A rise in tracked conversions does not, by itself, mean advertising created more sales. It can mean the business is recognising purchases it previously missed. Google’s latest measurement tools make that distinction especially important when deciding whether to increase a budget.
Google announced Data Strength Uplift and the general availability of Meridian GeoX on September 10, 2026. This guide, reviewed September 14, explains the different questions those developments raise. It does not report a campaign experiment or promise that every advertiser can run a useful geographic test.
- 01Observation and causation answer different questions.Recovering missing conversion signals can improve measurement without creating new customers.
- 02GeoX supports a different kind of evidence.Geographic experiments estimate effects against a comparison, rather than simply crediting recorded conversions.
- 03Feasibility comes before a test budget.You need suitable outcomes, geographic variation and an experiment that can detect a decision-relevant effect.
- 04Annotate the collection change.A new integration or event mapping creates a break in the series; date it before anyone reads the rise as campaign improvement.
01 — Practical guidanceWhat Google announced
The September 10 announcement describes new data connections, Data Strength Uplift and improvements to Meridian. Google positions these as complementary tools for improving data foundations and making better investment decisions. It also says GeoX is generally available globally. Google defines Data Strength Uplift as additional conversions recovered by the first-party data setup; it is not itself a count of sales caused by advertising.
Do not collapse the release into one claim that Google can now prove every sale. An attribution system connects observed outcomes with advertising interactions under its rules. A causal experiment asks what would have happened under a different advertising condition. Better inputs can support both, but the resulting numbers are not interchangeable.
For an existing account, begin by recording what changed in collection and reporting. A new integration, event mapping or eligible data source can create a break in the historical series. Annotate that date before interpreting a rise as campaign improvement.
02 — The decisionMatch the method to the decision
The reference below separates four common decisions. It is an editorial decision aid, not a statement that one method replaces the others. A business may need a collection audit before it can trust the outcome series used in an experiment.
Google describes GeoX as an open-source geo-experimentation solution. Its availability makes a method accessible; it does not remove the need for a suitable design, credible comparison and interpretable outcome. Use the technical documentation to establish feasibility for the actual business.
| Question | Evidence to seek | What it cannot establish alone |
|---|---|---|
| Are conversions missing? | Event reconciliation and data-collection checks | How many sales advertising caused. |
| Which interactions receive credit? | Attribution definitions and reported paths | What customers would have done without exposure. |
| Did changed advertising affect sales? | A credible incrementality experiment | A universal effect across future periods and markets. |
| How should channels share a budget? | Experiment-informed modelling and business constraints | Guaranteed returns from a single historical estimate. |
03 — MethodA tracking change can move the report without moving sales
Consider an illustrative business with 100 completed orders. Its previous tracking implementation captured 60 of them; a repaired implementation captures 80. Reported conversions rise by 20, or about 33.3% relative to the earlier 60. The business still has 100 orders. These are invented teaching numbers, not Google results.
The repair is valuable because the account now sees more of what happened. But a claim that it generated 33.3% more sales would confuse observation with business growth. Reconcile to an independent order or lead record, using consistent definitions for cancellations, duplicate submissions and qualified enquiries.
When historical reporting changes, preserve both the old and new definitions. If an AI reporting assistant summarises the account, give it the change log alongside the performance data. Otherwise it may confidently attribute the discontinuity to a creative or bidding change that happened nearby.
04 — The evidenceWrite the geo-experiment brief before choosing settings
Start with a business decision: for example, whether an increase in a selected channel produces enough additional qualified demand to justify its cost. Specify the outcome, eligible locations, exposure change, test period, comparison method and the smallest effect that would alter the decision.
Geographies need enough usable variation and stable measurement to support the chosen design. Spillover between locations, a nationwide promotion or a change in store availability may complicate interpretation. Document these risks before the experiment rather than using them selectively to explain an unwelcome result.
A small advertiser with few sales or little geographic separation may not have a useful test at the desired scale. Do not invent a universal spend threshold. Review the design’s ability to detect the relevant effect and seek a narrower question or longer observation window where appropriate. Our incrementality guide covers the underlying distinction.
