New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
Training evaluation survey questions for every Kirkpatrick level. Pre and post examples, behavior-anchored prompts, and the question architecture funders accept.
A questionnaire labelled training evaluation usually contains end-of-session reaction items: how relevant was the content, how was the pace, how clear were the materials. Reaction questions are Level 1. The evaluation framework has three more levels above it. Each one calls for a different question shape, a different cadence, and a different decision tag. This page is the question bank: thirty examples across the four Kirkpatrick levels, the paired pre and post pairing rule, and the persistent-identity discipline that lets the levels connect on the same participant. Workforce training, clinical training, and pharma sales enablement examples throughout.
Kirkpatrick's four levels are not optional layers. They build on each other. A Level 4 result requires Level 3 application, which requires Level 2 learning, which the Level 1 reaction influences. Each level needs questions written for that level, asked at the cadence that level requires.
A questionnaire that only asks Level 1 reaction questions tests the first assumption and leaves the next three unmeasured. The questions for the remaining three levels are different in format, different in cadence, and run on the same participant identity to keep the chain connected.
The four-level pathway, originally Donald Kirkpatrick (1959), expanded with the New World Kirkpatrick Model (Kirkpatrick & Kirkpatrick, 2016). The participant identity discipline is the thread that lets the levels connect on the same person across instruments and time.
Every training evaluation question belongs at one of four Kirkpatrick levels. The level the question sits at determines the format, the cadence, and the decision it can feed. The four definitions below cover what each kind of question is and how it differs from the others.
A reaction question is asked at the end of a training session and scores how the participant experienced it: relevance, clarity, pace, application moment. Most are Likert-formatted on a one-to-five or one-to-seven scale, with one or two paired open-ended prompts asking what produced the rating.
Reaction questions are the easiest level to ask and the easiest to misuse. They reveal whether the session felt useful, not whether anyone learned anything. A questionnaire that only contains reaction questions is a feedback survey labelled as evaluation. The worked example on this page contains six reaction items per end-of-session form.
A learning question scores knowledge or skill change. The format that produces a real measure is paired pre and post: the same scenario or rubric-scored item asked before training and again at the end, on the same participant identity. The post score alone is not a Level 2 measure. The delta per person is.
Scenario items beat recall items. A one-paragraph case followed by an action question, scored against a published rubric, captures what the participant actually understood. Recall items capture what was memorized. The Level 2 question bank in the worked example uses six scenario items, scored on a four-point rubric, asked twice.
A behavior question asks whether the learning was applied on the job. It runs thirty, sixty, or ninety days after training, never at the end of training itself. Effective Level 3 items are anchored to specific moments: in the past thirty days, how many times did you use the intake protocol from training. Anchored counts beat self-rated frequency scales because the moment is concrete.
The application moment named in the end-of-training reaction question seeds the behavior question. The participant remembers what they committed to apply. Self-report items pair with manager observation when the program has manager visibility into job tasks. The Level 3 bank in the worked example uses five anchored-count items plus one open-ended prompt about barriers.
A results question scores whether trained behavior moved a downstream outcome. It is tied to an existing operational metric, not a new survey indicator. For a workforce training program: placement rate, retention at six months, employer satisfaction. For a clinical training: case-resolution time, patient outcome scores. For a sales training: conversion rate, deal size, retention.
Level 4 questions are usually pulled from operational systems three to twelve months after training, not asked in a survey. The "question" is a metric definition, a date range, and a comparison cohort. The worked example pairs five outcome metrics to the ninety-day cohort report.
A training evaluation question earns its place when it sits inside three connections: the decision the answer feeds, a paired open-ended item that explains the rating, and a persistent participant identity that links the question to the same person's answers across other waves. A question missing any of those three is what we call an Orphan Question. Most published question banks, even the well-written ones, are collections of orphan questions. The fix is structural rather than editorial.
Reaction, learning, behavior, results. Each level needs a different format, a different cadence, a different rubric. A questionnaire that mixes the four without level tags is a feedback survey pretending to be an evaluation. The tag tells the analyst which decision the answer can feed and which it cannot.
A 1-to-5 confidence rating without an open-ended counterpart is a number no one can interpret once the cohort closes. Place the paired open-end immediately after the rating, never at the end of the survey where dropout peaks and response quality collapses. Pre training survey questions and post-training survey questions pair ratings with open-ends symmetrically across both waves.
