Hypothesis
The relationship your team wants to test between the action and observed problem.
Impact measurement
After a change, check what is different.
Compare observations before and after your team’s work. Check whether observed information changes and decide whether to continue, adjust or investigate further.
Your brand · periods compared
ComparisonActionLimitations
Has the update been reflected?
Example question
After the page update, does AI describe compatibility correctly?
What you read
Example check: after clarifying compatibility on the product page, repeat the same questions under the same conditions. Compare answers and sources to see whether the information appears correctly. The comparison shows what changed, without guaranteeing that your update was the only cause.
The next step
Record the change and compare answers and citations before and after, keeping the same conditions.
A before-and-after comparison does not isolate seasonality, changes in sources or other factors. An observed change alone does not demonstrate a causal effect.
What you get
Define what to observe first.
Choose two collections from the same set of questions, with compatible market, language, models and collection methods. The comparison places available presence, recommendation and citation measures side by side, showing numerators and denominators. Failed or unreviewed answers are excluded from both periods; they are not counted as brand absences.
- A precise topic
- Choose the aspect being addressed: support, quality, delivery or another documented experience.
- Comparison baseline
- Define the period, sources and inclusion criteria before reading the change.
- A record of the work
- Record completed activities and external events that may affect the result.
- Outcome criterion
- Establish which observation would help decide whether to continue, adjust or investigate further.
What you get
Two versions. A measure defined in advance.
For a landing page hosted on TrustInsider, compare two previously published versions with the same form and privacy notice. Define the objective, sample size and stopping rules before starting. Consenting visitors are assigned to either version with equal probability; results describe that consenting audience.
The measure counts new contact requests submitted within 15 minutes of a recorded exposure. Exposure records browser rendering, not the attention of a unique person. No winner is declared during collection, and no variant automatically replaces the page. The comparison does not establish an effect on AI visibility or advertising campaigns.
| Step | What you check | Usable result |
|---|---|---|
| Assignment | Consent and two predefined versions | 50:50 probability, version retained for the browser |
| Exposure | Acknowledgement after version rendering | One record per assignment |
| Attributed request | New contact within the 15-minute window | Unique count without attributing sales or AI visibility |
Hypothesis and baseline
Define the result before you act.
If you update a page to clarify compatibility, the first check concerns that explanation. If you improve support, you need evidence about response times and resolution. The indicator should follow the problem addressed.
The verification plan defines a baseline, intended action, observation window and interpretation criterion. For AI, it keeps questions, model and method comparable; for the website, it distinguishes visibility, visits and measured events. Stating the criterion in advance reduces the risk of choosing only the number that appears to improve.
Reference
Initial evidence and the conditions needed to repeat the comparison.
Outcome
The relevant measure and circumstances in which the result will remain inconclusive.
Levels of verification
Done. Observed. Yet to be proven.
Completing a task, observing a change and attributing it to the task are three different conclusions. A useful report makes them identifiable rather than reducing them to one success indicator.
Measurement connects the action to evidence of implementation and subsequent collections. When seasonality, campaigns or new sources may play a role, the report preserves these explanations. Controlled comparisons and causal analyses require an appropriate design: a simple before-and-after comparison does not replace them.
| Level | Evidence required | Permitted conclusion |
|---|---|---|
| Task completed | Applied version, date and verification of the affected content or process. | The action was executed under the documented conditions. |
| Change observed | Comparable collections and an indicator relevant to the topic. | The measure changed within the sample, subject to the stated limitations. |
| Effect attributed | A method capable of assessing alternative explanations and external factors. | The action’s contribution can be assessed only within that method’s limitations. |
Next decisions
Every outcome helps you choose.
A stable result may indicate that more time, a different check or another action is needed. Even an inconclusive outcome is useful when it clarifies which information is missing.
If a new observation still describes a feature inaccurately, return to the question and cited source: another passage on the page may need clarification. If the description is correct, record the outcome and choose the next issue to address. Keep the dates and conditions of both checks.
- Observed outcome
- Hypothesis to reconsider
- Plan decision
Related paths
Choose your next step.
Check your brand’s presence, compare results or choose content improvements: continue with the feature you need.
Explore every intelligence tool
Choose the tools to see where your brand is found and where it is missing.
ReadFrom question to action
From the question to checking the outcome: context, sources, hypotheses and action plans in the TrustInsider analytical workflow.
ReadA transparent method, not a black box
TrustInsider’s criteria for interpreting reputation and trust: provenance, samples, reviews and the limits of conclusions.
ReadFind the right answer
Find guides, FAQs and methodology to clarify indicators, coverage and access, with dedicated routes for data requests.
Read