The practical answer
Record the source, collection date, period and metric definition next to each imported value. Choose comparable current and baseline periods, and mark partial or missing periods. Calculate changes with explicit denominators and check large movements against the source export.
Plan the input and the review
- Record the source, collection date, period and metric definition next to each imported value.
- Choose comparable current and baseline periods, and mark partial or missing periods.
- Calculate changes with explicit denominators and check large movements against the source export.
Choose the operation for this step
- Inspect domain traffic metrics: Keep monthly total visits, organic-search traffic, authority scores and collection dates separate; missing estimates are not zero and estimates are not your own measured users.
- Review analytics exports: Import the intended GA4, Search Console or tool-event CSV, then check the metric and period mapping; export-start events do not prove completed downloads.
- Compare reporting periods: Review the baseline, unit and anomaly notes, then keep original values and sources beside calculated changes in the exported report.
An editorial example to compare
Make the acceptance condition visible
A public monthly traffic estimate is compared with seven days of measured search clicks and described as user growth.
Each measure has its own definition and period. Comparable click periods are compared directly; the estimate is labeled separately with source and date.
More clicks and a lower CTR can both be true
This is an illustrative dataset, not a KitNelo traffic report. Compare October 1–7 with October 8–14: two complete seven-day periods with the same Search Console property, search type and filters. The baseline has 2,000 impressions and 80 clicks; the current period has 4,000 impressions and 120 clicks. Clicks increased by (120−80)/80=50% and impressions doubled, while CTR fell from 4% to 3%.
The evidence supports more clicks from a larger impression pool, with a smaller share of impressions producing clicks. It does not establish that the title became worse. New queries, countries, devices or pages can change the mix. Group by query and page to locate the additional impressions before investigating titles and snippets within comparable segments.
| Question | Calculation | Accurate wording |
|---|---|---|
| Did clicks increase? | (120−80)/80 | Up 50%, or 40 more clicks |
| What share of impressions became clicks? | 80/2000; 120/4000 | 4% to 3%, down 1 percentage point |
| What was the relative CTR change? | (3%−4%)/4% | Down 25%, not down 1% |
Label an event ratio before calling it a user conversion
A separate illustrative dataset contains 60 task_started events, 45 task_succeeded events and 30 export_started events. Successful processing divided by starts is 75%. Export starts divided by successes is approximately 66.7%. Export starts divided by processing starts is 50%. Each is a valid aggregate calculation, but each describes a different stage. Put the denominator in the label.
One person can start, retry or export repeatedly. These totals cannot establish that 75% of users succeeded, or that every downstream event belongs to an upstream event in the same period. Cross-period completions also affect the comparison. Reconstruct a task-level funnel from linkable operation records when that is the question; a totals-only CSV cannot supply missing links. The growth analyzer maps downloads to export_started. That observes export initiation, rather than confirmed file saving or reader satisfaction.
Repair dates and duplicate records before interpreting a period report
The period report includes both endpoints and rejects overlapping periods. For October 8–14, 2026, the preceding equal period is October 1–7. Do not include October 8 in both. Normalize timestamped records to YYYY-MM-DD in the intended business timezone before importing the date field.
Keep the value column numeric: separate “1,200 visits” into 1200 and a unit note. When two exports contain the same event, investigate a stable key before adding their totals. Record original rows, exclusion reasons and retained rows after cleanup. Mark a period missing two days as incomplete instead of filling blank days with zero. Keep third-party estimates in the website metrics tool in a separate series with their provider and observation date. They do not belong inside a first-party event funnel.
Check the final result and preserve evidence
Google describes Search Console as a way to inspect search performance. Preserve the source’s metric definitions when combining reports.
Check the denominator before describing a change. Reconstruct totals before averaging rates.
Acceptance checks
- Comparison periods and units match.
- The report distinguishes estimates, observations and missing values.
- Percent change and percentage-point change are labeled correctly.
Eight-day exercise: take the unfinished Monday out
All daily records are fictional and use one timezone. Monday October 12 has 40 organic-search UV and eight distinct users completing the workflow; the complete Monday October 19 has 50 and ten. October 26 at 10:00 has only 12 UV and two completing users. Comparing that partial day with a full 50-UV day and announcing a 76% decline is premature. The two complete Mondays show +25% UV, +25% completing users and the same 20% user conversion.
Do not add daily UV and call the sum weekly unique visitors: the same person may return on several days. An interval-level unique-user report must deduplicate across the entire interval. The exercise supports single-day comparisons; it contains no identities from which interval uniques could be reconstructed. Show complete-day observations when a week is unavailable, without calling three days a steady trend.
| Comparison | Calculation or treatment | Supported statement |
|---|---|---|
| Complete Monday 10/12 vs complete Monday 10/19 | (50−40)/40 = 25% | Organic UV rose 25% between these complete Mondays |
| 10/26 morning vs all of 10/19 | Exclude the partial day; await complete data | A daily comparison is not yet supported |
| 10/19 completing users vs eligible search UV | 10/50 = 20% | User conversion within the same source and user scope |
After changing tracking, check whether the question changed too
Suppose completion included samples before October 20, but subsequently required real input, successful processing and the final report step. Fewer events can reflect a stricter definition. Mark the measurement change and retain both definitions; do not interpret the break as product deterioration or an SEO penalty. The pack labels the regimes legacy and real-only, without offering a directly comparable conversion lift across them.
Event success is successful processing attempts divided by started attempts under one definition. User conversion requires distinct users with at least one completion divided by eligible users in the same period, source and user scope. One user repeating the process three times is not three converted users. If completion users cannot be restricted to google/organic, report source UV and sitewide events separately rather than dividing mismatched populations.
On the search side, follow Google’s Search Console explanation to separate crawling, indexing and search performance before comparing analytics. Deployment, index eligibility, impressions, clicks and real completion are different states. A weekly review needs actual counts, data cutoff and a verifiable next change; publication volume cannot guarantee rankings or UV.
A review statement that can be checked
Traffic grew steadily for a week after publication, proving the new blog works.
Two complete Mondays have 40 and 50 source-matched UV and eight and ten completing users; user conversion is 20% on both. Two observations and an intervening tracking change do not establish blog causation or sustained growth. Compare complete periods and inspect target-article impressions and clicks next.
Common questions
Can authority scores be added to visits?
No. They describe different constructs and units; present them separately.
What if the baseline is zero?
Ordinary percent-change division is undefined. Report the actual values and explain the situation instead of forcing a percentage.
Does a tool completion event mean a person accepted the file?
No. Processing completion, export initiation and human acceptance are distinct observations.
Further reading & sources
Put the guide into practice
Open the relevant tool, preview the result and check your exported file.
