An AI usage survey is a ten-minute anonymous questionnaire that measures how often your people use AI, how deeply, how safely, and what's stopping them. Scored properly, it gives you two readings: where the business sits, which tells you what to fix first, and where each person sits, which tells you what training they need. Run it before you design any AI training, not after.
Everything you need is below: the seventeen questions, the points for each answer, the scoring, and the tools to run it in.
Why does AI training miss when you skip the survey?
Because it's pitched at an average employee who doesn't exist. In most rooms, a few people have used ChatGPT or Claude every day for a year, often on a personal account, and a few have never opened either. Teach the middle and you lose both ends.
Leaders usually can't see the spread. McKinsey asked executives what share of their staff used generative AI for at least 30% of their daily work: they guessed 4%, and 13% of employees said they did (McKinsey, January 2025). In the global study by KPMG and the University of Melbourne, 57% of employees said they hide their use of AI and present AI-generated work as their own (Melbourne Business School, April 2025).
The same study found 47% of employees had received any AI training, and 40% said their workplace had a policy or guidance on generative AI. So people are using AI without guidance, and often without telling anyone. Training designed without knowing that is a guess.
You can't design AI training for a team you haven't measured.
When we start AI training with a business, this survey comes first. It answers two different questions, and the training needs both.
What should an AI usage survey measure?
Four things, which we call Intersect's Four Rs: Reach, Routine, Risk and Readiness. Most AI surveys stop at the first one and ask who uses AI and how often. That measures adoption. It doesn't tell you what to teach.
Reach and Routine add up to a Usage score. Risk becomes a Safety score. Readiness isn't scored at all, because it's the brief for your training.
What questions should an AI usage survey ask?
Seventeen, which takes about ten minutes. Two are about the person's team and role, so you can compare groups. The rest map to the Four Rs, and every scored answer carries between 0 and 4 points.
Copy the bank into whichever tool you use and keep the wording. It's written so nobody has to confess to anything to answer honestly.
How do you score an AI usage survey?
Each person gets two scores out of 100, one for usage and one for safety. Export the responses to a spreadsheet, convert each answer to its points, and apply two formulas.
Here's an illustrative example. Someone in finance uses ChatGPT a few times a week (3), drafts with it and then reworks the result (2), ticks four tasks (2) and saves 30 to 60 minutes a week (2). That's 9 out of 16, a Usage score of 56: a Regular.
They use a personal account only (0), have pasted confidential figures in once (2), usually check the output (3), would tell their manager depending on the work (2) and haven't read the rules (2). That's 9 out of 20, a Safety score of 45. A Regular with a safety flag gets the safe-use session before anything else.
The business reading: Intersect's AI Usage Map
For the business view, take the median Usage and Safety score for each team and place it on a simple two-by-two. Where a team lands tells you what to do first.
How do the scores shape AI training for employees?
Levels decide who goes into which track. The map decides the order. The Readiness answers decide what goes into each session.
The blockers in question 14 tell you whether training is even the fix. If "I'm not sure what I'm allowed to do" comes top, you need rules before a course, and our one-page AI policy template is a place to start. If it's "I don't have access to a good tool", training will stall until you pick one: here's how to set up a shared AI workspace in an afternoon.
If it's "I don't know what it's good for in my job", build the sessions around the answers to question 16. Those are tasks people already want gone, so the examples land.
Your power users are your cheapest trainers, and question 17 finds them. Put managers through early too. Gallup found that employees in AI-adopting organisations who strongly agree their manager supports their team's use of AI are 1.7 times as likely to use it a few times a week or more (Gallup, US data, May 2026).
How do you run the survey so people tell the truth?
Promise an amnesty, make it genuinely anonymous, and show people what happened next. Without all three, the people using AI on personal accounts will answer the way they think you want.
Anonymous comes down to settings and one habit. Don't collect names or emails, don't ask anything that would identify someone in a small team, and never report a group smaller than five. Microsoft's own guidance warns that a form sent to a few specific people can be matched to its answers even with names switched off (Microsoft).
If you already pay for business accounts on ChatGPT, Claude or Copilot, set the survey's Reach answers against what your admin usage data shows. A big gap between the two is your shadow AI.
Which survey tool should you use?
The one your people already sign in to. That's Microsoft Forms if you run on Microsoft 365, or Google Forms if you run on Google Workspace. Both come with what you already pay for, and both can be set so they don't record who answered.