Identical wording, identical scale, identical response options. Pre, post, thirty-day, sixty-day, ninety-day. Lock the instrument at version 1.0 before Wave 1. A questionnaire designed to evaluate training reach loses comparability the moment any of those vary.
Facilitator review, curriculum cut, barrier intervention, funder- report line. Every question maps to one of these. Questions collected in case we need it produce the forty-seven-question instruments no one analyzes. Decisionless questions are the single largest source of survey fatigue and shallow questions getting answered first while hard ones get truncated.
Email addresses change. Names abbreviate. Participant-remembered access codes get lost. The identity has to be assigned at enrollment by the system and embedded in every personalized wave link. This is the only structural fix for the matching break that ends pre-and-post analysis three weeks into cleanup. No analyst process replaces an identity that was never assigned.
Level 3 behavior items and Level 4 results items require waves that run weeks or months after the program closes. Adding them as afterthoughts six weeks after completion produces fifteen-percent response rates and no baseline to match against. Commitment to follow-up is asked for at enrollment, alongside identity capture.
Every principle above prevents one or more of these. Each failure survives any question-bank template. The fix is architectural, not editorial.
Reaction data presented as evaluation. Satisfaction averaging 4.3 out of 5 and the funder still has no answer to the question they asked. Level 1 is the floor, not the ceiling.
Pre and post both exist. The identity chain between them does not. Thirty to forty percent of records fail manual matching by name and email. No cleanup recovers an identity that was never assigned.
Open-ended responses sit in a CSV column no one codes. The reasoning behind every number is collected and then discarded. Qualitative data that is not themed is effectively not collected.
Pre-survey used 1-to-5. Post-survey used 1-to-7. The delta is an artifact of instrument drift, not a measurement of change. Locked scales are a prerequisite, not a refinement.
Six choices come up every time someone writes a training evaluation questionnaire. The first column names the decision. The middle columns show the broken default and the working alternative. The last column names the consequence: what each choice decides about the data the questionnaire produces.
The first decision controls all the others. If you anonymize each wave, the pairing rule cannot fire, the anchored-moment rule cannot connect to a known learning baseline, and tied metrics cannot match a person to an outcome. Identity discipline is the spine that lets the rest of the matrix work.
A 240-participant workforce training program. Eight weeks. Outcomes: case-management certification and job placement. Below: the scenario, the instrument set, and thirty specific evaluation questions distributed across the four Kirkpatrick levels with the cadence each level requires.
We had a feedback survey running for three years. End-of-program scores stayed in the high fours. Job placement rates barely moved. The funder asked whether the training was actually working. The honest answer was that we had no idea, because we had only ever asked Level 1 reaction questions. The questionnaire we now run spans all four levels and it has changed which curriculum modules we keep.
Workforce training program lead, post-rebuild cycle.
Five instruments across the cohort lifecycle. One participant identifier carries across all five. Levels chain because the same person fills every form.
Each question lists the format, the scoring approach, and the decision the answer feeds. Items are written for case-management certification training, but the structure transfers to clinical, sales, and professional development training.
The thirty questions above are written once and run as a connected instrument set for every cohort. The pairing, the manager-observation join, the rubric coding, and the operational-metric pull are structural properties of the platform, not procedural steps an analyst repeats.
Workforce programs, capacity-building cohorts, and credential programs each carry different content and different funder questions. They all need the same three-wave question architecture to produce evidence. The question bank changes. Persistent participant identity, locked scales, and paired open-ends do not.
Funder requires pre-to-post evidence disaggregated by gender and site, plus ninety-day post-exit employment data. The question bank stays identical across cohorts. The architecture is what determines whether funder reports take three days or four minutes.
Google Forms for intake with no identity beyond row number. SurveyMonkey for post with separate login and separate export. Generic email for the ninety-day with responses untagged to participant. Analyst reconciles on name and email in a spreadsheet. Thirty to forty percent of records fail matching on the first pass.
Persistent participant identity assigned at enrollment. Every wave delivered through a personalized link tied to that identity. Open-ended responses themed as they arrive. Disaggregation by gender and site is a filter rather than a project. Funder report generates in minutes.
The participant is both an individual and an institutional representative. Funder asks whether the practice was adopted at the organizational level, not only whether the individual learned it. The question bank adds organizational-adoption items at ninety days and six months, alongside the individual-level items at end-of-program.
