Write each rate as “counted action ÷ eligible audience.” Compare the same metric and segment over complete periods. A lower combined rate can come from a different traffic mix even when each segment performs exactly as before.
Separate the four signals before calculating a rate
An app owner can see more exposure, fewer page visits and a different download total in the same week. Those observations are not automatically contradictory. First ask which event each counter represents, and whether a person can contribute more than once.
- ExposureSomeone was shown the app. Check repeat versus unique counts.
- Page visitSomeone opened a product page. This is a different event.
- Button clickSomeone chose an action. Completion still needs evidence.
- Completed actionA download or acquisition was recorded under that report’s rules.
Keep website activity in another row. Someone clicking a store link on a website has left for a store; that click alone cannot establish a completed app download, an active user or a purchase. For example, a visit to the Frugal product page is a website visit, not a Frugal installation. This guide uses invented numbers, not Moocsoft product analytics or customer results.
Apple: keep impressions, page views and downloads distinct
Apple’s metric definitions distinguish repeated impressions from unique-device impressions. Product-page visits are included in impressions, so do not add the two counters as separate audiences. Total downloads combine first-time downloads and redownloads. The conversion-rate numerator also includes pre-orders; a pre-order is not counted again when its download occurs.
| Metric | Example count | How to use it |
|---|---|---|
| Unique device impressions | 10,000 | Denominator for this conversion example |
| Product page views | 800 | A separate page-view counter, not the denominator above |
| First-time downloads | 80 | Keep separate when evaluating first-time acquisition |
| Redownloads | 20 | Included in total downloads |
| Total downloads | 100 | 80 + 20 |
With these assumptions, 100 ÷ 10,000 × 100 = 1%. Dividing 100 by 800 gives 12.5%, but that is not the reported Apple conversion metric. These aggregate counters also do not establish that those 100 downloads followed those 800 page views. Do not label 12.5% a measured page-to-download conversion without an appropriate linked report.
Apple’s acquisition guide supports examining source, territory and device. Keep those filters consistent when comparing periods. Save the displayed unique count for the chosen window instead of assuming that adding daily unique counts produces unique people for the entire month.
Google Play: a click is now a separate outcome to check
Google’s current listing-performance guidance, checked November 10, 2026, describes a shift to click-focused reporting in 2026. The listing analysis reports visitors, clicks and CTR; the relevant actions can be Install, Open or Pre-register. Completed acquisitions remain available in other reports, including Statistics. Read the actual selected metric, not just the word “conversion” in an old spreadsheet.
For a deliberately simplified, install-only example, assume 1,000 eligible visitors and 120 unique people clicking Install in the same period and segment. 120 ÷ 1,000 × 100 = 12% is the click rate for that example. If a separately matched outcome report records 90 completed acquisitions, write 90 in its own column. Do not relabel all 120 clicks as installs or infer why the remaining actions did not become recorded acquisitions.
Do not compare an older acquisition-based percentage with a newer click-based percentage as though only the listing changed. Keep the old export’s metric name and definition. If the population or action differs, start a separate series rather than joining the lines into one trend.
Check whether the audience mix changed
The following independent arithmetic example uses two mutually exclusive visitor groups. Each visitor contributes at most one click within a period. It illustrates weighted rates, not the deduplication rules of a particular console.
| Period | Segment | Visitors | Clickers | Segment rate |
|---|---|---|---|---|
| A | Search segment | 1000 | 100 | 10% |
| A | Referral segment | 1000 | 20 | 2% |
| B | Search segment | 500 | 50 | 10% |
| B | Referral segment | 1500 | 30 | 2% |
Period A has 120 ÷ 2,000 = 6%. Period B has 80 ÷ 2,000 = 4%. The search segment remains at 10% and the referral segment at 2%; more of the second period’s audience came from the lower-rate segment. The combined rate fell by 2 percentage points, or about 33.3% relative to its previous level, without either segment’s rate falling.
The unweighted average of 10% and 2% is 6%, which is wrong for period B. Add eligible counts and counted actions before dividing, but only when the groups are compatible and do not double-count people. If the console uses overlapping groups, use its aggregate rather than inventing a deduplication rule.
Copy a small comparison worksheet
| Record | What to write down | Why it matters |
|---|---|---|
| Window | Start/end dates, report time zone and latest complete date | A partial day is not a complete prior day |
| Metric | Exact label, numerator, denominator and unique-count rules | Similar labels can describe different actions |
| Population | Platform, territory, source, listing and new/returning filters | Audience changes can move the overall rate |
| Counts and rate | Raw eligible count, raw action count and reported percentage | Small counts and rounding remain visible |
| Product change | Release version or listing edit and its actual publication time | Preparation time is not exposure time |
| Outcome | Download/acquisition report separately from click intent | A click is not proof of installation or revenue |
| Next check | One question, its owner and the evidence needed | The response stays tied to an observable problem |
Save a private export before making a change. Keep credentials, user identifiers and unredacted account screenshots out of a public issue or portfolio. Ten clicks from 100 visitors and 11 from 100 visitors differ by one person; that alone is not a reliable reason to announce a winning design.
Choose the next action from the evidence
If exposure drops while a comparable rate holds steady, investigate the discovery segment and report definitions first. If the same audience is visiting but the matching action rate falls, review the promise, screenshots, availability and purchase explanation. If reported clicks hold up while completed acquisitions fall, investigate the completion metric, supported devices and release behavior before rewriting all the copy. These are diagnostic questions, not proven causes.
For a new mobile app, put the measurement definitions beside the store submission checklist. At handover, agree who owns the reports and follow-up through the app handover checklist. A development brief should name the user action that matters and the evidence available to check it; it should not promise a ranking, download total or revenue result.
Common questions
Are product page views the denominator of Apple’s App Store conversion rate?
No. Apple uses unique device impressions. Total downloads and pre-orders form the numerator under its counting rules. Keep page views as a separate metric and preserve the filters for the selected report.
Does an Install click in Google Play prove a completed acquisition?
No. A button click records intent. Use the completed-acquisition report for the outcome and check that its period and population match the question you are answering.
Can conversion fall even when each traffic source performs the same?
Yes. In the fictional example, the two segments remain at 10% and 2%, but their weights change. The combined rate falls from 6% to 4%. Compare segment counts and rates before blaming a listing edit.
Can a website store-link click be reported as an app download?
Not from that event alone. Keep website visits and outbound clicks separate from store-reported downloads, acquisitions and purchases.
References and further reading
Published by Moocsoft, a Flutter development studio. This guide uses AI assistance. How we prepare guides, label examples and handle corrections.