Claims, numbers and provenance
Keep the calculation and its conditions beside the number.
A number becomes a useful claim only when the reader can tell what was counted, compared and excluded. Preserve the metric definition, denominator, time window, method and uncertainty so that a percentage does not imply more than the evidence establishes.
Free learning from Semantec SEO.
Define the metric and comparison.
Name the quantity in ordinary language and record the system that produced it. Search clicks, website sessions, enquiries and sales describe different events. Do not relabel one as another because it makes a result sound stronger.
For a ratio, identify the numerator and denominator. For growth, identify the baseline and later figure. Record units, filters, inclusion rules and any change in measurement. A comparison using different event definitions cannot be repaired by adding a percentage sign.
If a tool has its own metric definition, consult its current documentation. For example, Search Console’s Performance report documentation explains its search metrics and dimensions; it does not establish that another analytics system counts the same event.
Preserve time, scope and traceability.
State both date ranges, their lengths and the relevant market, device, page set or audience filters. Check that the periods and collection methods are comparable. Mark partial periods and changes to tracking, consent, availability or page scope.
Keep a record of the original values, export or observation reference, collection time, formula, rounding and person responsible. A screenshot alone may hide filters or omit rows needed to reproduce the calculation.
Trace the published sentence back to that record. If the source is corrected or the metric definition changes, identify every dependent page, chart and summary. A new page update date does not turn old observations into current evidence.
Worked example: repair an unsupported growth headline.
Consider the fictional headline “Traffic grew 50%.” No measurement is reported by Semantec SEO in this example. The invented record below demonstrates how to make a claim inspectable.
Suppose the teaching record defines the metric as sessions in a fictional analytics export, with one unchanged site scope and filter set. It contains 100 sessions for 1–30 September 2026 and 150 for 1–30 October 2026. Both comparison windows are thirty days.
The absolute change is 150 minus 100, or 50 sessions. Relative growth is (150 minus 100) divided by 100, multiplied by 100: 50%. The denominator is the earlier 100 sessions, not the later 150.
A bounded teaching sentence is: “In this fictional dataset, recorded sessions rose from 100 in 1–30 September 2026 to 150 in 1–30 October 2026, an increase of 50 sessions or 50% relative to September, using the same stated scope and filters.” This demonstrates reporting; it is not a client result or forecast.
The calculation does not establish why sessions changed, that every visitor was satisfied, or that sales increased. A claim that a content change caused the increase would require evidence capable of addressing other explanations.
| Field | Fictional figure |
|---|---|
| Metric | Recorded sessions in the stated teaching export |
| Baseline | 100 sessions, 1–30 September 2026 |
| Later period | 150 sessions, 1–30 October 2026 |
| Absolute change | 50 sessions |
| Relative formula | (150−100) ÷100 ×100 =50% |
| Unknown | Cause, visitor satisfaction and sales effect |
Handle zero baselines and percentage points.
A baseline of zero makes the ordinary relative growth formula undefined because it divides by zero. Report the absolute change and the values instead of inventing an infinite or arbitrary growth percentage. If the baseline is missing, obtain it before publishing relative growth.
A rate moving from 2% to 3% rises by one percentage point and by 50% relative to the original rate. Those statements describe different comparisons. Show the underlying counts and denominator where they affect interpretation; a rate alone can conceal a small or changing population.
Check rounding at the end of the calculation. Record unrounded values in the working record where available. Do not combine percentages with different denominators as if they were counts or average them without a method that fits the question.
Describe uncertainty without inventing precision.
Measurement can be incomplete, sampled, filtered, delayed or affected by collection changes. Name the known limitations and how they affect the claim. Distinguish an observed change from a supported explanation and a proposed hypothesis.
A small observed difference may not justify a broad assessment. Do not add a confidence interval, statistical significance claim or causal label without the method and data needed to support it. An unexplained score is not a substitute for that evidence.
Use wording that matches the scope: what the supplied records show, for which period and under which conditions. When the records are insufficient, hold the comparison or publish a narrower factual statement.
Check permitted use as well as accuracy.
A calculation can be correct while its publication is unauthorized or misleading about the people represented. Check permission to use client records, personal information, screenshots, quotations and supplied datasets in the intended context. Remove or mask unnecessary identifying details and seek the appropriate responsible review when access rights are uncertain.
Record who supplied the material, what use was permitted and any limits on quoting, aggregation or attribution. Private access is not automatic permission to make a public case study. Generated substitutes must not be presented as real customer evidence.
Semantec SEO’s claims and substantiation policy sets its own business controls. This lesson explains a reporting method without converting those house duties into promises by other organizations.
Make the final statement reproducible.
Keep metric, values, denominator, dates, formula, source record, qualifiers, owner and review state together. Have another reviewer reproduce the arithmetic and compare the wording with what the records establish.
Use source selection for the claim record and original evidence for recording a test. Return to the evidence and useful contribution overview or learning overview.