The short answer
To find out which parts of a keynote actually landed, you need three things lined up on one timeline: a record of what you said and when, a minute-by-minute measure of how the room responded, and what people remembered afterwards without being prompted. Applause and a satisfaction score cannot do this, because both describe the talk as a whole and both are dominated by how it ended. The work starts about four weeks out, when you write down the three to five moments you want the audience to take away and decide how you will know each one worked. On the day you capture a timecoded recording, a room signal such as an observer log, live polls or an engagement timeline, and a short open-question pulse before people leave. Within 48 hours you mark each moment as landed, partly landed or missed, and two weeks later a one-question recall email tells you what stuck.
Four weeks before: decide what “landed” means
A moment can only land or miss against an intention, so write the intention down while the talk is still being built. List the three to five moments you most want the audience to carry out of the room — a line, a story, a demo, a number, a call to action — and for each one write a single sentence that finishes “we will know this landed if…”. Good tests are observable: the room goes quiet, people photograph the slide, the poll answer shifts, someone quotes the line back two weeks later. Then decide which signals you will capture, because some need lead time. Anything that relies on attendee consent, such as emotion analysis, has to be written into event registration before registration opens, which for most conferences means now. Ask the organiser at this stage what they already collect for your session, such as app check-ins, poll data and room counts, and whether you can have it afterwards.
- Write: three to five target moments, each with a one-line “landed if” test.
- Ask the organiser: what session data they collect and whether you will get it.
- Book anything that needs consent: opt-in wording must be in registration before it opens.
One week before: build a timestamped run sheet
The analysis afterwards is only as good as your ability to say what was happening at minute 17, so turn the finished talk into a run sheet. Rehearse it once against a timer and record the planned minute and slide number of every section change and every target moment. Mark the places you expect a response — a laugh, a reveal, a question to the room — because an expected response that does not arrive is one of the most useful things you will learn. Confirm that the session will be recorded with a visible clock or timecode, and if the venue is not recording, arrange a phone on a tripod at the back. Write a three-question pulse to run in the final minute: which moment do you remember most, what one idea will you use, and what was unclear. Open questions beat a one-to-five rating, because a rating tells you how people felt about the talk, not which part of it did the work.
- Run sheet: minute, slide number and expected response for every target moment.
- Recording: timecoded, from walk-on to the last question.
- Pulse: three open questions behind a QR code on your final slide.
On the day: capture three signals
On the day your only job beyond speaking is to make sure three signals are captured against the same clock. First, the recording, started at walk-on so its timecode matches your run sheet. Second, a room signal. The simplest is a colleague in the back third of the room who notes, every two minutes, what they can see: phones out, notes being taken, laughter, people leaning forward or drifting out. Live polls add a data point wherever you ask a question, and an audience engagement timeline gives you a measured curve for every minute and every seat rather than one person’s impressions. Third, the pulse. Put the QR code on the final slide and ask people to answer before they stand up, because responses fall away quickly once the room empties into the coffee break. Finally, write down anything that changed from the plan — a section cut for time, a demo that failed, a late start — while you still remember it.
If you use EchoDepth Events: the Standard tier measures the room from anonymous behavioural signals — footfall, dwell, movement against stillness and optional audio response — with no consent forms. The Deep tier adds emotion signals such as confusion and delight but requires delegate opt-in through registration, which is why it belongs in the four-weeks-out step. See keynote engagement analytics.
Within 48 hours: line the signals up
Do the analysis while the talk is still fresh, and do it moment by moment rather than for the talk as a whole. Put the run sheet, the room signal and the recording side by side on one timeline. For each target moment, ask three questions: did the room signal rise or hold at that minute, did the recording show the response you expected, and did anyone name that moment in the pulse without being prompted? Two out of three is landed, one is partly landed, none is missed. Then look for the stretches you did not plan to examine: any run of three minutes or more where the signal fell away, and what you were saying at the time. These drift points are usually not the weakest material; they are more often a transition, a long setup or a slide with too much on it. End with three lists — keep, fix and cut — and a single sentence on what you would change first.
| Target moment | Room signal | Recording | Named in pulse? | Verdict |
|---|---|---|---|---|
| Opening story (min 2) | Held | Laugh as expected | Yes, 11 mentions | Landed |
| Market number (min 9) | Rose | Quiet, phones up | No | Partly landed |
| Framework slide (min 17–21) | Fell for 4 min | Read the slide aloud | “Unclear” x6 | Missed — fix |
| Call to action (min 38) | Rose | Clear, on time | Yes, 7 mentions | Landed |
An illustrative worksheet. Use one row per target moment, plus a row for every unplanned drift point.
Two weeks after: the recall test
The strictest test of whether a moment landed is whether anyone still remembers it a fortnight later without being reminded. Two weeks after the keynote, send one short email to attendees, through the organiser if you do not hold the list, with a single question: what is the one thing you remember from the talk? Do not list your points, and do not offer multiple choice, because recognition is far easier than recall and will flatter every section. Count which target moments come back unprompted and in whose words. A moment that scored well in the room but never appears here was entertaining rather than memorable, while a moment that appears here despite a flat room signal was probably doing quiet work worth protecting. Compare the result with your 48-hour verdicts, update the keep, fix and cut lists, and file the whole pack with the run sheet so the next version of the talk starts from evidence rather than memory.
What each signal can and cannot tell you
| Signal | What it tells you | What it cannot tell you | Effort |
|---|---|---|---|
| Applause and the feel of the room | Overall warmth at the end | Which minute did the work; heavily shaped by the close | None |
| One-to-five session rating | Satisfaction with the talk as a whole | Which part landed; ratings cluster near the top | Low |
| Open-question pulse | What people remember as they leave | What happened in minutes nobody mentions | Low |
| Live polls | Participation and opinion at poll moments | Anything between polls | Low |
| Observer log | Visible attention cues every two minutes | More than one person’s view of part of the room | Medium |
| Timecoded recording | What you actually said and when | How the audience responded | Low |
| Engagement timeline | A measured, minute-by-minute curve across the whole room | Why a minute worked — you need the recording for that | Low on the day; needs booking |
| Two-week recall email | What actually stuck | Anything from people who do not reply | Low |
Questions speakers ask
Can I do this without any technology?
Yes. A timecoded phone recording, a colleague keeping a two-minute observer log and a three-question pulse on paper or a free form will give you a usable moment-by-moment verdict. What technology adds is coverage: one observer sees part of the room, while a measured timeline covers every seat for every minute.
Is it legal to measure the audience?
Polls, pulses and observation are ordinary event feedback. Anonymous behavioural measurement that stores no personal data does not need consent; anything that analyses individual emotion does, and must be set up through registration. Our GDPR guide for event analytics explains the line.
What if the organiser will not share session data?
Then rely on what you control: your recording, your observer and your pulse. Ask again afterwards with a specific request, such as the poll results and the room count for your slot, which organisers can usually share even when attendee-level data is off limits.
For the measured version of this, see what an EchoDepth keynote engagement timeline looks like or book a 30-minute demo on real session data.
Published 27 September 2026.