In a team of eight, a survey is only anonymous if nobody can work out who wrote which answer. That depends on what you collect, how you slice the results and how you report them, not on a promise in the invitation email. For AI use and skills, an anonymous survey of every staff member gives you a map of what people actually use and where they are stuck. We built our AI readiness assessment around that idea: an anonymous staff survey that produces a scored report and a 90-day action plan.
Calibreforumcalibreforum.auTurn AI Use Into a Clear PlanCalibre Forum’s AI readiness assessment gathers staff insights and produces a scored report with a clear 90-day action plan.VisitTL;DR
- Anonymous means no identifying information is collected at all. Confidential means you know who answered and promise to keep it hidden. Only the first gives staff real cover.
- Small teams are the hard case. Department, role, tenure and free-text comments can identify a person even without a name.
- Australian public-sector and ethics guidance sets minimum group sizes before results are shown. Use one for your own reporting, and report the whole organisation rather than team by team.
- Ask about tools used, tasks, confidence, training, barriers and data handling, then agree what you will do with the answers before you send the survey.
- Share what you found and what changes, within weeks. Staff judge the next survey by what happened after this one.
Table of Contents
- Anonymous and confidential are not the same thing
- Why small teams make anonymity harder
- Designing the survey so nobody can be picked out
- What to ask when the topic is AI
- Getting honest participation
- From answers to a plan
- What our assessment does for a small team
- FAQ
- Sources
Anonymous and confidential are not the same thing
Most surveys sold as anonymous are confidential. The difference matters most when your team is small.
The Australian National University’s research ethics guidance says a survey is anonymous when the participant’s identity and responses cannot be identified, even by the researcher, because no identifying information was given. Confidentiality is different: the person has supplied identifying or potentially identifying details, and the researcher promises to keep them hidden. The same guidance says that if people are identifiable in the raw data, calling them anonymous is not accurate, even when they are not named in the final report.
It also warns that blanket promises of strict confidentiality are unhelpful, because there are legally appropriate situations where confidentiality can be broken, such as when data is subpoenaed. If your survey tool logs email addresses or login names, you are running a confidential survey, and your wording should say so.
For a manager, the practical test is simple. Ask whether anyone in your business, including the person who administers the survey, could look up who gave a particular answer. If yes, it is not anonymous. Staff usually assume the worst about that question, and they are right to.

Why small teams make anonymity harder
A 200-person company can break results down by department and still hide individuals. A team of eight cannot. Three things give people away.

Demographic filters. One survey question about role and another about tenure can narrow a group to a single person. The Australian IRIS position statement on employee surveys (published December 2025) says small team sizes, demographic filters and verbatim comments can make individuals identifiable even in supposedly anonymous data, and that this creates risks of retribution and career impacts.
Free-text comments. Writing style, a project name or a specific incident can point to the author. The IRIS statement lists verbatim comments among the risks.
Well-known views. The ANU guidance notes that identification by third parties is more likely in small populations, or where a person’s views are already well known. If everyone knows who is sceptical about a new tool, an answer that sounds like them will be attributed to them.
Minimum group sizes you can borrow
You do not need to invent a rule. Two Australian sources give you numbers:
| Source | Minimum for showing group results | Free-text comments |
|---|---|---|
| NSW Public Service Commission, People Matter survey | 10 or more respondents in the group | 30 or more responses |
| IRIS position statement | Do not publish where there are five or fewer respondents | Not specified |
The NSW survey also collects no name, date of birth, email address or employee ID. For a team of 12, a threshold of 10 would stop you reporting by department at all, and that is the point. Report the whole organisation, and break results out only where the group is large enough.
The UTS Human Technology Institute’s 2025 guide to worker engagement on AI makes the same point from the AI side: it recommends comparing teams or AI-literacy levels only when doing so will not jeopardise anonymity, particularly with small samples.
Designing the survey so nobody can be picked out
Follow these steps in order. Each one is a place where small-team surveys leak identity.
1. Decide what you need to learn before writing questions
Every question is a potential identifier and a burden on the respondent. If you cannot say how an answer will change a decision, cut the question. For an AI survey, the useful decisions are usually: which tools are in use, which tasks they are used for, what training is needed and what data is being put into them.
2. Collect the minimum demographics, ideally none
Skip role, tenure, age and location unless you will genuinely report on them and the groups are large enough. In a small business, “whole organisation” is often the only reporting level that is safe. This is where a survey can give you real visibility without anyone being exposed.
