–90). Chart grouping uses the database day boundary; confirm the deployed timezone. Do not assume the beginning of a selected day, a site filter or exact agreement with a provider UI timezone.

UI label or observationMeaning / numeratorDenominatorProvider/scope and timeIncomplete/failed evidenceSourceExampleLimitation
ImpressionsSum of imported impression countsNone; countConnected supported provider accounts, workspace/window aboveNo import is unknown coverage, even if displayed as 0performance snapshots100 + 200 = 300Aggregate ad delivery, not exact prompt or unique-user count
ClicksSum of imported click countsNone; countSame results scope/windowImport failures leave incomplete evidenceRecorded evidence2 + 3 = 5Not conversions or organic referrals
SpendSum of imported spend micros; one currency unit is 1,000,000 microsNone; monetary totalSame results scope/window; preserve account currencyCurrency is absent from inspected results payload; confirm display and account comparabilityRecorded evidence2,500,000 micros = 2.5 units in the verified currencyDo not sum different currencies or invent conversion into dollars
Conversions fieldStored field is summed, but inspected importer supplies fixed zero rather than importing conversionsNone; not a supported observed conversion metric in this flowSame results scope/windowTreat as unavailable until measurement/import support is verifiedRecorded evidenceA displayed 0 is not evidence of zero conversionsDo not calculate conversion rate, CPA or ROAS from this field
Creative/variation resultsImported totals grouped by linked creative ID, displayed with current version title; up to 50 groupsNone; counts and spend aboveWorkspace/window; provider external ad linkageUnlinked ads may remain; exact text/URL matching can be ambiguousRecorded evidenceA linked ad contributes five clicks to creative CNot immutable version attribution; not proof the displayed current title earned those clicks
Last importedLatest imported-at value among selected snapshotsNone; timestampResults scope/windowNull/no snapshots means no confirmed import in that scopeRecorded evidenceImport time after reporting windowNot the same as provider event time or connection last-synced time
Fresh render allowanceEffective account render consumptionEffective allowance for that usage periodWorkspace/account period; not AI provider scopeReconcile reservations, failures and diagnostics with usage rulesaccount usage and render contract50 used of an effective 500 = 10% of allowanceCrawler Fresh card is a different population; cache hit does not use fresh-render allowance
Tracked prompt allowanceActive monitored questionsEffective prompt capacityCurrent account/site setupA failed response does not remove the prompt slotRecorded evidenceTwo questions across three providers still use two question slotsNot a response or API-call count
Page-monitoring URL allowanceConfigured monitored public URLsEffective URL capacityCurrent account configurationRun completion is separateRecorded evidenceThree configured URLs use three URL positionsNot prompt capacity
Generated content allowanceRecorded generation usage under effective account rulesEffective content-generation allowanceApplicable account usage periodConfirm failure/reservation treatment; saving is not publicationcontent/usage recordsOne generated draft is not one published articleDo not infer unused allowance from deleting a draft
Ads creditsPB operation credits consumed/reserved under current tariffEffective credit allowance and periodPB workspace; separate from ad-provider spendCompleted Proof can consume credit for any result; operational exception releases its reservation in inspected flowoperation ledger and account displayCopy 1 + optional image 1; completed Proof 1 in inspected revisionReverify tariff; opening saved local preview is not generation; no provider spend equivalence

Read the account's effective limit and period before interpreting usage. A legacy subscription, a trial and a current plan can have different entitlements. Do not derive them from a public plan name alone.