Readiness scores don't prove adoption stuck
Three months after a Copilot or ChatGPT Enterprise rollout, someone on the leadership team asks a simple question: is this actually sticking, or did people try it twice and go back to old habits? You need AI adoption survey questions that check what happens after launch, not before it.
Most teams already ran a pre-launch readiness survey: are people trained, willing, confident. That's useful for planning a rollout. In our view, a readiness score nobody repeats after go-live is really a pre-launch mood check, not an adoption measure, and treating it as one misleads the board.
Once people have real exposure to the tool, the question changes from "will you use it" to "are you still using it, and for what." That calls for a different, shorter survey built around actual use, not intentions.
More than three-quarters of respondents say their organization now uses AI in at least one business function, according to McKinsey's July 2024 Global Survey on the state of AI. That tells a board AI has landed somewhere in the company. It says nothing about whether your finance team, your support desk, or your design group kept using it past week two.
Who to survey, and when after go-live
Send this survey to everyone who had access to the tool, not just the early champions who show up in every demo. Include managers separately: their answers reveal whether they're modeling use or quietly sticking to the old process themselves.
Send the first wave four to eight weeks after go-live, once people have had enough exposure to form a real opinion, then repeat it around the 90-day mark. If you're mapping out your AI adoption survey for 2026, put both checkpoints on the calendar before the rollout even starts.
We'd rather see one short, consistent survey run twice than three different one-off polls. A single comparable wave tells you almost nothing; a repeated one shows you a trend line, which is the only thing worth presenting to a board.
The AI adoption questionnaire: four things worth asking
These are the AI adoption survey questions worth building into the questionnaire once people have had real exposure to the tool. Four themes cover most of what a Head of L&D needs: how often people use it, what it replaced, what's blocking it, and what people believe it changed.
| Theme | What it checks | Example question | Pair it with |
|---|---|---|---|
| Usage frequency and depth | How often and how deeply people use the tool | In the past two weeks, how often did you use the tool? | Login or usage logs, if available |
| Task substitution | Which tasks moved to AI versus still done manually | Which of these tasks have you handed to AI: drafting, summarizing, coding, research, other? | Task completion records or workflow audits |
| Perceived barriers | What's blocking fuller use | What stops you from using AI more: time, trust, training, access, policy? | Help-desk tickets, training attendance |
| Perceived productivity and confidence | Self-reported time saved and confidence | How much time do you believe AI saved you on a typical task last week? | Manager observation, output review |
On task fit, HeySurvey's post-implementation survey guide reports that a study of 112 users found perceived usefulness mattered about 50% more than ease of use in predicting whether people kept using a system after launch. Ask what AI replaced, not just whether people liked using it; a tool people enjoy but don't need will fade fast.
On barriers, keep the question open-ended: "what stops you from using AI more?" with options like time, trust, training, access or policy. The answers decide which intervention you run next. AI Adoption Score's barriers survey, the AI Adoption Barriers Survey, is built around exactly that question.
On perceived productivity, expect a gap between what leaders assume and what employees report. Windmill's HR guide, published in January 2026, cites McKinsey's Superagency research finding that C-suite leaders estimate only 4% of employees use AI for 30% or more of their daily work, while the actual share is estimated at roughly three times higher.
Expect the same self-report gap here: see the limits below before you present any "time saved" number as fact.
The questions below are editorial suggestions to adapt, not a validated instrument; word them for your own tools and roles before sending.
Sources: 31 Post Implementation Survey Questions, Measuring AI Adoption: Survey Questions for HR
How to read the response scale, without overtrusting it
Keep response options simple and consistent across waves. Use a short frequency scale for usage questions, listed below, and a five-point scale from "not at all" to "a great deal" for confidence and usefulness questions rather than a vague 1-to-10 slider.
- Frequency: daily, a few times a week, a few times a month, rarely, never
- Confidence: not at all, a little, somewhat, mostly, completely
- Open barrier question: short free text, capped at one or two sentences
We're skeptical of teams that average these scores into one adoption number and stop there. An average of five-point confidence ratings hides the fact that one department never adopted at all while another lives inside the tool daily; look at the spread before you look at the mean.
Where self-reported answers run out of road
Self-reported use and actual logged use don't always line up. A 2021 meta-analysis of 106 effect sizes in digital-media-use research found that self-reported use correlated only moderately with logged data, and that self-reports were rarely accurate; treat survey answers as perception, not proof, and check them against usage data where you have it.
Where you have login counts, usage logs, or task-completion records, compare them against the survey answers rather than trusting either alone. A gap between what people report and what the logs show isn't noise to explain away; it's a signal worth asking about directly.
We still think the survey is worth running even with that gap. It's the only channel that tells you why usage is low, not just that it is, and a usage log can't ask a follow-up question.
Repeat the check, don't just run it once
Run this as the counterpart to whatever readiness survey you used before launch, so you can see what changed between "ready" and "actually using it." Keep the core question set identical across waves; swap in new questions only for things you've already decided to fix.
The right AI adoption survey questions for a 90-day check are the same ones you used at 30 days, plus whatever follow-up your barrier answers pointed you toward. That consistency is what turns one survey into a trend line a board will actually trust.
Stop asking whether people feel ready. Start asking what they actually did this week, and keep asking it on a schedule.
Frequently asked questions
How soon after launch should we send a post-rollout AI adoption survey?
Send the first wave four to eight weeks after go-live, once people have had real exposure to the tool, then repeat it around the 90-day mark using the same core questions so you can compare waves instead of reading one-off results.
Should we survey people who never used the tool?
Yes. Employees who never logged in are often the clearest sign of a blocked rollout, whether the blocker is access, training or confidence; leaving them out of the sample hides exactly the gap a Head of L&D needs to see.
Can we reuse our pre-launch readiness questions after rollout?
Not as they stand. Readiness questions ask about intentions and confidence before anyone used the tool, while a post-rollout check needs to ask about actual frequency, task substitution and friction, which readiness questions aren't built to capture.
How do we know if a 'time saved' answer is reliable?
Compare it against output records, help-desk data or manager observation where you have them. As a practical check, see whether the same person gives a similar answer when you ask again in the next wave, rather than treating one answer as final.
