Why an AI staff survey needs more than a direct question
Ask a team 'are you using AI tools?' in an email from the managing director and you will get a tidy answer. A few people will say yes, with a careful explanation. Most will say no or not reply. It will not be the true picture.
The reasons are ordinary. The question sounds like the lead-in to a rule. People worry that admitting a shortcut will mark them as lazy, or that admitting a lot of use will mark their job as replaceable. The person who built a clever spreadsheet with a chatbot's help does not want it taken away. The person who has not touched any of it is afraid of looking behind.
An AI staff survey only works if people believe the answers are safe. Everything below is in service of that.
Neutral wording: ask about work, not about AI
Questions that ask about the work and the tools used to do it get fuller answers than questions that name a technology people have feelings about. Neutral wording also avoids leading people toward a particular answer. 'How much time do you waste on this?' invites a complaint. 'How much of your week does this take?' invites a figure.
Here are some before and after pairs.
Avoid double-barrelled questions, such as 'How do you feel about AI and automation?' People do not know which half to answer. Avoid loaded words like 'risky' or 'time-wasting'. And offer 'none of these' and 'prefer not to say' options, so nobody is forced into an answer that is not true.
- Instead of 'Do you use ChatGPT?', try 'Which websites, apps or tools did you use last week to get a piece of writing started?'
- Instead of 'Are you worried AI will take your job?', try 'How do you feel about changes to how your work gets done?' with a scale and a free-text box.
- Instead of 'Do you enter client data into AI?', try 'Which of these kinds of information do you handle in your tasks?' and separately, 'Which tools do you use for those tasks?'
- Instead of 'Which AI training do you need?', try 'What would make the tasks that fill your week quicker or less fiddly?'
Anonymity, and what 'anonymous' has to mean
Promising anonymity is easy. Delivering it takes a few decisions. The biggest is the smallest group you will ever report on.
If you show results for a group of one, you have named a person. Two is nearly as bad, because each knows the other's answer and can subtract their own. A minimum of three people per reported group is a widely used rule of thumb. It is a judgement call rather than a legal line, and a higher number is safer in a small business where people know each other well.
Small businesses face a specific problem: slicing. A 30-person firm with a team of two in finance cannot fairly report 'finance' separately. The fix is to merge small groups into larger ones, such as 'operations and support', or to report only for the whole organisation. Watch combinations too. 'Part-time, accounts team, over five years' may describe exactly one person even if each filter alone looks safe.
Free-text answers are the other leak. A comment can identify the writer through a specific incident or a turn of phrase. Show comments without the group label, or summarise themes instead. And think about the plumbing: if the survey requires a personal login that the owner can see, it is not anonymous whatever the cover note says.
If you cannot meet the standard you are promising, promise less. 'Your manager will not see your individual answers' is honest. 'Totally anonymous' that is not is a promise you will break.
Explain the arrangement to staff in a sentence they can verify: 'Results are shown to the owner as groups of at least three people. Nobody sees an individual's answers.' Specific beats warm.
What to ask
Offer a list of common tool types plus a free-text 'other'. People recall categories better than brand names, and 'other' is where the interesting answers live.
Keep the survey short enough to finish on a phone in under ten minutes. A workable shape:
- Role and context: what is your role, and what does a normal week look like?
- Tasks: which tasks fill most of your week, and which are repetitive or fiddly?
- Tools: which tools and websites do you use for them, including any you set up yourself?
- Information handled: what kinds of information pass through your hands?
- Attitude: how do you feel about change at work, and what would help?
- Needs: what is missing, unclear or in the way?
Policy, systems and insurance questions belong with the owners and managers, not staff. Keeping them separate stops staff from feeling that they are being asked to account for decisions that were never theirs.
What to tell people beforehand
Say what the survey is for in plain words, and keep it true. 'A short survey about how work really gets done here and which tools help or get in the way' is neutral and honest. Do not call it something it is not, and do not dress it up as a wellbeing survey if it is not one.
Then say what happens to the answers: who sees them, in what form, and what they will not be used for. If you can say 'the results will not be used to decide who is needed in a role', say it, and only if you mean it. Write it down in the invitation so you cannot quietly retreat from it later.
Tell people how long it takes, that it works on a phone, and when it closes. Have the owner or a manager send the invitation, but do not hover. Do not ask people in person whether they have done it.
Staff are not fools. If the survey is suspiciously vague, people will assume the worst. A plain purpose statement and a stated promise do more for honesty than any clever framing.
Running it well
Avoid the end of the financial year, the week of a major deadline, and the week after any restructure. Open for about a week, send one reminder, and thank people on close. If response is low, ask why before pushing. Low response in a small team is itself information about trust.
Be careful with proxies for participation. 'Who has not done it yet?' is a question that quietly breaks anonymity. Track only the overall completion rate.
Calibre Forum is built on this approach: each person answers a short survey on their phone, the owner sees results only in groups of three or more, and staff are told the results are not used to decide who is needed in a role. The same principles apply if you build your own with a general form tool, provided it has no personal login and gives you control over what is shown.
Let people complete it in work time. Asking them to do it in their lunch break signals that it is not important, and the people who most need to be heard usually have the least spare time.
Reading the results and acting visibly
Look for gaps, not individuals. Where do owner assumptions and staff reports differ? Which repetitive tasks come up across several teams? Which tools appear that you had never heard of? Where is anxiety high, and what do those people say they need?
Then act, and let people see you act. Share a short summary within a few weeks: what you heard, two or three things you will do, and when. Pick at least one change people will notice, such as fixing a clumsy process or setting up a properly supported tool. The next survey will be more honest because of it. If the first one produces silence, the next one will too.
Be ready for answers you do not like. Someone will say the new system is worse than the old one, or that they do not trust management to keep the promise. Those answers are the survey doing its job.
Common questions
What should an AI staff survey ask?
Ask about each person's role, the tasks that fill their week, the tools they use (including ones they set up themselves), the kinds of information they handle, and how they feel about change. Keep wording neutral and focused on work rather than on AI itself. Put policy and insurance questions to owners and managers separately.
How many people do I need for a staff survey to be anonymous?
Report results only for groups of at least three, and merge smaller groups into larger ones. Three is a common rule of thumb, not a legal rule, and a higher number is safer in a small firm. Watch combinations of filters and free-text comments, which can identify one person even when the groups look big.
Should I tell staff the survey is about AI?
You must not mislead them, but you can describe it neutrally and truthfully, for example as a short survey about how work gets done and which tools help or get in the way. Naming AI in the invitation tends to produce guarded answers. Whatever wording you use, be clear about who sees the results.
Can I run an AI staff survey with a free form tool?
Yes, if you set it up carefully. Turn off anything that records names or email addresses, avoid personal logins, set your own minimum group size when you report, and do not look at raw rows alongside who was sent the link. Be honest with staff about the limits of the anonymity you can provide.
This guide is general information, not legal, insurance or financial advice. Check your own obligations with a qualified adviser.