Your last workshop scored 4.6 out of 5 on the feedback form, and three months later nobody can find the framework in their daily work. The problem is rarely the trainer; it is the training evaluation questions you asked. Questions about comfort collect applause. Questions about application predict behavior change.
The money at stake is not trivial. US organizations spent $102.8 billion on employee training in 2023 — an average of $874 per learner (Training magazine, 2023 Training Industry Report) — and the commonly cited Ebbinghaus forgetting curve holds that people lose about 50% of new material within 24 hours without reinforcement. Only 13% of employees are engaged at work worldwide (Gallup, State of the Global Workplace 2023) — the long-run price of development that stops at satisfaction.
This guide hands you a 25-question training evaluation survey built on Kirkpatrick's four levels, the timing and scoring rules that make answers usable, and the mistakes that make evaluation theater.
Key Highlights
- An evaluation survey should measure reaction, learning, behavior and results — satisfaction alone predicts nothing.
- US training spend hit $102.8 billion in 2023, an average of $874 per learner (Training magazine, 2023).
- People lose about 50% of new material within 24 hours without reinforcement (Ebbinghaus, commonly cited).
- Only 13% of employees are engaged at work globally (Gallup, 2023) — the cost of development that never reaches behavior.
- The bank below holds 25 questions: 5 reaction, 5 learning, 7 behavior, 5 results and 3 action.
- Reaction surveys go out within 48 hours; behavior questions wait 30–60 days (illustrative cadence).
Training Evaluation Questions: The Direct Answer
A training evaluation survey earns its keep when every question maps to a level of evidence. Kirkpatrick's four levels — reaction, learning, behavior and results — are the standard map, and Phillips added a fifth that asks whether the program paid for itself. Most organizations collect level 1 and stop; the bank below walks up all five.
The design principle is simple: ask about observable events, not feelings. Did the person apply the skill? Did a manager see it? Those answers survive a budget review.
A 4.6 satisfaction average does not. The table below distributes the 25 questions across those levels.
| Level | What it measures | Questions | When to ask |
|---|---|---|---|
| 1. Reaction | Relevance, clarity, intent | 5 | Within 48 hours |
| 2. Learning | Skill and confidence gained | 5 | Last day, re-checked at day 30 |
| 3. Behavior | On-the-job application, observed | 7 | Day 30–60 |
| 4. Results | Business metric movement | 5 | Day 60–90 |
| 5. ROI | Was it worth the cost | 3 | End of quarter |
Level 3 carries the most questions because it predicts everything else: if behavior never changes, the reaction score was entertainment and the results line stays flat. The level mechanics live in our Kirkpatrick model guide.
Key takeaway: weight your questions toward behavior — five questions per level beats twenty questions about comfort.
Why Satisfaction Scores Waste the Budget
The smile sheet is the default because it is easy, and useless for the same reason. Satisfaction correlates with catering as much as with learning. A cohort can love every session and change nothing — 13% of employees are engaged worldwide (Gallup, 2023) while training budgets grow.
Unmeasured spend compounds quietly. At the national average of $874 per learner (Training magazine, 2023), a 30-person cohort is a $26,220 purchase that deserves evidence. Typical 2026 US ranges we see quoted (illustrative): $50–$150 per person per year for off-the-shelf e-learning, $200–$400 a head for a one-day custom workshop, $800–$1,600 for a multi-week cohort and $3,000–$6,000 for a three-month coaching engagement. The midpoints sit below.
None of those numbers is provable without a measurement plan agreed before launch. In our work with a 90-person roofing company, the supervisor program scored 4.7 while crew turnover stayed flat for two quarters — the survey was measuring lunch, not lift.
Key takeaway: an evaluation that only measures satisfaction is a review of the venue — price the measurement plan into the program, not after it.
The 25 Training Evaluation Questions That Predict Behavior Change
Here is the bank. Never send all 25 at once — a survey longer than five minutes dies in inboxes. Three rules govern the wording: ask about observable events, include one manager-reported item in the behavior wave, and put open-text questions last.
