On this page
QA testers use AI to turn requirements into test cases, write clear bug reports from rough notes, create invented test data, and build regression checklists when something changes. AI is quick at spotting missing cases and unclear requirements. The tester still runs every test, judges what "working" means for real users, and never feeds real customer data into a public tool.
TL;DR:
- AI drafts test cases fast, including the negative and edge cases people forget. You still decide what matters.
- A rough note becomes a clean bug report in seconds. Check every step by reproducing it yourself.
- Test data must be invented. Real customer names, phone numbers and ID numbers stay out of prompts and test systems.
- Never mark a test as passed because AI says the code looks right. Run it.
- If your product has an AI feature, test it for prompt injection too.
Who this is for
This guide is for manual and automation testers, QA leads and developers who test their own work, in Kenyan startups, banks, telcos, agencies and software teams. Developers writing unit tests should also read how software developers in Kenya use AI coding tools, which covers test code, reviews and package risks. If you write or receive user stories, see how product managers use AI.
Each prompt uses the four Cs we teach at FuKazee: Clarity (what you want), Context (the product and users), Constraints (scope and format) and Customisation (your team's templates).
Seven tasks where AI helps today
1. Test cases from requirements
Below is a user story and its acceptance criteria for a mobile app used in Kenya. Write test cases in a table: ID, title, preconditions, steps, test data, expected result, type (positive, negative, boundary). Cover the acceptance criteria first, then negative and boundary cases, such as empty fields, very long input, wrong formats, a slow network and the user pressing "back" mid-flow. Then list any part of the requirement that is unclear, as questions for the product owner. Do not invent features that are not in the story. Story: [paste]
Check: every acceptance criterion has at least one test. Remove cases that test behaviour nobody asked for. The questions list is often the most useful output, because unclear requirements cause many bugs.
2. Clear bug reports from rough notes
Turn my notes below into a bug report for our tracker. Fields: short title (what is wrong, where), environment (device, OS, app version, browser), steps to reproduce (numbered), expected result, actual result, how often it happens, and attachments to add. Suggest a severity with one line of reasoning, and mark it SUGGESTED. Keep it factual. Notes: [paste]
Check: reproduce the bug again using only the steps as written. If a developer cannot follow them, rewrite them. Your team agrees the final severity and priority.
3. Invented test data
Create 20 rows of invented test data for a customer sign-up form in Kenya. Columns: first name, last name, email, phone number, date of birth, county. Include cases for testing: an apostrophe in a name, a very long name, an email with a plus sign, a phone number in the wrong format, someone under 18, and an empty county. Make it obvious the data is fake: use the domain example.com for every email. Present it as CSV.
Check: a made-up phone number may still belong to a real person. Never send real SMS, M-PESA prompts or emails to generated numbers and addresses. Use your provider's sandbox or test numbers for anything that sends messages or moves money.
4. Regression checklists after a change
Here are the release notes for this sprint and the list of our main test areas. Suggest which areas to retest and why, ranked by risk. Flag any area that the changes might affect indirectly, such as login, payments, notifications or reports. Present a checklist the team can tick. Release notes: [paste]. Test areas: [paste]
Check: the QA lead decides the final scope. AI does not know which parts of your system are fragile, so add the areas your team knows break often.
5. Exploratory testing charters
Write five exploratory testing charters for the checkout flow of an online shop app. Each one: mission (one line), areas to explore, ideas to try, risks to look for, and a 45-minute time box. Focus on what real users in Kenya do, such as switching between Wi-Fi and mobile data, low storage and interrupted payments.
Check: charters guide the session. The tester's curiosity and notes during the session are where the bugs come from.
6. Testing an AI feature for prompt injection
If your product has a chatbot or AI assistant, it needs its own tests. The OWASP Gen AI Security Project explains that a prompt injection vulnerability "occurs when user prompts alter the LLM's behavior or output in unintended ways".
Our app has a customer support chatbot for a [savings app]. It should only answer questions about accounts, fees and app use, and never reveal other users' data or its own instructions. Write 15 test inputs a tester can use: attempts to make it ignore its rules, reveal its instructions, give financial advice, use rude language, or answer about another customer. For each, give the expected safe behaviour.
