The short answer A scheduled observation, an execution attempt, a completed response, a usable measurement, and a billable unit are not automatically the same thing. Preserve each denominator, then apply the controlling vendor policy exactly as documented.
Evidence flag, not an overcharge verdict A variance means the reconstructed record and supplied invoice do not yet agree. It can reflect missing rows, taxes, weighted engines, retries, credits, plan changes, or a policy the reviewer has not supplied. This calculator does not interpret a contract, determine entitlement, allege fraud, or contact a vendor.

Separate measurement status from billing status

Measurement evidenceScheduled, attempted, completed, usable, no-answer, other completed, failed, and retry counts remain visible as separate fields.
Supplied billing policyEach event type receives the unit weight documented by the reviewer. A blank or unknown rule stays unresolved rather than being guessed.
Human decision boundaryThe output flags arithmetic and evidence conflicts. It does not decide whether an invoice, credit, waiver, refund, or contractual remedy is valid.

A failed attempt is missing measurement. A valid answer that contains no AI answer, citation, or brand mention may still be a completed and billable observation under one provider's policy. Another provider may waive it. The event record and billing rule must therefore be evaluated separately.

Browser-local schedule-to-invoice worksheet

Reconstruct the billing period

Use the executed agreement, provider documentation, usage export, run history, invoice, and credit memo. All numbers and policy weights below are reviewer-supplied.

01 · Review identity

Name the exact service and period

Enter the organization.
Enter the provider or tool.
Enter the invoice identifier.
Enter the service-period start.
Enter the service-period end.
Enter a three-letter currency code.
Name the reviewer.
02 · Supplied policy and invoice

Record the billing rule before calculating units

Supply the controlling policy source.
Enter zero or a positive amount.
Enter zero or a positive unit allowance.
Enter zero or a positive rate.
Enter zero or a positive amount.
Enter zero or a positive amount.
Enter zero or a positive invoice total.

Unit weights from the controlling policy

Enter the number of billable units assigned to one event of each type. Use 0 only when the supplied policy explicitly excludes that event. These categories are added, so do not give a scheduled row and its outcome a weight unless the policy genuinely charges both.

Supply a non-negative weight.
Use for capacity reserved or schedule-based charging.
Supply a non-negative weight.
A completed observation that passes the declared usability rule.
Supply a non-negative weight.
A valid completed result reporting no AI answer or other declared null outcome.
Supply a non-negative weight.
A completed result excluded by the reviewer's usability rule for another reason.
Supply a non-negative weight.
A provider error, timeout, blocked request, or other non-measurement attempt.
Supply a non-negative weight.
An additional retry charge, separate from the outcome weights above.
03 · Delivery and usage evidence

Reconcile each run or homogeneous cohort

One row may represent one execution or a cohort only when every observation in it shares the same engine, status rules, unit weights, and source. Completed must equal usable + no-answer + other completed. Attempted must equal completed + failed. Retries are additional attempts and must not create extra final observations.

DateRun or cohort IDPrompt or cohortEngine / market ScheduledAttemptedCompletedUsable No answerOther completedFailedRetries Provider unitsExpected unitsEvidence sourceAction
Blank rows are ignored. Active rows require a run ID and all count fields.
04 · Calculate and export

Run the evidence checks

Examples are illustrative reviewer inputs, not statements about any named provider. Loading an example replaces the current form.

Complete the review identity, supplied policy, invoice, and at least one evidence row.

Formulas and evidence rules

  • Delivery completion = completed observations ÷ scheduled observations. A retry attempt does not create another final observation.
  • Usable completion = usable observations ÷ scheduled observations. The reviewer must state the usability rule.
  • Unusable share = (completed but non-usable + failed attempts) ÷ attempted executions. It describes the supplied execution record, not brand visibility.
  • Expected units = the sum of each observed event count × its reviewer-supplied unit weight. The calculator does not infer vendor policy.
  • Expected overage = max(0, expected units − included units).
  • Expected invoice = base subscription + expected overage × supplied rate + supplied tax/pass-through − documented credit or waiver.
  • Cost per scheduled observation = supplied invoice total ÷ scheduled observations.
  • Cost per usable observation = supplied invoice total ÷ usable observations. It remains unavailable when the usable denominator is zero.
Count relationships catch ledger defects For each row, completed must equal usable + completed no-answer + other completed; attempted must equal completed + failed; completed cannot exceed scheduled; and attempted cannot exceed scheduled + retries. A duplicate run ID or a missing evidence reference is surfaced before arithmetic is treated as review-ready.

Why the provider policy has to be supplied

Current providers use materially different billing units and failure rules. These examples are evidence that no universal default is supportable; they are not endorsements and should not replace the policy controlling your invoice.

  • Keyword.com: How AI Visibility Credits Work defines a run as search term × AI engine × refresh event, assigns engine-specific weights, resets unused monthly credits, and says there are no automatic overages.
  • Apify: AI Search Visibility Tracker documents one implementation where completed no-answer checks are billable while failed rows are not. That distinction is specific to the published actor policy.
  • OpenRouter: Zero Completion Insurance waives requests with zero completion tokens and a blank or error finish reason. Output-bearing requests and separate downstream charges may require different evidence.

Record the source, version, and effective date that applies to the reviewed period. If a public page changed after the invoice period, preserve the contemporaneous contract, order form, export, screenshot, or written provider answer instead of silently applying today's rule backward.

Frequently asked questions

How do you audit an AI visibility monitoring invoice?

Preserve the service schedule, execution ledger, outcome status, provider-reported units, controlling billing policy, included allowance, overage rate, credits, taxes, and invoice total. Reconstruct the units under the supplied policy, then investigate every row or amount variance.

What is cost per usable AI visibility observation?

It is the supplied invoice total divided by completed observations that meet the reviewer's declared usability rule. It is an operational cost measure, not a visibility score or proof of business value.

Should a failed AI answer be recorded as zero visibility?

No. A failed retrieval or provider error is missing measurement, not evidence that the tracked brand had zero visibility. Keep it in the execution and billing record, but out of the visibility-result denominator unless the declared method says otherwise.

Is a completed no-answer result usable?

Sometimes. A valid observation that proves an AI feature did not appear can be decision-useful and billable. A completed response that cannot support the declared visibility metric may be non-usable. Define the rule before classifying rows and keep billing treatment separate.

Does a variance prove an overcharge or a refund right?

No. A variance is an evidence request. Contract interpretation, refund eligibility, tax treatment, and legal conclusions require the controlling agreement and qualified human review.

Does this page upload invoice data?

No. The calculator, copy action, and Markdown download operate in the browser. There is no upload or storage endpoint.