Measurement and diagnosis
Measure the task you intended to help.
Traffic can show that people reached a page. It does not, by itself, show that they understood the answer, completed an action or became a suitable customer. Define the question before choosing the measure.
Free learning from Semantec SEO.
Separate coverage, visibility and outcomes.
Coverage asks if the page answers its assigned task with supported information. Search visibility asks how the page appears in the available search records. A reader outcome asks what a person could understand or complete. A business outcome asks what happened after a useful action.
These questions need different evidence. An editorial review can identify a missing explanation without measuring search performance. Search Console can report impressions and clicks without proving that a visitor understood the page. A project enquiry can arrive without establishing which content caused it.
Start with a decision: retain an explanation, repair a contact route, investigate a change or collect better evidence. Record the relevant page, audience, dates, source and owner. Do not build a report merely because a dashboard exposes a number.
Define each state and the evidence it needs.
A measurement plan names an observation before choosing an event or outcome. A count of page views, people, sessions, messages or accepted projects has a different unit. State the population, period and source so another reviewer can understand what the number actually represents.
The dictionary below defines possible states. It is not a claim that tracking is installed, that records have been collected or that a person moved through every state. Unknown evidence remains unknown. Do not infer a later state from an earlier one.
| State and unit | Evidence needed | What it cannot establish |
|---|---|---|
| View: recorded page view | A defined collection source, page identity, observation window and stated coverage. Report recorded views separately from distinct people. | That every visit was recorded, that the reader understood the answer or that one view represents one person. |
| Relevant interaction: recorded action | The exact trigger, such as opening a particular disclosure, and a test that the event corresponds to that action. | That the interaction was useful, intentional in every case, or evidence of comprehension. |
| Understood answer: assessed response | A participant response to a stated task, compared with predefined answer criteria and recorded with the test conditions. | Population wide success, later business impact or understanding of topics outside the tested question. |
| Opened draft: action to open a draft | An observed or appropriately recorded attempt to open the email draft, with the actual route and collection limits. | That a mail application opened successfully on every device, that a message was sent or that the business received it. |
| Sent message: message transmission | Evidence from the actual sending service or an appropriate sender record, under the agreed access and privacy limits. | Receipt by the business, suitability, acceptance or completed work. |
| Received enquiry: received message | A permitted receiving record with the agreed time window and duplicate handling. | That the enquiry is valid, that a project fits or that a particular page caused it. |
| Valid enquiry: reviewed qualifying message | The business’s agreed fit criteria applied by an authorized reviewer, with duplicate, spam and unsuitable states recorded separately. | An accepted project, contract, delivered change or revenue. |
| Accepted project: accepted engagement | The actual approval or agreement record and its accepted scope, subject to appropriate access. | That work was delivered, a recommendation was implemented or a later result occurred. |
| Completed work: accepted delivery | A scoped delivery or release record and the relevant acceptance checks. | Search processing, customer benefit, revenue or causal impact without their own evidence. |
Coverage review remains a separate editorial question: does the resource answer its assigned task? Separate topic coverage from reader outcomes when deciding what this measurement can support.
Hypothetical plan: a free resource.
Imagine a free explanation intended to help a business owner distinguish a topic from a query. This is a proposed measurement plan, not collected user research or a report of results.
The learning question is: can a reader apply the distinction to a new example? A task test could ask participants to identify the subject and the wording used to find it. Record the prompt, answer criteria, sample and the circumstances of the test.
An event recording a link click can show that a recorded click occurred. It cannot establish that the explanation was understood. Time on page can be influenced by distraction, reading speed or an idle tab. Treat such signals as clues to interpret, not substitutes for the learning question.
If an analytics event is proposed, define its trigger and permitted fields before deployment. A later finding should identify the data source and collection period. Until collection happens, report the event as proposed or not run.
Test whether the reader can apply the distinction.
A proposed comprehension task needs a question with an assessable answer. In the existing fictional free resource example, ask a reader to apply topic and query to a new situation rather than to report whether the article felt helpful.
The exercise below is a manually written teaching plan. No participants have been recruited, no answers collected and no success rate measured.
- Task prompt
- “A person wants to learn how to maintain a bicycle and types ‘how to clean a bicycle chain’. Identify the wider subject and the exact search wording. Explain why the two are related but not interchangeable.” This is an invented task, not an observed search.
- Answer criteria
- The response identifies bicycle maintenance as the wider subject, identifies the supplied wording as the query and explains that one topic can lead to several differently worded questions. Accept a sound explanation in the reader’s own words rather than requiring a memorized sentence.
- Participant context and sample
- Record the intended audience, relevant prior knowledge, recruitment method and actual number of participants. Those values are NOT ACQUIRED for this plan. Do not label a convenience sample representative of all visitors.