Each participant answered separately with no institutional link. When the individual leaves the partner organization, the record is lost. Adoption evidence reduces to anecdote from annual grantee reports. Funder asks which organizations adopted and the answer requires re-surveying the cohort eighteen months later.
Every participant tagged to the partner organization at enrollment. Organizational adoption is a six-month rubric- based survey on the same connected record. Open-ended responses themed for barrier and enabling-condition patterns. Which organizations adopted becomes a report filter, not a research project.
The credential holds value only if post-program rubric scores can be traced to the same participant's learning trajectory. Rubric scoring is the Level 2 anchor. Employment outcome at ninety days is the Level 4 proof. Cross-cohort comparison requires the rubric and the question bank to stay locked across quarters.
Rubric scored in a spreadsheet and the participant survey in a form tool. Matching rubric to survey requires assessor- provided identifiers. No cross-year comparison because rubric versions drift. Employer verification arrives as a PDF attachment that never gets linked back to the participant record.
Assessor opens the participant record and scores the rubric directly. Rubric version locked across cohorts and year- over-year comparison holds. Employer verification captured through a link to the same record. Credential value is auditable because every score traces back to a participant and an assessor.
Three program shapes, one question architecture. The question bank changes by program type. Persistent participant identity, locked scales, and paired open-ends do not.
A training evaluation question scores a training outcome at one of the four Kirkpatrick levels. Reaction questions score how the session felt. Learning questions score what knowledge changed. Behavior questions score what got applied on the job. Results questions score whether trained behavior moved a downstream outcome. The level the question maps to determines the format, the timing, and the decision the answer can feed.
A training evaluation questionnaire is a set of evaluation questions organized by the level each one measures. A real questionnaire spans more than one level. The end-of-program form covers reaction and learning. A separate thirty-day or sixty-day form covers behavior. A quarterly or annual report covers results, often pulled from existing operational data rather than a survey. A questionnaire with only Level 1 reaction items is a feedback survey under a different name.
Examples by level: Reaction, on a one-to-five scale, how relevant was today's session to your current work. Learning, before training, what is the first action you would take if a participant disclosed unsafe housing, and the same question asked after training. Behavior, in the past thirty days, how many times did you apply the intake protocol from training. Results, what is your three-month case-resolution rate compared to last quarter. The worked example on this page lists thirty specific questions organized into four Kirkpatrick levels.
Sample questions span four levels. Reaction Likert plus paired open-ended at the end of every session. Learning paired pre and post on the same knowledge items, run before and after training, with persistent participant identity so the delta computes per person. Behavior at thirty and sixty days, anchored to specific job moments. Results pulled from existing operational data three to twelve months post-training. Question wording for each level is on this page; the worked example contains thirty samples.
Training feedback is one of the four levels of training evaluation. Feedback covers reaction, what Kirkpatrick calls Level 1: did the session feel relevant, was the pace right, were materials clear. Training evaluation is the larger frame and includes learning, behavior, and results in addition to reaction. Most surveys labeled training evaluation only ask reaction questions. A complete evaluation has a different question set for each level and runs at the cadence each level requires.
Decide which Kirkpatrick levels you will measure. Write reaction items only after you have written the learning, behavior, and results items the higher levels need. Pair every Level 2 learning item across pre and post. Anchor every Level 3 behavior item to a specific application moment. Tie every Level 4 results item to an operational metric you already track. Keep persistent participant identity across all four waves so the levels connect on the same person. Total instrument set runs four to six pieces, not one form.
Level 1, reaction: how participants felt about the training. Level 2, learning: what knowledge or skill changed. Level 3, behavior: whether the learning was applied on the job. Level 4, results: whether the applied behavior moved a downstream outcome. Levels build on each other: a strong Level 4 result requires Level 3 application, which requires Level 2 learning. Most training evaluation questionnaires only cover Level 1, which is why the framework feels truncated when published reports try to claim impact from feedback alone.
Pre and post training evaluation questions are paired knowledge or confidence items run before training begins and after training ends, on the same participants. The pairing is the measurement: the post score alone says nothing about change unless the pre score is on file for the same person. Same wording, same scale, same scenarios, asked twice. Persistent participant identity is the discipline that makes the pairing work. Without it, you are comparing two anonymous group averages, which is not a Level 2 measure.
