AI fluency for UX researchers: separate observation and inference

Record what happened before explaining it

An observation should describe the action or statement that the session actually contains.

The fictional note says a participant opens the Reports page, pauses for twelve seconds and clicks Help. The moderator then points to Saved reports, after which the participant opens the requested file. The note does not say why the participant paused or whether they had seen the feature before.

The GOV.UK Service Manual recommends recording what was seen or heard separately from its interpretation. Apply that distinction here: preserve the click sequence and moderator intervention, then place possible explanations in another column. Do not let AI rewrite the pause as confusion unless the evidence supports that wording.

Sources: [3]

Keep assistance attached to the task outcome

Completion after a hint differs from unassisted discovery.

The participant eventually opens the file, but only after the moderator points to the control. A summary that says the participant found the saved report independently would lose a material part of the session. Record the assistance and where it occurred rather than reducing the attempt to completed or failed.

Ask AI to summarize the sequence in three sentences with the intervention included. Compare the result with the fictional note. Do not use the twelve-second pause as a general usability threshold or claim that every pause of that length signals a problem. It is one supplied observation.

Generate competing explanations without inventing evidence

A hypothesis helps choose a follow-up; it does not replace the missing observation.

Possible explanations include an unclear label, unfamiliar page layout or a mismatch between the assignment wording and interface terminology. The session does not settle which explanation is correct. Write each as a question to investigate and identify what evidence would distinguish it from the others.

AI may suggest follow-up questions, but it must not fabricate a participant quote or a second session to strengthen the finding. Keep synthetic variants explicitly fictional. If a new question asks what the participant expected to find, use it as a planned prompt, not as an answer already obtained.

Choose a design action proportionate to the finding

One session can motivate investigation without establishing the best redesign.

A reasonable next step in this exercise is to examine whether Saved reports is understandable and visible in the task context. Do not conclude that the entire navigation must be rebuilt. Describe the proposed change or test and the uncertainty it is intended to reduce.

Introduce another fictional session in which the participant opens Saved reports immediately but cannot recognize the file names. Keep that as a different observed difficulty. Combining both into users cannot find reports may conceal that the problems occur at different steps and need different follow-up evidence.

Hand over a traceable finding

The reader should be able to distinguish the session record from the researcher’s recommendation.

Deliver the observation sequence, assistance note, competing explanations and proposed follow-up. Attach each statement to the relevant fictional session. Preserve the difference between the first navigation difficulty and the later file-recognition difficulty instead of using a broad theme to erase the distinction.

Review whether the worker avoids invented motivations and retains the moderator’s contribution to completion. Customer insights work groups reported feedback across people; this UX task examines the sequence of an observed interaction and the limits of what that sequence can establish.

Sources and scope

NIST: AI risk management. Skills England: workplace AI foundations.

These references provide background, not validation or endorsement of this exercise. The case details, calculations and suggested review questions are original instructional material. Use them to discuss observable work, not to infer customer outcomes, professional credentials or performance in every setting. Before adapting the exercise, confirm the relevant facts, approved tools, data permissions and decision owners. If you change the case, revisit the expected answers and checks as well. These examples describe practice tasks, not a promise that a particular product includes the fictional features.

Sources: [1] [2]