- Method
- With permission, present the resource and then the same task prompt. Record the answer and the conditions needed to interpret it. Decide in advance whether help is allowed and what counts as a partial answer; keep partial, incorrect, unanswered and unassessable responses separate.
- Interpretation and next check
- A misunderstanding in a small test can identify a candidate explanation to improve. It does not prove the same frequency among the whole audience. Preserve the actual prompt, response basis and limitations before proposing another test.
A click on a related article may indicate interest or confusion. Time on the page can include an idle tab. Neither replaces this application question or proves that the distinction was learned.
Hypothetical plan: a project enquiry.
A service page has a different task: help an appropriate visitor describe work and request a discussion. A useful enquiry definition might require a working contact route, relevant project information and a way for the business to assess fit.
Distinguish opening an email draft, sending a message, receiving it and accepting a project. A mail link can open a draft without proving that the person sent it. A successful form submission can still be spam, a duplicate or an unsuitable request.
Record the business’s agreed definition of a valid enquiry. If later contact records are available, identify who can review them and how they are linked lawfully to the measurement question. Do not collect message contents or personal data merely because a reporting tool permits it.
Delivery is another state again. A proposal accepted by a client does not prove that the recommended website change was made. Keep release records, checks and later outcomes distinct.
Keep receipt, fit and acceptance separate.
The business must define a valid enquiry before a reviewer counts one. For an explicitly fictional website review service, a qualifying message could contain a reachable reply route, an identifiable website and a question within the agreed service scope. A real business must approve its own criteria; these are not Semantec SEO’s measured acceptance rules.
A duplicate message is not an additional distinct enquiry. Spam can be received without qualifying. An unsuitable request may be genuine but outside the service scope. Record these dispositions separately and name the person permitted to review the receiving records.
Keep the contact route’s observable states aligned with the interface. A draft opening, a sending attempt, a receiving record and a fit decision need their own evidence. Define contact and confirmation states before counting the interface’s success wording as proof of receipt.
State the denominator before calculating a rate.
Fictional arithmetic only: suppose an approved measurement record contains 10 distinct valid enquiries attributable under its stated matching rule to 200 eligible recorded visits during the same illustrative window of 30 days. The chosen enquiry rate is 10 ÷ 200 × 100 = 5 percent. These are stipulated teaching values, not site, Search Console or Semrush measurements.
- Numerator
- Ten distinct messages that were received and passed the defined relevance and contact criteria in the window. Exclude spam, duplicate messages and requests outside the stipulated scope. Do not replace this with all attempted submissions.
- Denominator
- Two hundred eligible recorded visits to the stated service page in that same window. In this fictional definition, exclude staff tests, known automated traffic and visits outside the stated page or period. State whether repeat visits are counted; this example counts eligible visits, not distinct people.
- Matching and collection limitation
- The example stipulates an approved matching rule. If the permitted records cannot connect qualifying enquiries to those visits reliably, do not calculate or label this as a page enquiry rate. Show the two scoped counts separately and record the missing basis.
- Different questions
- Valid enquiries divided by all received messages measures the share that qualifies. Valid enquiries divided by all site visits uses a wider population. Neither is interchangeable with the 5 percent example, and none proves that the page caused the enquiry.
Write the collection plan before collecting.
Name the decision, the minimum information needed and the actual source that could supply it. A proposed event remains proposed until its collection is authorized, implemented and tested. Writing an event name here does not install analytics, activate tracking or request a new permission.
Define consent dependencies, access and a retention decision with the accountable organization before release. This lesson does not determine a universal legal basis or retention period. Collect no message contents, email addresses or other personal details merely to fill a dashboard.
- Proposed event
- Use a clear internal name such as resource_related_link_click for a click on a specified learning route. Define exactly which page and link qualify. The name is an example specification, not evidence that an event exists.
- Trigger and allowed fields
- Record the intended action and minimum permitted fields, such as a stable resource identifier, a named link identifier and a suitably scoped collection period. Review the actual URL and identifiers for personal information before permitting them. Exclude message text and contact details from this example event.
- Source and coverage
- Name the collection system and its actual coverage once known. Rejected consent, blocked scripts, repeated events and device differences can affect what is recorded. A missing event is not automatically proof that no action occurred.
- Owner, access and retention
- Assign an accountable measurement owner and an authorized reviewer. Record the agreed access, retention period and disposal method; all remain unresolved in this proposed example until the organization accepts them.
- Release test
- In an authorized environment, perform the exact action, inspect whether it creates one correctly scoped record and test a nonqualifying action. Check excluded fields, consent states and failure behavior. Record browser, date, expected and actual result. Until tested, mark the event NOT RUN.
Use privacy and behavioral evidence to limit what is necessary and distinguish observation from inference. A proposed review schedule does not create an unattended monitoring service.
Read the available search records with their limits.
The Search Console Performance report documentation explains impressions, clicks and grouping choices. The query dimensions documentation describes privacy omissions and missing rows.
