1. The device is nearest the loudest person
Every room has someone everyone strains to hear, and they are the person the transcript will lose. Put the microphone nearer them, not in the geometric middle of the table.
If you are wearing the device, this means sitting nearer the quiet speaker rather than opposite them.
2. Air conditioning, and everything like it
Continuous background noise — an air conditioning unit, a fan, traffic through an open window — does not sound loud to a person because we filter it out unconsciously. A microphone cannot, and every word competes with it.
A meeting room with the AC on full is the single most common reason a recording in this region comes back poor. Turning it down for the twenty minutes that matter is worth more than any hardware.
3. Two people talking at once
Speaker separation works by finding the boundaries between voices. Overlapping speech has no boundary, so both speakers get merged or one gets dropped.
This is unavoidable in a real conversation, and it is why the useful output is a summary rather than a verbatim record. But the moments that matter — a decision, a number, a commitment — are worth saying into a gap rather than over someone.
4. The phone on the table doing something else
A phone that is recording and also receiving notifications is a phone whose microphone is being interrupted. On some devices a notification sound is recorded at full volume directly over whatever was being said.
This is a real advantage of a separate device, and we would say so even if we did not sell one.
5. Nobody said what the meeting was
A recording with no opening context produces a summary with no framing. Ten seconds of “this is the handover call for the Marina project, with Ahmed and Sara” gives the summary a subject, the speaker labelling two names, and you something searchable in six months.
It is the cheapest improvement on this list and the one people most reliably skip.