3. Check how the tool treats identity
Look for no login, no email-linked invitations that tie answers to a person, and no individual response exports. Ask who has access to the data. The WGEA’s guide to surveying staff says anonymity supports employees expressing their views honestly and that, to build trust, you should make explicit which staff will have access to the data. If you are comparing tools, this explanation of how survey software protects employee confidentiality shows the kinds of safeguards to look for.
4. Handle open comments deliberately
Open text is where the richest insight sits and where the biggest risk sits. Options: read comments only in aggregate themes, remove names and project details before sharing, or limit open text to one question about barriers. The NSW threshold of 30 responses for comments shows how cautious large organisations are about this.
5. Do not chase individuals
You want a high response rate, but pressure undermines the survey. The IRIS statement warns that monitoring response rates can cross from quality assurance into surveillance, especially when teams are singled out for low participation. Send one reminder to everyone, and leave it there. Do not announce which team is behind.
6. Say honestly what the survey can and cannot protect
Use the ANU’s suggestion: qualify promises with phrases like “as far as the law allows”. A short, accurate explanation beats a sweeping guarantee. Also decide in advance what happens if a comment describes something serious such as harassment. IRIS recommends defined thresholds and pathways that preserve anonymity where possible, so write that down before launch.
What to ask when the topic is AI
A good AI survey covers tools, tasks, skills, attitudes and risk. The Indeed Australia guide to building an AI upskilling program suggests assessing AI fundamentals and output checking, attitudes to adoption, workflow and task suitability, approved tools, and policies on privacy, acceptable use and confidentiality.
Turned into plain questions for staff:
- Which AI tools have you used for work in the past month? (Include ones you pay for yourself or use on a personal account.)
- Which tasks do you use them for, and how often?
- How confident are you checking an AI answer before it goes to a client or colleague?
- What stops you using AI more? (Not knowing how, not having access, rules you are unsure about, worry about mistakes or job impact.)
- Do you know what information you are allowed to put into these tools?
- What training would help most?
- Which tasks would you hand to AI if you were sure it was allowed and safe?
Research from Finland’s VTT on AI use at work found that respondents cited privacy and security and reliability as concerns, and lack of familiarity, organisational restrictions and lack of training as reasons for not using AI. Those are good prompts for your barrier options, though your own staff may give different answers.
Question 1 matters for another reason: it surfaces tools you did not approve. Our guides on shadow AI in the workplace and on what staff can paste into AI tools under the Privacy Act explain why that visibility matters.
Using the survey as an AI skills gap assessment
A skills gap assessment compares what your people can do now with what your planned AI use requires. The survey gives you the first half. For the second, Australia’s National AI Centre says that before introducing AI, organisations should check whether they have people able to use and oversee AI tools, whether workers have been meaningfully consulted, what training teams need, and where human judgement still matters.
In practice, read your results in three layers:
- Capability: how confident staff are in checking outputs and handling data, and whether a few people carry most of the knowledge.
- Need: which tasks staff want help with, and what training they ask for.
- Gap: the distance between those and what you plan to roll out.
The Indeed guide recommends combining short surveys with interviews, manager workshops and a review of existing tools and policies. That suits a small team well: use the survey for the honest, anonymous picture, then discuss the themes openly in a team meeting.
Getting honest participation
Anonymity is a design feature. Trust is what makes people use it.
The UTS guide advises engaging workers in a psychologically safe, interactive setting, listening to their views and incorporating them into decisions. A few habits follow from that:
- Explain the purpose in two sentences. Say what you want to learn and what you will do with it, such as choosing training and setting tool rules. Do not frame it as a performance check.
- Name who sees the data. The WGEA guidance says to be explicit about access. If only the owner sees aggregate results, say so.
- Pick the format for your team. The WGEA notes paper surveys can feel more anonymous to respondents, while online surveys are easier to distribute and analyse. In a very small team, online with no login is usually the practical choice, provided you explain it.
- Treat answers as self-reported. People may overstate or understate skill. Pair numbers with comments and a follow-up conversation. The Society for Industrial and Organizational Psychology recommends treating surveys as part of ongoing listening, pairing quantitative results with qualitative context and making follow-up commitments visible.
The last point is where most surveys fail. If staff see nothing change after the first survey, the second gets guarded answers. For ideas on building that trust through open results and visible action, this piece on reducing workplace stress and building team trust is a helpful companion.
