CONVERSION

Session recording note synthesis

Use when you have watched a batch of session recordings and have messy notes to turn into patterns.

session-recording-note-synthesis.md

Download .md

You are synthesising raw notes from session recordings into patterns. Be conservative: a pattern needs repetition, not one memorable session.

RAW NOTES (one session per block): {{SESSION_NOTES}}
NUMBER OF SESSIONS WATCHED: {{SESSION_COUNT}}
WHAT I WAS LOOKING FOR: {{RESEARCH_QUESTION}}

Output in this order:
1. PATTERNS - each with: description of the behavior, number of sessions it appeared in out of the total, the page or step where it happened, and a plausible explanation. Only include behaviors seen in at least three sessions or ten percent of sessions, whichever is larger.
2. ONE-OFFS WORTH NOTING - interesting single sessions, clearly labeled as anecdote and not evidence.
3. CONTRADICTIONS - places where sessions behaved in opposite ways, and what might explain the split (device, source, intent).
4. WHAT THE RECORDINGS CANNOT TELL US - list the questions that need a survey, interview, or analytics query instead.
5. NEXT ACTIONS - separated into "fix directly, no test needed" and "worth testing".

Constraints: do not infer intent or emotion from mouse movement. Do not convert counts into percentages that imply more precision than the sample supports. If my notes are too thin to support a pattern, say the sample is insufficient rather than producing one anyway. No em dashes.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{SESSION_NOTES}}
{{SESSION_COUNT}}
{{RESEARCH_QUESTION}}

Getting a better result

Timestamp and label each note block by device and traffic source so it can find the contradictions.
Watch sessions that converted as well as ones that did not, or every pattern will look like a problem.
The "cannot tell us" section is usually the most useful part - it stops recordings being over-read.

Questions about this prompt

When do I synthesise notes rather than present the sessions themselves?

After watching a batch, when the temptation is to show the one session everybody will remember. The threshold is deliberate: a behaviour needs three sessions or ten per cent of the total, whichever is larger, before it counts as a pattern. Memorable single sessions still appear, labelled as anecdote rather than as evidence.

What do I need in front of me before running it?

One note block per session, the number of sessions watched and the question you were trying to answer. Label each block with device and traffic source, since the contradictions section works by finding the split that explains opposite behaviour. Include sessions that converted, or every pattern you get back will look like a problem.

What comes back, and which part is worth acting on?

Patterns with session counts, one-offs marked as anecdote, contradictions, what the recordings cannot tell you, and next actions split into fix directly and worth testing. Section four is the one to read twice. It lists the questions that need a survey or an analytics query, which stops recordings being stretched past what they show.

What is the mistake that costs me here?

Reading intent or emotion off mouse movement. The prompt refuses to do it and so should you, because a wandering cursor means nothing reliable. The other trap is thin notes: if the sample cannot support a pattern you will be told so, and rewriting the notes to force one just launders a guess into a finding.