Sopact is a technology based social enterprise committed to helping organizations measure impact by directly involving their stakeholders.
Copyright 2015-2026 © sopact. All rights reserved.
Training evaluation survey questions for every Kirkpatrick level. Pre and post examples, behavior-anchored prompts, and the question architecture funders accept.
Most training surveys stop at reaction: a 4.3 out of 5 that answers a question no funder or sponsor is asking. This page is a set of prompts you paste into Claude, ChatGPT, or Gemini to draft questions across all four Kirkpatrick levels, make them pre/post matched, and turn the results into evidence. A prompt writes the questions in a minute. The same person answered twice, linked by one ID, is what makes them count.
The short answer
Two definitions to keep straight before you write a single item — then the four levels every strong evaluation covers.
Training evaluation survey questions are the items used to measure whether a training program worked — not just whether people liked it. A complete set runs across four Kirkpatrick levels: reaction at session end, learning as paired pre/post scenarios, behavior at 30 to 90 days, and results as a tied operational metric. Most surveys ask only the first level and call it an evaluation.
Reaction, Learning, Behavior, and Results. Reaction asks how the experience landed; Learning measures what changed in knowledge or skill, baseline to exit; Behavior tracks what people actually did differently on the job; Results ties the program to an operational outcome a sponsor cares about. Each level uses a different question format and a different cadence.
Build it live
A question bank gives you items to pick from. These give you a working session. The order matters — each prompt builds on the answer before it.
Start from what you teach. A sentence about the course is enough for the tool to draft both levels.
Create a course evaluation survey with Likert items and open-ended questions, mapped to Kirkpatrick Level 1 (reaction) and Level 2 (learning). My course is [course name and topic] for [audience], running [length]. For Level 1, use 1-to-5 Likert items on relevance, clarity, and confidence, each paired with one open-ended "what produced that rating." For Level 2, write three to four short scenario questions that test applied understanding, not recall.
The delta is only real if the same person answers the same item twice on a locked scale. This sets that up.
Turn the Level 2 scenarios into a matched pre/post pair. Use the exact same scenarios and the same rubric at intake and at end-of-program, and keep every Likert scale locked at 1-to-5 across both waves. Explain how I'd score the per-person delta, so the change is a real measure rather than a scale artifact.
Behavior is what people did on the job. Anchored counts beat self-rated frequency, which just measures personality.
Add a Level 3 behavior follow-up to run 30, 60, and 90 days after training. Use anchored-count questions tied to a specific application moment — "in the past 30 days, how many times did you use [skill] where it applied" — not self-rated frequency scales. Pair each with one open-ended question surfacing the barriers that stopped people from applying it.
Level 4 is not a survey question. It is an operational metric with a date range and a comparison cohort, defined now so attribution holds up later.
Help me define the Level 4 results measure. It is not a survey question — it is an operational metric that existed before training, with a date range and a comparison cohort. For my program, the outcome that matters is [describe it]. Propose one or two tied metrics I could pull, the source system for each, and the cadence, defined now so the attribution is auditable later.
A number with no reasoning behind it produces an average no one can interpret. This adds the why, and a rubric to theme it.
For every Likert item in my survey, write the paired open-ended question that asks what produced the score, worded to get a specific anchor rather than "it was good." Then give me a short coding rubric — four to six themes — I could use to tag the open-ended answers consistently across the whole cohort.
Leading questions, double-barreled items, scale drift, and outcomes with no instrument behind them. This finds and fixes them.
Review my whole survey for the things that quietly break training evaluation: leading or double-barreled questions, vague items that can't be acted on, scale drift between pre and post, and outcomes I'm claiming with no instrument behind them. Rewrite each weak item and tell me in one line what was wrong with the original.
Collection was never the bottleneck — analysis was. This computes the delta and themes the open-ends in one pass.
I've collected responses. Here is the data: [paste or describe it]. For each Level 2 scenario, compute the pre-to-post delta per participant and flag anyone who didn't move. Theme the open-ended answers using the rubric, and tell me which themes show up most among the people whose scores didn't improve.
The four-level architecture transfers; the scenarios and the Level 4 metric do not. This swaps them for your setting.
Rewrite this question bank for [sales enablement / clinical training / a nonprofit workforce program / compliance training]. Keep the four-level structure, but swap in scenarios, application moments, and a Level 4 metric that fit that setting. Flag any level that doesn't apply to a short, single-session format.
The honest part
A GenAI tool can write every question across all four levels — that is the easy 5%. The other 95% is what links one person's answers across waves so the numbers mean something.
What the prompt gives you — the 5%
What a data system gives you — the 95%
Without the shared ID, the same participant is two anonymous rows in two separate exports — and the delta cannot be computed at all. The ID is the difference between a 4.3 average and proof that this person changed.
A prompt writes the questions. A record links the answers. Five hundred post-surveys you read ten of prove nothing. The same person answered at intake and again ninety days later, held together by one ID, with the reasons themed beside the scores — that is the evidence a sponsor accepts.
Your next evaluation
A prompt drafts the four-level survey in a minute. Sopact Sense is where those questions become a connected learner record — pre to post to behavior to results, one ID holding it together.