Most items are 1–5 scales. Questions 5, 10, 17, 22, 24 and 25 are open text, question 11 is yes or no, question 12 is a count and question 19 is multiple choice.
| # | Question | Predicts |
|---|---|---|
| 1 | This session addressed a problem I face this month | Intent to apply |
| 2 | I know exactly what I am expected to do differently | Clarity for transfer |
| 3 | The facilitator made it easy to ask questions | Environment quality |
| 4 | The pacing kept me engaged | Attention and retention |
| 5 | What is the one thing you will try first? | Written commitment |
| 6 | I could complete the core task without help by the end | Real skill built |
| 7 | The practice exercises matched my actual work | Transfer design |
| 8 | I understand why the method works, not just how | Depth under variation |
| 9 | Rate your confidence to perform this skill today | Day-30 baseline |
| 10 | What part of the material is still unclear? | Next-session fix |
| 11 | I have applied the skill at work at least once | Application started |
| 12 | How many times did you use it in the past two weeks? | Habit frequency |
| 13 | My manager has seen me use the new method | Observed, not self-reported |
| 14 | The skill fits my real workflow without workarounds | Design fit |
| 15 | Something at work made the skill hard to use | Barrier, reverse-scored |
| 16 | I know where to get help when I get stuck | Support structure |
| 17 | What made applying it harder than expected? | The fix list |
| 18 | The skill has already saved time or reduced errors | Leading result |
| 19 | Which metric does this skill touch most directly? | Business line mapped |
| 20 | Customers or internal partners have noticed the difference | External evidence |
| 21 | I would recommend this program to a peer in the same role | Perceived value |
| 22 | What result do you expect this skill to move this quarter? | Results target |
| 23 | Was the time invested worth the outcome so far? | ROI judgment |
| 24 | What would you cut, and what would you double? | Redesign input |
| 25 | Who else on the team needs this program next? | Next-cohort pipeline |
Question 15 is reverse-scored on purpose: agreement signals a problem, so its answer belongs in the fixes column, not the average. Question 13 beats any self-report because managers observe behavior participants forget. Questions 12 and 19 are the two your CFO will ask about first.
In our work with a 140-person logistics firm, the behavior wave found the top barrier was system access, not skill — the practice environment used a login the floor did not have. One question, one fix, and application scores moved within a month (our observation, not a controlled study).
Key takeaway: steal the wording, not just the count — observable events, one manager-reported item and open questions last are what make answers usable.
Timing and Response Rates: When to Ask
Timing decides whether the answers are honest. Reaction questions go out within 48 hours while memory is intact; learning confidence gets re-checked at day 30; behavior questions wait 30–60 days; results questions run at 60–90 days. Asking a behavior question on the last day measures optimism, not behavior.
Response rates decay faster than most L&D plans assume; the illustrative pattern below matches what we see in mid-market teams. Send the same day or lose a third of the responses, and keep each wave to five to seven questions.
| Wave | Questions | Send | Owner |
|---|---|---|---|
| Reaction | 1–5 | Within 48 hours | Coordinator |
| Learning re-check | 9 | Day 30 | Manager |
| Behavior | 11–17 | Day 30–60 | Manager |
| Results | 18–22 | Day 60–90 | Participant and finance |
| Action | 23–25 | End of quarter | Program owner |
The pre and post pairing matters as much as the timing: confidence at day 30 means nothing without the day-zero baseline, and metric movement means nothing without the starting value. The full method lives in our pre and post assessment guide.
Key takeaway: same-day sends and 30-day behavior checks — timing errors cannot be fixed by better questions afterwards.
Scoring the Answers: Thresholds That Trigger Action
Scores without thresholds are decoration. Set the action rule before launch: reaction below 4.0 of 5 on relevance means the objectives were wrong; behavior below 50% applied at day 30 means the reinforcement failed; results flat at day 90 means the module goes back to design.
Behavior scores respond to follow-up more than to delivery quality, the finding most teams miss. The illustrative model below is what we see when manager check-ins run weekly versus when they do not: the day-90 gap is roughly threefold.
When behavior stalls, the fix is usually reinforcement design, not a better trainer. Spaced refreshers, application assignments and manager check-ins keep skills alive; the design levers live in our learning transfer guide. Report the delta per cohort — averages bury the teams that need help.
