Understanding why verification vendors disagree

Two verification vendors can disagree because they observed different impressions or signals, applied different rules, or reported at different times. A percentage gap alone cannot show whether coverage, rule, timing or evidence caused the difference, or which vendor is correct.

Diagnose the gap in that order: align the population, compare the rule, align the reporting cut-off, then test disputed cases against evidence. If a report omits its denominator, rule version or reporting status, record the field as missing rather than assuming the reports are comparable.

A vendor comparison may name DoubleVerify, Oracle's MOAT or Grapeshot. The names alone do not show that reports use the same impression population, rules, reporting cut-off or evidence, so compare those fields before interpreting the gap.

Diagnose a verification gap in order

Reconcile the population first

Ask each provider for the campaign, dates, time zone, formats, placements, tag or SDK deployment and impression count used as its base. One provider may receive an event the other never sees.

A player may execute one verification resource but not another, and a restricted environment may expose only some signals. For viewability, keep viewable, non-viewable and undetermined counts separate. A measured rate describes measurement coverage; a viewable rate describes the share of measured impressions that passed the rule.

Create a reconciliation sheet by format and placement. Put each provider’s eligible, observed and unknown counts beside the relevant ad-server count. The ad server is a reference for investigation, not automatically the correct verification denominator. Record where counts are taken and whether invalid-traffic filtration has already been applied.

Check whether both reports use the same inclusion and exclusion boundaries, not merely the same campaign label. If a report gives a rate but not the counts behind it, record that breakdown as missing; the rate alone cannot show whether the gap began with coverage or classification.

Different eligible counts point to a population boundary to investigate. If eligible counts match but observed counts differ, investigate delivery or signal access before treating the gap as a difference in verdicts.

Population reconciliation sequence

Compare the decision rule next

For viewability, compare the applicable pixel and time criteria, in-focus treatment, large-ad exceptions, player geometry and qualifying interactions. Providers can produce different results even when they cite the same headline threshold if their measurement access or permitted methods differ.

For invalid traffic, ask which general and sophisticated methods each service applies and when. Some sophisticated findings require later analysis, so a provisional report and a final report can differ. Traffic remaining after filtration is not proven human.

For content suitability, compare taxonomy version, category mapping, property or content granularity and any adjacency rule. A domain-level label and a page-level classification describe different evidence. A rule that restricts an entire category can also disagree with one that flags only defined adjacent content.

Write down each service’s rule and the evidence it used, rather than relying on a shared label. A matching label does not establish that both services had the same measurement access or applied the rule in the same way.

For matched cases, classify a difference as a rule issue when the written criteria or mappings differ. If the criteria match but the inputs or signals do not, investigate evidence; if the reports use different cut-offs or statuses, investigate timing.

What a disagreement can point to

Investigate a sample of disputed impressions

Select a documented sample from the affected placement and period. Request the identifiers or other permitted matching keys, timestamps, rule versions and evidence each provider can share.

For content cases, preserve the state available when the decision was made. For video, record the player and any server-side insertion details. Ask the publisher or platform about missing signals where it controls the integration.

Classify each difference as a population gap, signal gap, rule difference, timing difference or unresolved case. Leave an impression unresolved when the available evidence cannot support pass or fail.

Compare matched cases at a common reporting cut-off. If a finding changes between report statuses, check timing; if it stays stable but the providers classify the same case differently, compare their rules and evidence.

If the providers cannot share a permitted matching key, do not treat unmatched cases as impression-level proof. Keep them separate as unresolved and compare the aggregate counts for the same placement and period.

Disputed-impression investigation checklist

Decide what happens next

Agree which metric and provider governs a commercial decision, what discrepancy triggers review and what evidence could change a finding. Correct a broken tag or mapping when supported by the investigation.

If disclosed policies differ, use the rule agreed for the campaign and retain the other result as context. Escalate material unresolved cases with the providers and seller. The goal is an explainable decision for the same inventory, not identical numbers from every system.

Record the placement, period, denominator, rule, report status and evidence reviewed in the outcome. Keep unresolved cases distinct from pass or fail so a later comparison can use the same basis.

Decision and documentation checklist