From answers to a plan
The point of an AI survey is a decision. The UTS guide recommends prioritising opportunities by feasibility, impact and alignment with business objectives, while addressing workforce capability, workflow suitability, tool access and governance gaps. It also advises testing solutions with workers, assigning follow-up actions, explaining what changed and why, and revisiting outcomes at intervals.
A simple version for a small business:
- List the findings in four groups: tools in use, skills, barriers, data handling.
- Score each potential action from 1 to 3 on feasibility, impact and fit with your goals.
- Pick three to five actions for the next 90 days, each with an owner and a date.
- Tell staff what you found and what you are doing, in the same week you decide.
- Repeat the survey after a quarter to see whether confidence, tool use and data handling moved.
Ninety days is long enough to run a training session, write a one-page tool policy and approve a short list of tools. It is short enough that staff remember the survey.
A note on privacy rules
Survey data and AI tool use both touch Australian privacy law. The Privacy Act 1988 defines a small business by annual turnover of $3 million or less, subject to statutory exceptions, so whether the Act applies to you depends on your business and activities. The Office of the Australian Information Commissioner’s Australian Privacy Principle guidelines cover collection, use and disclosure, security, and access and correction. If you are unsure whether the Act covers you, ask your lawyer. A truly anonymous survey that collects no personal information also reduces what you have to manage.
Our view
We built Calibre Forum around one belief: you cannot manage the AI use you cannot see, and staff will not show you it unless they are sure they cannot be identified.
That is why our assessment surveys every staff member anonymously, without installing software or collecting individual data. A map of tools and workflows drawn from the whole organisation is more honest than a handful of manager opinions, and in a small business it is the only kind that works, because there is nowhere to hide in a team of ten.
We also think a survey should end in action, not a dashboard. So the scored report comes with a 90-day plan: what to fix first, who needs what training and where the security gaps are. Staff see their answers lead somewhere, which is what makes them honest the next time.
Anthony
What our assessment does for a small team
Running this properly means designing the questions, protecting anonymity, scoring the answers and turning them into a plan. Our AI readiness assessment does that for Australian and New Zealand businesses. Every staff member completes an anonymous survey on the AI tools they use and how they work, and you receive one scored report that maps capabilities, flags risks and sets out a tailored 90-day action plan with operational and security recommendations. You can see what is inside on the report contents page and the how it works page.

The one-off report is A$1,990. If you want to re-measure each quarter, the membership is A$2,388 per year and covers recurring assessments and reports. We also have a guide on running an AI staff survey honestly and anonymously.
FAQ
How many people do you need for an anonymous survey?
There is no legal minimum for a private business, but published standards give you a guide. The NSW public service survey shows group results only with 10 or more respondents, and the IRIS position statement suggests not publishing groups of five or fewer. In a small team, report on the whole organisation instead of splitting by team.
Can an anonymous survey still identify someone?
Yes, through combinations of answers such as role plus tenure, or through free-text comments. Keep demographic questions to a minimum, skip any breakdown that leaves a small group, and handle comments as themes. The ANU guidance notes the risk is higher in small populations.
Should staff be told the survey is anonymous or confidential?
Use the word that is true. If you collect no identifying information, say anonymous. If you collect names or emails and restrict access, say confidential, and explain who can see what and the limits, such as legal requirements.
How often should we run an AI staff survey?
Once gives you a baseline. A repeat each quarter shows whether training and policy changes are working. The Society for Industrial and Organizational Psychology treats surveys as part of ongoing listening, with follow-up commitments made visible.
Will staff worry the survey is about their jobs?
Some will, especially when AI is involved. Say plainly that the aim is to understand tool use, training needs and barriers across the organisation, that results are reported in aggregate and what you will do afterwards. Then do it.
Sources
- Confidentiality and Privacy, ANU
- FAQs, NSW Public Service Commission
- Position Statement: Employee Surveys and Ethical Review, IRIS
- From Invisible to Involved: A Guide to Worker Engagement on AI, UTS Human Technology Institute
- How to effectively survey your staff, WGEA
- How to Build an AI Upskilling Program in 9 Practical Steps, Indeed Australia
- VTT Research Report on AI use at work
- Get ready for AI, National AI Centre
- When Engagement Data Lies, SIOP
- Privacy Act 1988, compilation of 10 June 2025
- Consolidated APP guidelines, OAIC
Recommended
- How to Run an AI Staff Survey Honestly and Anonymously
- Staff Using ChatGPT at Work: What an Owner Should Do Next
General information only. Articles on this blog are written with the help of AI writing tools and published by Calibre Forum. They are not legal, insurance, tax or financial advice, so check anything important with a qualified adviser.