Locate the decision carried by the words
Start with a place where language changes what someone chooses. This might be a menu label, a field instruction, or a confirmation prompt. Record what the team intends the person to understand and what action should follow. A wording preference alone is not yet a research question.
The surrounding screen matters. A short label may work when its context is clear and fail when the same words appear in a different part of the workflow.
Ask for interpretation before explanation
A review can ask people to describe what they expect a label or action to mean before revealing the intended behavior. That account is more useful than asking whether the words are clear after the researcher has explained them.
Use task material that is realistic enough to support a decision, while avoiding unnecessary operational information. The plan should identify which parts of the interface are being examined and which context is intentionally held constant.
Compare alternatives for a reason
Alternative wording should represent a meaningful difference: the object affected, the scope of the action, or the expected next step. Changing several unrelated features at once can make it difficult to understand why a response changed.
A comparison can consider interpretation and task behavior together. A person may say that they like a phrase but still choose an unintended action. Record both observations rather than treating preference as a complete measure of understanding.
Turn observations into a usable revision
The output should identify the wording, the observed interpretation, and the consequence for the task. A recommendation might clarify a label, add context, or show that language alone cannot resolve a missing product distinction.
A language inventory can also note related screens where a term needs consistent use. The scope should make clear which material was examined. A finding about one workflow should not be presented as proof that the same wording will work for every audience and situation.