05 — ExampleReport uncertainty with the recommendation
A useful readout states the estimated effect, its uncertainty and the conditions under which it was measured. It also explains whether the result changes the decision. An inconclusive test should not be rebranded as proof that the channel has no value, or that it definitely works but needs more budget.
Separate measured findings from proposed action. A promising estimate may justify another bounded investment while a weak result may justify improving the data or redesigning the test. The MMM comparison explains how modelling and attribution can contribute different pieces of the same decision.
For paid-media planning, keep the final recommendation tied to the actual economic outcome. More reported events, stronger data coverage and more incremental customers can all be useful developments. They should appear as different findings in the report.
Before comparing a new report with an older one, review the marketing attribution guide and record the attribution definition used in each period.
As-of date. September 14, 2026. Sources reviewed for this article; the article is dated September 13, 2026.
Method. Primary-source review of Google’s September 10 announcement and GeoX overview. Four decision rows and one explicitly illustrative arithmetic example; no campaign data was collected.
Sources. Google measurement announcement; Meridian GeoX overview.
Limits. Availability is not suitability. No minimum spend, universal effect size or advertiser ROI is inferred from the launch.
06 — The failuresFour readings that confused the report with the business
The announcement puts a measurement tool and a causal tool side by side, and the readings below mix them up in the ways the guide anticipates. They are illustrative, like the 100-order example they reuse, and report no campaign.
The uplift read as growth
The repaired implementation captures 80 of 100 orders instead of 60. The monthly report shows conversions up about a third and the summary says sales grew. The business still has 100 orders; the account can see more of them.
The unannotated break
The new data connection went live mid-month and nobody dated it. An AI reporting assistant, given the performance data without the change log, attributes the step change to a creative refresh that happened the same week.
Availability as suitability
GeoX is generally available, so a small advertiser with few sales and little geographic separation runs a test at the scale it wanted. The design cannot detect the effect that would change the decision, and the result is read anyway.
The inconclusive rebranded
A geo test returns a wide interval around a small estimate. One reading calls the channel worthless; another says it works and needs more budget. The honest readout was that the test did not change the decision.
Each reading answers a question the data cannot answer. The brief in the next section fixes the question before the settings are chosen, which is the fourth section’s point.
07 — The methodThe geo-experiment brief in seven lines
The brief section lists what a geo-experiment needs before anyone opens the settings, and it fits on one page. The decision: the business choice the result will inform, such as whether an increase in a selected channel produces enough additional qualified demand to justify its cost. The outcome: the measure that stands for that demand, reconciled to an independent order or lead record with consistent handling of cancellations and duplicates. The eligible locations and why they have enough usable variation and stable measurement. The exposure change: what is varied and by how much. The test period. The comparison method, checked against the GeoX technical documentation for this design. And the smallest effect that would alter the decision, which decides whether the design can detect anything worth acting on.
Beside the seven lines, the risks: spillover between locations, a nationwide promotion inside the window, a change in store availability, and any collection change the first section asks to be annotated. Writing them before the experiment, as the brief section says, stops them being used selectively to explain an unwelcome result afterwards.
An illustrative brief reads: decision, whether to increase one channel’s budget by a stated share next quarter; outcome, qualified leads reconciled to the sales system; locations, a stated set of regions with comparable baseline volume; exposure, the increase applied to the test regions only; period, a stated number of weeks; comparison, the design chosen from the GeoX documentation for that structure; smallest decision-relevant effect, the lift that would cover the added cost; risks, a planned promotion in week three, noted. If the seventh line cannot be met by the design, the brief says so and the fourth section’s advice applies: a narrower question or a longer window, not a test run for its own sake.
More reported events, stronger data coverage and more incremental customers are different findings. Data Strength Uplift speaks to the first two; only a credible comparison speaks to the third. A readout that keeps them on separate lines is much harder to misread as the other.
08 — Next stepMore tracked conversions is not more sales
Write the seven lines for the test you are considering before touching the settings, and date the last collection change on the same page. If the seventh line cannot be met, that is a useful finding too, and cheaper than the test.
Decide which number would change your budget
First establish whether the report changed because the business changed or because measurement improved. Then choose a method capable of answering the investment question. A well-defined outcome and an honest comparison are more useful than a larger conversion total without an explanation.