Five reputation measurement mistakes, starting with the displayed rating

Reputation management measurement mistakes often begin with a label that promises more than its underlying record can show. This non-ranked list identifies five errors that an England business can check in a review report without assuming any particular software supplier has made them.

What to take away

Selection method and boundaries

The research cut-off is 6 September 2026. Entries were chosen by reading the authoritative records cited below. Each must show a distinct reporting error tied to its definition or stated limitation.

Inclusion requires a practical correction and a source readers can inspect. The scope is review and response reporting for England operations. Global platform documentation is labelled as such; UK research stays within its original geography.

The list excludes provider rankings, unsupported frequency claims, vendor performance promises and allegations against businesses. Ordering follows the reporting process: rating, interaction, attribution, evaluation, then external comparison.

Numbers identify entries rather than severity; selection did not rest on payment or promised favourable coverage. Publisher commercial relationships are unverified and need disclosure review before publication; this is a documentary checklist, not a prevalence survey or product test.

1. Treating every average as the displayed rating

Google's review-score record, checked at the cut-off, defines the score from published ratings. An analyst's filtered monthly mean may answer a useful question, but it is a different series.

Keep the capture date and platform beside the displayed figure, give a recalculated measure its own title, describe included records and preserve the denominator. This entry covers Google's global definition and supplies no England rating target.

2. Counting a call click as a completed conversation

The authoritative Business Profile performance page defines its calls measure as call-button clicks. That event cannot confirm whether anyone answered.

Correct the label in the dashboard. Obtain a separate operational record for connected calls if the decision requires one. Preserve unmatched events instead of assuming every click becomes a customer. This product definition was checked on 6 September 2026 and is not a study of English callers.

3. Calling allocated credit incremental sales

Google's attribution record, accessed on the research date, describes models that assign credit across recorded paths. A change in credited share after switching models is not a change in completed sales.

Re-export the same event population under a fixed convention and identify the model visibly. Keep claims about incremental effects outside that allocation table until an appropriate evaluation supports them.

4. Claiming impact from a before-and-after chart

HM Treasury's analytical methods annex, from the May 2026 Magenta Book update, sets out methods using comparison conditions to estimate effects. A period change alone leaves competing explanations unresolved. Record concurrent changes and ask an evaluator whether a credible counterfactual can be established. Where it cannot, describe the movement and its timing without assigning a causal amount to review handling.

5. Applying product-review research to local services

The Department for Business and Trade's fake-review study record, published on 25 April 2023, explicitly excludes services from its findings. A local service firm cannot use those product-review estimates as its own expected fake-review rate.

Attach category and geography limitations to any citation, and omit the numerical comparison when populations do not match. The research remains useful context for deciding what needs investigation.

Apply the checklist to a report

For each affected chart, record the current claim, the observable record and the corrected wording. Assign someone to repair the definition or source mapping. Review waiting responses separately from answered cases, so an apparently improved timing statistic cannot hide unresolved work.

Recheck the revised export against its input records before circulation. Keep a change note that tells readers which earlier conclusions should no longer be used.

Before you act

Common questions

Why is a recalculated average not the same as the displayed rating?

Google's review-score record defines the score using published ratings. An analyst's filtered monthly mean may answer a useful question, but it is a different series. Keep the capture date and platform beside the displayed figure, give the recalculated measure its own title, and preserve the denominator.

What does a call-button click actually measure?

The Business Profile performance page defines its calls measure as call-button clicks. That event cannot confirm whether anyone answered. Correct the label in the dashboard and obtain a separate operational record for connected calls if the decision requires one, preserving unmatched events rather than assuming every click becomes a customer.

Can a before-and-after chart prove that review handling caused a change?

No. HM Treasury's analytical methods annex sets out methods using comparison conditions to estimate effects. A period change alone leaves competing explanations unresolved. Record concurrent changes and ask an evaluator whether a credible counterfactual can be established. Where it cannot, describe the movement and its timing without assigning a causal amount.