Six tools for running an AI usage survey
| Tool | Best for | Keeping it anonymous | Cost |
|---|---|---|---|
 Microsoft Forms | Companies on Microsoft 365 | Choose "Only people in my organization can respond" and untick "Record name" | Included in Microsoft 365 |
 Google Forms | Companies on Google Workspace | Leave "Collect email addresses" off. People may sign in, but names aren't recorded unless you collect emails | Included in Workspace |
 Tally | A cleaner-looking form for free | No sign-in needed. Don't add name or email fields | Free plan: unlimited forms and responses, fair use |
 SurveyMonkey | Teams that want built-in charts and analysis | Check the collector settings before sending, and don't ask for names | Free plan stops at 10 questions, so this survey needs a paid plan |
 Typeform | One question at a time, polished look | Don't ask for names or emails | Free plan caps at 10 responses a month |
 Culture Amp | Companies already running engagement surveys on it | Built-in minimum group size hides results for small groups | Paid platform |
Checked against each vendor's own pages on 9 November 2026. Whichever you use, export the answers to a spreadsheet to score them.
Pay for something else only if you need what it adds. Culture Amp, for example, hides results for any group below a minimum size you set, so nobody can filter their way to one person's answers (Culture Amp). With the free tools, you enforce that rule yourself when you cut the data.
Where does this break?
- Self-reporting flatters. Even anonymously, people overstate time saved and understate risky use. Treat question 7 as a direction, not a number for a business case.
- Small companies can't stay anonymous. If most teams have fewer than five people, team results identify individuals. Report company-level numbers only, or swap the survey for short conversations.
- The scores are ours, not a benchmark. The weights are our judgement. Use them to compare your own teams and track change, not to compare with another company. Question 3 uses frequency bands that line up with Gallup's, so that one answer can sit beside Gallup's US figures: 15% of employees use AI daily and 30% a few times a week or more (Gallup, May 2026).
- A survey finds people, not value. It tells you who uses AI and how. It won't tell you which processes are worth automating; that's the job of an AI readiness assessment.
- One broken promise ends it. If anyone is disciplined for something they admitted in the survey, nobody will answer honestly again.
What to do next
Pick your tool, copy the seventeen questions, and agree what you'll do with the answers before you send anything. Then put a date in the diary six months out to run it again. The second survey is the one that tells you whether the training worked.
Want us to run it and build the training around it?
We'll run the survey, score it, map your teams and build training tracks from the results, starting with anyone who needs the safe-use basics first.
FAQs
How long should an AI usage survey be?
About ten minutes, which is seventeen questions. That's short enough to fill in during a team meeting, which is the easiest way to get a good response rate. If you have to cut, keep questions 3, 5, 8, 9, 14 and 16.
Should an AI usage survey be anonymous?
Yes. Questions 8, 9 and 11 ask people to admit things. Without anonymity and a clear amnesty, you'll measure what people think you want to hear, which is the opposite of the point.
Should leaders take the survey too?
Yes, and ask them one extra question first: what share of their team they think uses AI weekly. Put the guess next to the real answer when you share results. If the guess is low, that's your case for training made for you.
Can we use AI to analyse the open-text answers?
Yes, if it's a tool the company provides and the answers contain no names. Ask it to group the answers to question 16 into themes and count them. Those themes are your first training examples and your first candidates for automation.
How often should we run it?
Every six months, with the same questions, so the change in scores shows whether training and tools are working. Add new questions at the end rather than rewording old ones.
Related reading
- The One-Page AI Policy: A Template You Can Adopt This Week
- How to Prompt ChatGPT and Claude: An AI 101 for Your Team
- What an AI Readiness Assessment Should Actually Cover
Sources
- McKinsey: Leaders underestimate employees’ AI use, from Superagency in the Workplace, January 2025
- Melbourne Business School: Trust, attitudes and use of AI, a global study 2025 (University of Melbourne and KPMG), April 2025
- Gallup: AI indicator, US employees, data to May 2026, accessed 9 November 2026
- Microsoft Support: Set up your survey so names aren’t recorded, accessed 9 November 2026
- Microsoft Support: Adjust your form or quiz settings in Microsoft Forms, accessed 9 November 2026
- Google Docs Editors Help: Publish and share your form with responders, accessed 9 November 2026
- Google Help: Collect respondents’ email addresses using Google Forms, accessed 9 November 2026
- Tally: free plan, unlimited forms and submissions under fair use, accessed 9 November 2026
- SurveyMonkey: plans and pricing, accessed 9 November 2026
- Typeform Help Center: Free plan, accessed 9 November 2026
- Culture Amp: Confidentiality protections in reporting, accessed 9 November 2026