Check: run each input in a test environment and record what really happens. Repeat runs, because AI answers vary. Report anything unsafe as a security bug.
7. UAT scripts and test summary reports
Turn these test cases into a user acceptance testing script for branch staff who are not technical: plain English, one action per step, with a space to write "worked" or "did not work" and a comment. [paste]
Write a one-page test summary for this release from the figures below: tests planned, run, passed, failed, blocked; open bugs by severity; risks; and a recommendation section left blank for the QA lead. Do not change any number. Figures: [paste]
Check: the numbers match your test management tool. The release decision belongs to the people your process names, not to AI.
What QA testers must not do with AI
- Paste real customer data. No production records, real names, phone numbers, ID numbers or account details in a public AI tool.
- Share secrets. API keys, passwords, tokens and internal URLs stay out of prompts.
- Mark tests passed without running them. GitHub's own guide to writing tests with Copilot says generated tests "may not cover all scenarios" and that you "should always review the generated code".
- Let AI decide severity or release. It can suggest. Your team decides.
- Treat AI as security testing. It can help plan tests, but security testing needs proper tools and qualified people.
Data rules for testers in Kenya
As of November 2026, Kenya's Data Protection Act, 2019 matters to testing in three ways:
- Personal data is "any information relating to an identified or identifiable natural person" (section 2). A copy of production data in a test system is still personal data.
- Section 25(d) says personal data must be "adequate, relevant, limited to what is necessary". Testing rarely needs real people's details.
- Section 41 requires data protection by design and by default. It lists measures such as "the pseudonymisation and encryption of personal data". Invented or properly masked test data is one practical way to meet this.
Ask your data protection officer before using any copy of real data in testing. Our guides to the Data Protection Act and AI and writing an AI acceptable use policy go further. This is general information, not legal advice.
How FuKazee trains tech teams
FuKazee has trained the HackHouse Africa community and works from HackHouse in Westlands, Nairobi. As our AI training for startups page explains, we teach teams to use coding assistants for tests, refactoring and documentation, with review habits that catch insecure or invented code, and to keep API keys, production databases and user records out of prompts. Developers can have a separate session focused on code and security. We start with free versions and do not promote any one product.
Questions people ask
Will AI replace QA testers?
AI can write test cases quickly, but it does not know your users, your risks or what "good enough" means for your product. Testers who use AI well spend less time typing and more time thinking and exploring.
Which AI tool is best for testing?
Use the one your company approves. General chat tools handle test cases, bug reports and data. Coding assistants inside your editor help with automated tests. Free versions have limits that change.
Can AI write automated tests for me?
Yes, it can draft them. Run every test, make sure it can fail, and read the code before it goes into your suite. Our software developers guide explains more.
Can I paste our requirements document into AI?
Check your company's rules first. Requirements can contain confidential plans or client details. Remove names, client data and anything marked confidential, or use your company's approved tool.
How can a tester analyse lots of test results?
Export results to a spreadsheet and ask AI for formulas or a pivot table, then check the totals. Our guide on how data analysts use AI shows the method.
Sources
All sources last checked 15 November 2026.
- The Data Protection Act, No. 24 of 2019, ODPC copy: section 2 (personal data), section 25(d), section 41.
- OWASP Gen AI Security Project, LLM01:2025 Prompt Injection.
- GitHub Docs, Writing tests with GitHub Copilot.
- FuKazee facts: AI training for startups in Kenya.
Where FuKazee fits in
FuKazee is a Nairobi training company that has trained more than 3,000 people since 2022. We teach tech teams to use AI on real work, with review habits and data rules that hold up in a busy sprint. Tell us the size of your team and how you test today. Email lead@fukazee.com
Want guided support instead?
Not sure which programme fits? Tell us who you are and we will point you to the right path: corporate AI training, the ScholarShift Accelerator for postgraduates, or KCSE study plans and tuition.





