Claude research uses observation to catch what interviews smooth over. Guests retell a purchase as a decision. In the room it is often a queue: a badge, a form, a person who is not at their desk, a tab that will not load, a phrase repeated until someone senior arrives. Those details rarely survive until the debrief. They matter because friction is frequently a handoff, not an opinion.
What a visit can support
A visit can support claims of this shape: in this setting, at this time, people did these things in this order, and this utterance recurred. It can support a comparison between what a brochure implies and what the floor required, if you saw both. It can give you better questions for the next decision-maker interview, because you are no longer guessing which artifact exists.
A visit cannot support “most buyers,” “the industry is moving,” or any rate. You did not draw a sample. You stood somewhere. Say where. If you saw three booths, you saw three booths. The discipline is the same one pattern finding applies to talk: repetition is interesting, and it is still local.
Where it is legitimate to stand
Useful scenes are ones you were invited into or that are openly public. An implementation call where the customer asked you to listen. A renewal meeting you joined as a note-taker with consent. A trade-show aisle. A lobby while you wait for a meeting that is on the calendar. A screen-share in which a buyer walks you through the procurement portal they already use, narrating what they click. Each of these can be written without sneaking.
Illegitimate scenes are private channels you were not given: a forwarded inbox, a Slack you were added to “just to look,” a badge you borrowed to wander a closed hall. Claude listening is not covert research. If the scene would embarrass you in the memo’s first line, do not enter it. The ethics notes in interview design apply here as well: consent, a plain statement of purpose, and no harvesting of personal data that the study does not need.
Limit
If you cannot name the door that let you in, you do not have a scene. You have a rumor.
How to watch a touchpoint
Pick one touchpoint and stay with it longer than is comfortable. At a booth, do not count badges. Watch the first ninety seconds of three conversations you can hear without intruding: who speaks first, whether a product is touched, whether a meeting is booked, whether the visitor leaves with nothing in their hands. In a lobby, watch what people do with the ten minutes before they are called: which document they reopen, whether they rehearse a sentence. On a renewal call, note the minute the talk leaves price and enters risk, and the artifact that is screen-shared when that happens.
Write in the present tense while you are there, then rewrite the same day into the six lines in the margin of this page. Separate what you saw from what you think it means. “The visitor folded the one-pager without reading it” is an observation. “They do not care about collateral” is already a theory, and it may be wrong; they may have read it on the walk over.
From a scene to a memo
A claude observation memo is short. It names the door, the sequence, two objects, one repeated sentence, and one absence. Absences are easy to invent, so restrict yourself to an absence you can anchor: you expected a security engineer in a pilot review because the invitation listed one, and the chair stayed empty. That absence is evidence about that meeting. It is not evidence that security is “disengaged industry-wide.”
Bring the memo to the interview rather than to the positioning slide. Ask the guest to correct it. If they say the empty chair is normal because security reviews a packet asynchronously, you have learned something about the path. That correction belongs beside the original note, not instead of it. Both are part of the record that later positioning synthesis is allowed to use.