Google Ads and GA4 conversions differ. What should your report say?

A conversion gap is a question to investigate, not a ready-made conclusion about campaign performance. For a growing business or an agency, the immediate task is to establish whether two totals describe the same thing. Only then can a report explain what changed and what needs attention.

The following reconciliation method is a MindVerve recommendation. It preserves the original evidence, separates definitional differences from demonstrated problems and gives AI a limited role in writing up verified findings.

The Google documentation cited here was checked on 17 September 2026, Bangkok time. This date records the documentation check, not a new product announcement.

Start with the report you actually need

Write down the business question before opening a second export. Are you comparing purchases recorded during a month, or outcomes associated with advertising interactions in that month? Those questions can require different reporting choices.

Keep the raw exports unchanged. Record the account or property, event or conversion action, measurement source, exact report, metric and dimension. A dimension is the category used to break down a report, such as a traffic source. A number detached from those labels is not enough to support a comparison.

Distinguish GA4 key-event totals, Google Ads conversions based on GA4 key events, and conversions measured natively in Google Ads. This setup shares conversion data between Ads and Analytics, so comparisons can match when their scope is aligned. Do not assume that all three labels identify the same total.

Make the date basis explicit in both exports

Record the date range, time basis, time zone and export timestamp. Google Ads distinguishes ad-interaction dates from conversion dates and provides conversion-time columns for comparisons by conversion date. Check the selected report; do not assume every Ads view uses one date basis.

Consider a hypothetical click on 30 June followed by a purchase on 2 July. Depending on the selected time basis, the purchase can appear in different reporting periods. This date-only example says nothing about which system or channel receives credit.

A purchase occurring days after a click is different from data taking time to process. GA4 processing can take 24–48 hours, and reports may change. Separately, attribution credit for key events can change for up to 12 days after recording. These are not guaranteed completion deadlines or a rule to delay every decision for 12 days. Mark recent extracts as provisional where appropriate and schedule a specific recheck.

Check what is counted before explaining why

Compare the included actions and their counting settings. Google Ads “One” counts once per ad click per conversion action, not once per unique person. “Every” counts each conversion for that action. Record the actual setting; there is no useful universal default to assume.

“Conversions” and “All conversions” can contain different actions. Primary actions feed the Conversions column and bidding when their associated standard goal is used. Secondary actions appear in All conversions and ordinarily serve observation, but secondary actions in a custom goal can still be used for bidding.

Next check the attribution model and eligible channels. Attribution determines how credit is assigned to interactions. Data-driven attribution can produce fractional credit. In GA4, reporting attribution concerns event-scoped traffic dimensions; Session source and First user dimensions do not behave in the same way. Record the report and dimension rather than describing Ads as always last-click or GA4 as always cross-channel.

Do not change settings just to make totals agree. A change may affect reporting or bidding. Tags and attribution windows are also possible causes to inspect, but a possible cause is not evidence that it explains this gap.

Keep one reconciliation record for each comparison

Use the following compact record beside the untouched exports. Complete it for both sides before calculating a difference; if a field is unknown, keep it visible and assign a check.

1. Identity: account/property; event/action; measurement source; exact report, metric and dimension.
2. Scope: date range; time basis; time zone; export timestamp; count settings; attribution model; eligible channels; goals and included actions.
3. Finding: difference; supporting evidence; status; owner; next check.

Calculate totals and differences with deterministic arithmetic, such as a checked spreadsheet formula. Classify each gap as explained by definitions, an implementation or configuration issue supported by evidence, or unresolved. For the hypothetical June–July example, a documented difference in time basis supports the first status; without that documentation, the cause remains unresolved.

The owner should be responsible for a concrete next check, such as confirming the selected time basis or inspecting a named action’s configuration. Do not average incompatible metrics, invent a normal discrepancy percentage or blame tracking or consent without evidence.

Give AI the findings, not the authority to choose a total

A responsible brief might read: “Use these approved, anonymised extracts and completed reconciliation records to draft the explanation. Separate verified findings from unresolved questions. Preserve provisional labels, cite the supplied evidence and propose questions for each owner. Do not infer missing causes or recommend account changes.”

Keep the model’s access read-only and limited to approved samples. A person validates the claims and any proposed changes; budget, bid and measurement changes are not delegated by default.

If the scope and value are still unclear, MindVerve’s AI Opportunity Sprint is the diagnostic option. If one reporting workflow is sufficiently defined, AI Workflow Launch covers one scoped implementation with tests, operating documentation and handover. Licences, API usage and additional integrations are specified in the quote. The next step is a checked reporting workflow, not a promise to repair every tracking issue.