Level 3 evaluation questions ask whether the learning was applied on the job. They run thirty, sixty, or ninety days after training, never at the end of training itself. Effective Level 3 items are anchored to specific moments: in the past thirty days, how many times did you use the intake protocol from training. Self-report items pair with manager observation when possible. The application moment named in the end-of-training reaction question seeds the behavior question, so the participant remembers what they committed to apply.
Level 4 results questions ask whether trained behavior moved a downstream outcome. They are tied to an existing operational metric, not a new survey indicator. For a workforce training program: placement rate, retention at six months, employer satisfaction. For a clinical training: case-resolution time, patient outcome scores. For a sales training: conversion rate, deal size, retention. Level 4 questions are usually pulled from operational systems three to twelve months after training, not asked in a survey.
Knowledge testing belongs at Kirkpatrick Level 2 and works best as paired pre and post. Use scenario items rather than recall items: a one-paragraph case followed by an action question, scored against a rubric. Recall items measure what was memorized; scenario items measure what was understood. Five to eight scenario items, each tied to a learning objective from the training, asked before training and after. The same items, the same scoring rubric, the same participants. The delta per person is the Level 2 score.
Run two instruments, not one. Instrument A at end of session covers Kirkpatrick Level 1 reaction: five Likert items on relevance, clarity, pace, confidence, and application intent, with one paired open-ended prompt asking what one moment was clearest and what one moment was unclear. Instrument B is the post-training knowledge test, paired against an identical pre-test: six to eight scenario items scored against a rubric, plus two open-ended prompts asking the participant to apply what they learned to a case. Same participant identity across both instruments. Levels 1 and 2 connect because the same person fills both forms.
Pharma sales enablement post-training questions span all four Kirkpatrick levels. Level 1 reaction at end of session covers content relevance and pace. Level 2 learning is paired pre and post on six to eight clinical-product scenario items, scored against a compliance rubric. Level 3 behavior at thirty and sixty days asks how often the trained talking points were used in customer conversations and whether compliance review flagged issues. Level 4 results pulls from CRM: conversation count by product, prescription lift, formulary coverage. The question discipline is the same as any other training evaluation; the rubric and the operational metrics are pharma-specific.
The Orphan Question is a training evaluation question that sits without three connections: the decision it feeds, a paired open-ended counterpart that explains the rating, and a persistent participant identity that links the question to the same person's answers across other waves. Most published question banks are compilations of orphan questions. They can be well-written and still produce data that cannot answer the questions funders and boards now require. The fix is architectural rather than copy-edit: assign a participant identity at enrollment, pair every rating with an open-ended reasoning prompt, and tag every question to a specific decision it feeds.
A retrospective pre-test is a post-survey item asking participants to rate their pre-program state after training, alongside their current state: looking back now, how would you rate your confidence in the skill before the program began. It corrects for response-shift bias, the phenomenon where participants revise their pre-program self-assessment downward once they understand how much there is to know. Use it in addition to, not instead of, the actual pre-test. Pair both readings to measure both actual change and perceived change. The item belongs on the post-survey alongside the matched post-confidence item.
A post-training survey delivered at session end typically reaches seventy to ninety-five percent response. Follow-up rates drop at predictable intervals: thirty to fifty percent at thirty days, twenty to forty percent at sixty days, fifteen to thirty- five percent at ninety days, when the follow-up runs through generic email broadcast. Personalized links tied to the participant record substantially raise the ninety-day response rate because the recipient recognizes the context. Response rate is a function of identity discipline as much as instrument quality.
Sopact Sense ships with a training evaluation question bank organized by Kirkpatrick level, with paired pre/post structure already built in and decision tags named on every item. Persistent participant identity carries across the pre-training baseline, end-of-program reaction, end-of-program knowledge post, thirty-day behavior follow-up, and ninety-day results indicator. The instrument set acts as one connected record per participant rather than five disconnected forms. Built-in qualitative coding handles open-ended scenario responses without manual recoding.
The training evaluation cluster covers methodology, the Kirkpatrick framework, sister feedback work, and the pre/post pairing pattern that Level 2 questions depend on.
The thirty questions on this page are a starting set. The discipline that matters is which level each question maps to, whether the Level 2 items are paired pre and post, whether the Level 3 items are anchored to a named application moment, and whether the Level 4 items are tied to a metric you already track. Sopact Sense ships the question bank with that structure built in and the participant identity that connects the five instruments across the program.
Author: Unmesh Sheth, Founder and CEO, Sopact