Key takeaway: decide the thresholds in writing before launch — a score that does not trigger an action is just a number.
The Mistakes That Make Evaluation Data Lie
Six mistakes account for most useless evaluation data. Surveys run too long, so completion collapses. Every question is a Likert agreement scale, so nothing observable gets captured. Behavior questions arrive before anyone has tried the skill, results get averaged across cohorts, non-responders get ignored, and nobody set thresholds.
Length is the quiet killer. The illustrative model below shows completion collapsing as question counts grow, which is why the bank runs in waves of five to seven questions. The missing responses on a long survey are not random — they are the people who needed the fix.
Non-response bias deserves special attention: the people who skipped the behavior survey are disproportionately the ones who never applied the skill, so glowing answers at a 60% response rate may still be a failing program. Chase non-responders once, then report the rate next to every score.
Key takeaway: short waves, observable wording and a chased non-responder list — the hygiene that keeps evaluation honest.
What Your Business Should Do
Days 1–30: Draft the survey before the next session.
- Write five questions for the level you currently ignore, usually behavior.
- Cut any question you cannot act on within a month of receiving the answer.
- Pilot the survey with five employees and check it takes under five minutes.
Days 31–60: Run the first full measurement cycle.
- Send the reaction wave within 48 hours of every session, same day if possible.
- Ask each manager to rate one observable behavior per participant at day 30.
- Record the starting value of one business metric the program should move.
Days 61–90: Act on the answers and re-measure.
- Fix the top two barriers the behavior wave surfaces, starting with question 15.
- Re-run behavior and results questions at day 60 and report the delta per cohort.
- Redesign or retire any module whose behavior scores never move.
Key takeaway: ninety days and one honest measurement cycle will tell you more than a year of satisfaction averages.
The Bottom Line
Training evaluation questions decide whether your budget produces evidence or applause. US companies spent $102.8 billion on training in 2023, about $874 per learner, and people lose roughly 50% of new material within 24 hours without reinforcement — a spend that size deserves measurement that reaches behavior. Weight the survey toward observable events, send it on the 48-hour and 30-day clock, and set thresholds that trigger fixes rather than debates. If you want this mapped to your own team, book a free strategy call or start with the free resource library.
Work With Zohaib Butt
1. Free Resource Library — Zohaibbutt.com/store
World-class training and coaching resources, all in one place. Download the assessments, templates and playbooks you need today.
2. Online Courses — learning.zohaibbutt.com
Self-paced leadership and coaching programs for managers and owners who want structure, not scattered tips.
3. Free Strategy Session — funnelsthatsell.com
Book your complimentary strategy call and we will map your biggest people-development bottleneck to a 90-day fix.
Free assessment pack (PDF)
Build a survey that predicts behavior change instead of collecting applause.
Download the PDF packFAQ
What are the best training evaluation questions to ask?
Ask questions that map to Kirkpatrick's levels: relevance for reaction, skill checks for learning, observed application for behavior and one business metric for results. This article's bank splits 5 reaction, 5 learning, 7 behavior, 5 results and 3 action. The behavior block does the heavy lifting, because questions 11–17 predict whether anything else moves.
How many questions should a training evaluation survey include?
Five to seven questions per wave is the practical ceiling, because completion collapses as length grows: 70% completion at five questions falls to 25% at twenty-five in the illustrative model. Never send all 25 at once; split them by level and timing. Reaction goes out within 48 hours, behavior waits 30–60 days, results run at 60–90 days.
When should you ask behavior questions after training?
Ask behavior questions 30–60 days after the program, once people have had real chances to apply the skill. Same-day behavior questions measure intention, which is close to worthless — people lose about 50% of new material within 24 hours without reinforcement (Ebbinghaus, commonly cited). Pair the day-30 check with a manager rating on one observable behavior.
What is Kirkpatrick's four-level evaluation model?
Kirkpatrick's model defines four levels of training evaluation: reaction (did people find it relevant), learning (did skills and confidence rise), behavior (did on-the-job application change) and results (did a business metric move). Phillips added a fifth level asking whether the program paid for itself. Most organizations only measure level 1, which is why satisfaction looks high while nothing changes.
