That inference reflects a possible locus of attention, not operational intent

What it is

Even when “probably looking at this region” is right, it is not “intends to do something to this region.” Locus of attention is a spatial estimate of processing resources; operational intent is a decision to take responsibility for an outcome. People stare at Delete to check how dangerous it is, at an ad because it flashes, at an already-rejected option to exclude it once more. Writing attention measurement as intent treats observation as authorization. It complements “the fixation point is not attention”: that leaf splits looking from attending; this splits attending from wanting to act.

Why it happens

Intent usually passes through decision and action preparation, and sometimes leaves a second trace on hand, voice, or dwell-confirm. Attention measurement stops a layer earlier. Reading the case against, comparing prices, checking for errors, all pile long AOI dwell on objects that will not be chosen—processing is deep, intent is negative. Salient regions attract attention without attracting intent. Waiting fixations park on any stable point; that is the eye resting, not a request for that point.

So attention measurement can predict “where they are more likely to look next, what they are more likely to recall”; the lift on “what they will click next” is limited, and on comparison and exclusion tasks it will systematically point at the wrong object.

Studying it

Collect AOI dwell and final choice together, and ask specifically whether the longest-dwelled region is the chosen one. Comparison, exclusion, and reading a warning are good materials for the split: the longest fixation often sits on the unchosen item or on the warning text. Add a delayed choice: look first, click allowed a few seconds later, and see whether the measurement still predicts after the delay. A usability heatmap with no choice data beside it cannot make this distinction. Do not use only a “looked then bought” funnel; that is a conditioned success sample.

Where it stops holding

On extremely familiar, almost automatic clicks (always the same switch), attention and intent coincide in space and the measurement looks like intent. The moment the task includes comparison, risk, or a new layout, the coincidence splits. When disabled users use gaze as their only command, measurement and command share a trace and must be split in protocol: which samples are measurement, which are commit. Ads and recommendations that treat attention as intent reward salience manipulation, not need.

Applying it

• Do not use the highest AOI alone as “what the user wants”; when you need intent, take a confirm, or at least cross with choice history.
• On comparison and warning UIs, treat long dwell as “processing,” default not “about to commit.”
• Verify on a task that requires excluding one item to choose another, and report the rate at which the longest AOI disagrees with the final choice.

Related

• Same group: C8.13.1 Gaze traces can infer currently attended UI regions without explicit action · C8.13.3 Using implicit attention to adapt the UI requires a clear boundary so it is not treated as command input · C8.13.4 Retention and use of implicit measurement data should differ from user-initiated interaction data
• Adjacent: C8.01 Fixation and saccade · C8.03 Midas touch in eye input
• Search: intent · attention versus action · AOI