On this page
UX designers use AI to sort research notes into themes, draft interview guides and test scripts, write and tighten microcopy, and run a first pass on accessibility. It makes the paperwork of design faster. It cannot tell you what real users do, so you still watch real people use your product, and you still make the design decisions.
TL;DR:
- Use AI to cluster masked research notes, then check every theme against the raw notes.
- Ask it to find leading questions in your interview guide before you meet users.
- It is good at microcopy options: error messages, empty states, buttons, in English and Kiswahili.
- Use it for a first accessibility pass, then test with real tools and real people.
- An AI "user" is not a user. Simulated feedback never replaces a real session.
This guide is for UX and product designers, UX researchers and the designer who is also the product manager at a Kenyan startup. If you work closely with a product manager, share our guide on how product managers use AI. For the visual side of the job, see how graphic designers use AI.
The four Cs
Each prompt below follows the four Cs we teach at FuKazee: Clarity (the job), Context (the product and users), Constraints (format and what to leave out) and Customisation (tone and who reads it). Any general assistant works, such as ChatGPT, Gemini, Claude or Microsoft Copilot. Square brackets are for your own details.
Six tasks where AI helps UX designers today
1. Research synthesis
After eight interviews you have pages of notes. AI can do the first sort.
Clarity: Group these interview notes into themes. Context: they come from eight interviews with small shop owners in Nairobi about how they track credit sales. Constraints: for each theme, give a one-line summary, how many participants mentioned it, and two short quotes copied exactly from the notes. Then list anything that contradicts the main themes. Do not invent quotes. Customisation: plain English for a product team. NOTES: [paste notes with names, phone numbers and business names replaced by P1, P2 and so on]
Check: open the raw notes and confirm every quote exists, word for word, and that the counts are right. AI tends to make themes neater than the evidence. Our guide to qualitative data analysis with AI goes deeper on coding and themes.
2. User interview guides
Here is my draft interview guide for a study on how parents in Kenya pay school fees by phone. Find any leading questions, double questions and questions that ask people to predict the future. Rewrite them as open questions about past behaviour ("Tell me about the last time..."). Keep the guide to 45 minutes. GUIDE: [paste]
Check: read the guide aloud. Run a practice interview with a colleague before you meet real participants.
3. Usability test scripts and tasks
Write a moderated usability test script for a mobile app that lets SACCO members check their loan balance. Include a short welcome that explains we are testing the app, not the person, five realistic tasks written as scenarios (no button names in the task), and follow-up questions for each task. Under two pages.
Check: every task matches what the prototype can really do, and none of the wording gives the answer away.
4. Microcopy
AI is strong at giving you many options to choose from.
Give me five versions of the error message a user sees when an M-PESA payment in our app times out. Context: the money may or may not have left their phone. Constraints: under 25 words, say what happened and the next step, no blame, no technical terms, do not promise a refund. Then translate the best two into simple, natural Kiswahili.
Check: the message matches what your system actually does. A fluent Kiswahili speaker reads the translation, because AI Kiswahili can sound stiff or shift a meaning. Test the final words with users; it is the cheapest test you will ever run.
5. A first accessibility pass
AI can review copy, suggest alt text and build a checklist. It cannot measure what a disabled user experiences.
Here is the text and structure of our sign-up screen: headings, labels, button text, error messages and a description of each image. Review it against WCAG 2.2 level AA. List likely problems with labels, instructions, error messages, alt text and reading order. For each, name the success criterion. Tell me what I must test with a tool or by hand, because you cannot see colours or sizes. [paste]
Check: use a proper contrast checker. WCAG 2.2 asks for a contrast ratio of at least 4.5:1 for normal text and 3:1 for large text, and, at level AA, pointer targets of at least 24 by 24 CSS pixels, with listed exceptions. Then test with a screen reader and keyboard, and with disabled users where you can.
6. Personas and journey maps from real research
Using only the research themes below, draft two personas and one journey map for a first-time user of our agri-input ordering app. Mark every detail that is an assumption, not a finding, as [ASSUMPTION]. THEMES: [paste checked themes]
Check: every detail traces back to research. Delete anything marked [ASSUMPTION] or turn it into a research question.
