Design a Level 1 reaction survey as the first step of a connected evaluation — every response tied to the participant's ID so reaction links forward to learning and behavior.
In short: A training feedback survey is the Kirkpatrick Level 1 (reaction) instrument — but a good one is not a standalone smile sheet. Design it as a short, six-question reaction set (relevance, content, facilitator, pace, likelihood to recommend, plus one open-ended "most and least useful") and tie every response to the same persistent participant ID you used at enrollment. Because reaction sits on the same record as learning and behavior, you can later see whether a low reaction score predicted drop-off or a weak Level 2 gain.
Reaction fatigue is real: the longer the smile sheet, the fewer honest answers you get at the end. Cover the drivers that actually move satisfaction and stop. Ask five closed questions on a consistent scale — relevance to the participant's job, content quality, facilitator effectiveness, pace, and likelihood to recommend — and exactly one open-ended question. A tight set completes in under two minutes and still gives you everything a driver analysis needs.
Use the same 1–5 scale for all five closed items so the responses are comparable and can be averaged into a single reaction score. Relevance to job asks whether the training applied to what the participant actually does. Content quality and facilitator effectiveness separate the material from the delivery — a common confound when scores are low. Pace catches sessions that ran too fast or too slow. Likelihood to recommend is your one summary loyalty signal. Keeping the scale identical is what lets Sopact Sense compute a per-participant reaction score you can rank and correlate.
The single most valuable question on a training feedback survey is open text: "What was most useful, and what was least useful, about this session?" The stars tell you the score; the comment tells you why. This is the field Sopact Sense codes into themes — content, facilitator, relevance, pace — so you learn which driver is pulling satisfaction down, in the participant's own words, not just that it is down.
Every reaction response must carry the persistent participant ID from enrollment — not a fresh anonymous submission. This is the design decision that turns a smile sheet into the first wave of a connected evaluation. With the ID in place, a participant's reaction sits on the same record as their Level 2 pre/post learning scores, their Level 3 behavior follow-up, and the Level 4 result. Sopact Sense treats the ID as the join key, so "did people who rated the facilitator poorly also fail to change behavior?" becomes a query, not a manual spreadsheet merge.
Write the reaction items with the later levels already in mind. Relevance-to-job at Level 1 foreshadows the transfer question you will ask at Level 3 ("were you able to apply this?"). Content quality points at the specific competencies your Level 2 pre/post test will measure. When the reaction survey is designed as the opening move of the same instrument family — same ID, aligned constructs — you avoid re-contacting participants to reconcile mismatched surveys later.
Once reactions are collected on the participant ID, paste this into the Sopact Assistant to turn raw feedback into a scored, themed, drop-off-aware read:
From this training feedback for [COHORT], code the open-ended responses into themes (content, facilitator, relevance, pace), give a reaction score per participant on their ID, rank the top two drivers of low satisfaction with a representative quote each, and flag reactions that predict drop-off.
Expected output: a per-participant reaction score joined to each ID; open-ended comments coded into the four themes with counts; the top two drivers of low satisfaction, each with a representative verbatim quote; and a flagged list of participants whose reaction pattern signals likely drop-off before Level 2. Input is the six-question reaction set on the participant ID; output is a scored, themed reaction table you can hand to the facilitator and carry into the Level 2 analysis.
GRADE: green | scored + themed + drop-off flags | needs a second cohort to confirm predictors; amber | scores + themes, weak drop-off signal | thin open-ended responses; red | average stars only | no ID, comments ignored, anonymous.
A Green result gives you a reaction score, coded themes, and drop-off flags all on the participant ID. Amber means you have scores and themes but too few open-ended comments to trust the driver ranking. Red is the classic anonymous smile sheet: an average star rating, no ID, and no way to connect reaction to anything that follows.
Read the comments, not just the stars. A 4.2 average hides which sessions and which drivers are weak. The open-ended "most and least useful" field, coded into themes, is where the actionable signal lives — always analyze it alongside the score.
Do not make it anonymous. Anonymity feels safe but it severs reaction from every later level. Without the participant ID you can never ask whether a poor reaction predicted no behavior change. Collect on the ID; protect confidentiality in how you report, not in how you collect.
Reuse the same scale you will use later. Keep the 1–5 scale consistent across the reaction survey and the constructs you carry into Levels 2–4, so scores stay comparable across waves and Sopact Sense can trend them on one record instead of reconciling different scales.
Five closed questions on a shared 1–5 scale — relevance to job, content quality, facilitator effectiveness, pace, and likelihood to recommend — plus one open-ended "what was most and least useful." Six questions total keeps it under two minutes and still supports a full driver analysis.
No. Tie every response to the persistent participant ID. Anonymity breaks the link between reaction and later learning and behavior, so you lose the ability to see whether a low reaction score predicted drop-off or a weak Level 2 gain. Confidentiality belongs in reporting, not in collection.
By sharing one participant ID and aligned constructs. Reaction sits on the same record as the Level 2 pre/post learning scores, the Level 3 behavior follow-up, and the Level 4 result, so Sopact Sense can correlate them directly — no manual survey merge required.
ChatGPT, Claude, and Gemini are fine for a quick test — but not for an answer you'll put in front of a funder or board. When it has to hold up, run it in Sopact Sense.
Try it in Sopact →
