On this page
HR can use AI to write clearer survey and exit interview questions, sort hundreds of open answers into themes, and turn those themes into an action plan. The safe order is simple: remove names and other clues first, then let AI help with the reading and writing. AI should never try to work out who said what, and it should never decide anything about a person.
This guide is for HR officers, people managers and business owners in Kenya who run engagement surveys, pulse checks or exit interviews. It gives seven tasks, each with a prompt you can copy and what to check.
Why this work needs extra care
Staff only tell the truth in a survey if they trust it. One leaked comment can end that trust for years. Kenyan law adds weight to this.
The Data Protection Act, 2019 defines anonymisation as removing personal identifiers "so that the data subject is no longer identifiable". Section 29 says that before you collect personal data you should, as far as practicable, tell people that data is being collected, why, and what their rights are. Section 32 says the organisation must be able to prove consent, and that a person may withdraw consent at any time.
Section 25 also says personal data should be "limited to what is necessary" for the purpose. And the Act's list of sensitive personal data includes health status, ethnic social origin, marital status and family details. Exit interviews often touch on exactly these: a health reason for leaving, a family move, a conflict.
In our view, this means: tell staff plainly what you are asking and why, make taking part genuinely optional, ask only what you need, and strip out identifying details before any AI tool sees the answers. This is general information, not legal advice.
Seven tasks where AI helps, with prompts
Each prompt follows the four Cs we teach at FuKazee: Clarity, Context, Constraints and Customisation. Replace the square brackets. Use invented or cleaned data only.
1. Write better survey questions
Weak questions produce weak data. "Don't you agree that management communicates well?" tells you nothing.
I am an HR officer at a [sector] organisation in Kenya with about [number] staff. Write a 15-question staff survey on [topic, for example workload, communication and growth]. Use a five-point scale from "strongly disagree" to "strongly agree" for 12 questions and add 3 open questions. Rules: one idea per question, no leading wording, plain English a junior staff member would understand. Do not ask for age, tribe, health or family details. Flag any question that could identify a person in a small team.
Check: read each question aloud. If it pushes towards an answer or asks two things at once, rewrite it.
2. Write the invitation and privacy note
Write a short message inviting staff to a [online / paper] survey. Explain in plain words: why we are running it, that taking part is optional, what we will and will not collect, who will see the raw answers, that results will only be reported for groups of [number] or more, and when we will share what we learned. Under 180 words. Warm, honest tone. Add a Kiswahili version.
Check: every promise in the message must be true. If your manager can see the raw responses, do not write that only HR can.
3. Prepare an exit interview guide
Write an exit interview guide for a 30-minute conversation with a staff member leaving a [role type] role. Context: we want honest reasons for leaving and ideas we can act on. Include 10 open questions in a sensible order, starting easy and ending with "what would have made you stay?". Add a short opening script that says the interview is voluntary and explains how notes will be stored. Constraints: no questions about health, family or personal life unless the person raises them.
Check: the guide fits your process. Offer a written option for people who will not speak freely face to face.
4. Clean the answers before AI sees them
Do this step yourself, in your spreadsheet. Delete names, staff numbers, email addresses and job titles that belong to one person. Remove the department column when a department has only a few people. Only then use AI to double-check, and only in a tool your organisation has approved for staff data.
Below are anonymised survey comments. Do not summarise yet. List any comment that still contains something that could point to one person: a name, a unique role, a specific date, an event only one person was involved in. Quote the risky phrase only. COMMENTS: [paste cleaned comments]
Check: remove or blur every flagged phrase. When in doubt, cut it.
5. Find the themes in open answers
Here are [number] anonymised answers to the question "[question]". Group them into 5 to 8 themes. For each theme give: a short name, how many comments fit it, two short quotes taken word for word, and one sentence on what staff seem to want. Put comments that fit no theme under "Other". Constraints: do not invent quotes, do not guess who wrote anything, and do not add themes the comments do not support. COMMENTS: [paste]
Check: pick ten comments at random and see if they sit under the right theme. Count a few themes by hand. AI often miscounts.
6. Turn themes into an action plan
Based on the themes below, suggest an action plan for the next three months. For each theme: one realistic action, who would usually own it (HR, line managers or leadership), a first step for this month, and how we would know it worked. Mark any action that needs budget. Keep it to one page. THEMES: [paste]
Check: leadership agrees the actions before staff hear about them. A plan nobody funds does more harm than no plan.
7. Tell staff what you heard and what you will do
Write a short message to all staff sharing the results of our [month] survey. Include: thanks, the response rate of [number], the three strongest positives, the three biggest concerns, and the actions we have agreed with dates. Plain, honest tone. Do not quote any comment that could identify someone. Under 250 words.
Check: the numbers match your report, and every action has a real owner.
For annual reviews of individual staff, see our guide on using AI for performance reviews. Surveys look at the whole organisation. Reviews look at one person, and they need different care.
What AI should not do in surveys and exit interviews
- Identify people. Never ask AI who wrote a comment, or to match writing styles to staff. That breaks the promise you made in the invitation.
- Predict who will leave, by name. Scoring individuals as "flight risks" from survey answers turns a listening exercise into surveillance. Section 35 of the Data Protection Act also gives people the right, with some exceptions, not to be subject to a decision based solely on automated processing that significantly affects them.
- Report tiny groups. Results for a team of three are not anonymous. Set a minimum group size before you start and stick to it.
- Invent quotes or numbers. Always check quotes against the original answers.
- Replace the conversation. AI can summarise what people wrote. A manager still has to sit down with the team and talk.
How FuKazee trains HR teams on this
FuKazee runs hands-on AI training for HR teams in Kenya, built on the adverts, policies and staff letters a team writes every week. Practice records are invented or masked throughout, and we show the free version of each feature first. Training runs on-site by arrangement or live online. See what an HR programme covers, or start with an HR manager's first 30 days with AI.
Questions people ask
Can I paste survey answers into ChatGPT?
Only after you have removed names and anything else that could point to a person, and only if your organisation's AI policy allows it. Free public tools are not the place for raw staff comments. For the legal side, read can staff use ChatGPT under Kenya's Data Protection Act?
Is an online survey really anonymous?
Not always. Depending on the settings, a form tool may record email addresses or sign-in details, and small teams can be identified from the department and role columns alone. Check the form settings and decide your minimum reporting group before you send it.
What is a good minimum group size for reporting?
We found no legal number. As a rule of thumb, we suggest at least five people per group, and more for sensitive questions. Choose a number that makes it hard to guess who said what, write it in your invitation and keep to it.
Should exit interviews be done by the line manager?
Often it is better if someone outside the person's reporting line runs it, such as HR. People are more open with someone who did not manage them. Offer a written form as well.
Sources
All sources last checked 15 November 2026.
- The Data Protection Act, No. 24 of 2019, ODPC copy: section 2 (anonymisation, consent and sensitive personal data), section 25 (principles), section 29 (duty to notify), section 32 (conditions of consent) and section 35 (automated decisions).
- FuKazee, AI training for HR teams in Kenya: how FuKazee trains HR teams.
Where FuKazee fits in
FuKazee is a Nairobi training company that has trained more than 3,000 people since 2022. We teach HR teams to use AI on real people work, with invented records and clear data rules. Tell us how many people are in your HR function and what you want AI to help with. Email lead@fukazee.com
Want guided support instead?
Plan AI training around your team's real work. Answer a few quick questions about your sector, team size and goals, and we will recommend the right programme and format. Sessions are built on your workflows and include a usage policy.