Why real users are still needed
AI can pretend to be a user and give you confident feedback. That feedback comes from patterns in text, not from a farmer in Kirinyaga with a cracked screen and patchy network. It will miss the things you most need to find: the button nobody notices, the word people misread, the step where trust breaks.
Use AI to prepare for research and to process it faster. Do not use it in place of research. A small number of real sessions will teach you more than any amount of simulated feedback.
What UX designers should not do with AI
- No raw recordings or identifiable notes in public tools. Names, faces, voices, phone numbers and workplaces are personal data. Mask notes first, or use only tools your organisation has approved.
- No invented quotes or numbers in research reports. If it is not in the notes, it is not a finding.
- No "AI said it is accessible". Accessibility is tested, not declared.
- No unreleased product screens in public tools without your employer's or client's permission.
- No AI personas presented as research. Label them as drafts or assumptions.
Data rules and accessibility in Kenya
Research participants have rights. Kenya's Data Protection Act, 2019 applies to interview notes, recordings and survey answers. Section 29 says that, before collecting personal data, you must tell people, as far as practicable, that data is being collected, why, and who else may receive it. Section 30 lists the lawful grounds for processing, starting with consent for specified purposes. Section 32 says the person collecting data must be able to prove consent, and that people can withdraw consent at any time. In practice: a clear consent form, a plain explanation of how recordings are stored, and a way to opt out. See can staff use ChatGPT under Kenya's Data Protection Act?
Accessibility is becoming a legal question too. Kenya now has the Persons with Disabilities Act, 2025. A June 2025 circular from the Head of the Public Service lists the enforcement of accessibility standards among the duties the Act sets out. Ask your lawyer whether and how the Act applies to your product. WCAG 2.2, the W3C's guidelines, is a practical standard to design and test against.
How FuKazee trains product and design teams
Our AI training for startups page describes the research work we teach: preparing interview guides, clustering notes into themes and looking for evidence that contradicts your favourite idea. Teams practise on masked or synthetic data, and we keep user records out of prompts. We cover ChatGPT, Claude, Gemini and Microsoft Copilot, starting with free versions. A designer who wants to learn alone can ask about our one-to-one AI training.
Questions people ask
Will AI replace UX designers?
AI speeds up writing, sorting and drafting. It does not sit with users, notice confusion on a face or weigh business trade-offs. Designers who use it well will move faster than those who do not.
Can AI run user interviews for me?
Some tools can ask scripted questions. They cannot follow up the way a skilled researcher does, and participants may say less to a bot. Use AI to prepare the guide and process the notes, and run the interview yourself.
Is it safe to upload interview recordings to an AI tool?
Only to a tool your organisation has approved, and only if participants agreed to how their recordings would be handled. For public tools, transcribe and mask first.
Can AI check colour contrast?
Not reliably from a description. Use a dedicated contrast checker or your design tool's accessibility features, then confirm with real testing.
What is the best AI tool for UX research?
There is no single best tool. A general assistant handles synthesis and writing well. Choose any specialist research tool by its data terms, not its marketing.
Sources
All sources last checked 15 November 2026.
- W3C, Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, 12 December 2024 version: 1.4.3 Contrast (Minimum), 4.5:1 and 3:1 for large text.
- W3C, Understanding Success Criterion 2.5.8: Target Size (Minimum): 24 by 24 CSS pixels, level AA, with exceptions.
- The Data Protection Act, No. 24 of 2019, ODPC copy: sections 29, 30 and 32.
- Executive Office of the President, circular OP/CAB 9/204, "Guidelines on the implementation of the Persons with Disabilities Act, 2025" (23 June 2025), published by the National Council for Persons with Disabilities: the Act's duties include enforcing accessibility standards.
- FuKazee facts: AI training for startups in Kenya.
Where FuKazee fits in
FuKazee trains founders, designers and product teams to use AI on real research and writing work, with clear rules about which user data stays out of public tools. Training runs at HackHouse in Westlands by arrangement, at your office, or live online. We assume nobody knows AI, avoid jargon and show the free version of each feature first. We have trained more than 3,000 people since 2022.
To talk about AI training for your design or product team, 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.