Record the property, filters, search type, dates and grouping. A query sample is not a complete transcript of visitor intent. An absent row can reflect reporting limits rather than zero demand.
Compare equivalent periods when appropriate and note changes in campaigns, seasonality, availability, tracking and the website itself. A rise after publication does not establish that the page caused the rise. A fall can also have more than one explanation.
When the baseline is missing or the collection method changed, state the limitation. It may be more useful to establish reliable measurement than to offer a confident before and after judgment.
Compare records at the same scope.
Keep the property, search type, dates, filters, page grouping and query omissions beside a Search Console observation. A keyword provider estimate describes a different source and cannot be substituted for an observed impression, click or business outcome.
Illustrative incompatible slices: one report covers mobile web clicks to a single service page during June; another covers impressions for the whole property across all devices during July. Dividing those values does not produce a meaningful ratio of clicks to impressions for either slice. Match the population, dimensions and period before calculating a ratio, and use a measure’s actual definition.
Query omissions also limit interpretation. An absent query row does not establish zero demand or a complete account of visitor intent. Read query and page observations with their reporting scope. If a visibility change could reflect discovery, crawl access or indexing, interpret that evidence first.
Record baseline quality, seasonality, campaigns, availability, collection changes and other releases before comparing periods. Keep favorable, unfavorable and inconclusive observations. A later rise or fall alone cannot isolate the page’s effect. Plan a test with guardrails when a controlled comparison is appropriate.
Turn findings into a bounded next action.
For each observation, write what was seen, what it may mean, what remains uncertain and the next check. A broken contact destination can justify a repair. An uncertain interpretation may justify more research. Neither requires an invented numerical score.
Keep positive, negative and inconclusive findings. Define who can approve a change, who will make it, and how a harmful change can be reversed. A planned review date is not a continuous monitoring service.
This resource introduces measurement and distinguishes learning, enquiry and delivery outcomes. Detailed experimentation, attribution, indexing evidence and performance diagnosis retain their focused scope. Use experiment guardrails before planning a test.
Return to the learning overview. For commissioned analysis, describe the decision and the records available.
Return a plan with a bounded next action.
These two fictional records demonstrate the different evidence obligations. They contain no observations of a real audience or business. The owner role identifies a responsibility to assign; it does not identify a person already commissioned to do the work.
| Task and decision | Source and measure | Limits and collection responsibility | State and next check |
|---|---|---|---|
| Free resource: decide whether the topic/query distinction needs clearer explanation. | The stated comprehension prompt and answer criteria; count assessed correct, partial, incorrect and unassessable responses separately, with the actual denominator of assessed responses and test period. | No participants or period acquired. A small convenience sample cannot establish a population rate. The assigned research owner must agree permission, minimum notes, reviewer access and retention before testing. | NOT RUN. Agree the task, audience and sample method; collect permitted evidence, then inspect which distinction remained unclear in those responses. A click count does not complete this task. |
| Service enquiry: decide whether the contact route needs repair or the scope explanation needs clarification. | Permitted sending/receiving and records from fit review, plus a separately scoped route test. Count received, duplicate, spam, unsuitable and valid enquiries by their agreed definitions and window. Calculate a rate using visits only if matching evidence supports it. | No contact records or attribution basis supplied. The assigned business owner approves fit criteria; an authorized reviewer of receiving records applies them. Access, consent dependencies and retention remain to be agreed; no message contents are needed in a public report. | NO DATA; collection and route tests NOT RUN. Verify the actual contact destination and receipt behavior, agree fit review, and keep any later acceptance or delivery record separate. |
For each eventual finding, preserve what was observed, what it might mean, what remains uncertain and who owns the next check. A planned review date is a responsibility in a record, not evidence of a running service. For commissioned analysis, discuss the available measurement records and the decision they need to support.
Choose the next evidence or method task.
The local explanation remains the starting point. These resources address distinct next tasks; this page does not establish that the complete learning branch has been reviewed or accepted.
- Coverage versus visibility metrics
- Review whether the assigned answer is complete without treating editorial coverage as a measured reader outcome.
- Topic and cluster health
- Inspect missing answers and learning routes across related pages before interpreting their combined health.
- Experiments and causation
- Define the comparison, baseline and competing explanations before attempting a causal interpretation.
- Prioritization with incomplete data
- Explain which repair should come first when the available evidence is incomplete.
- Privacy and observed behavior
- Limit collection to the agreed question and keep observed actions distinct from inferred personal intent.
- Search Console Query and Page Evidence: Read the Scoped Record
- Retain property, dates, filters, dimensions and omissions when reading a search record.
- AI Search Observation and Reporting: Evidence Without Guarantees
- Record the actual response, source trace and uncertainty rather than promising stable AI visibility.
Read URL observations in context.
Use the technical access and indexing resource to distinguish discovery, crawl access, indexing and dated inspection evidence before interpreting a change.