Back to all ArticlesChatGPT can generate a list of features, microcopy suggestions, a case study structure, or several flow variations in seconds. That does not mean it understands the product, its users, or the consequences of design decisions.
A prompt is not a magic spell. It is a way to give a conversation direction.
This is particularly important in UX, where an answer that sounds professional may still be based on false assumptions, non-existent evidence, or a process the real system does not support at all.
If you are looking for an idea for what to design, start with my collection of 100 UX/UI Design Prompts.
This article answers a different question:
How can you use ChatGPT during the design process so that it supports analysis, documentation, and decision-making instead of producing generic answers?
A Project Prompt and an AI Prompt Are Not the Same Thing
A project prompt gives you a starting point:
Design a platform that helps older adults manage medical appointments.
A ChatGPT prompt supports the work that begins afterwards. It can help you:
- identify assumptions,
- organise information,
- uncover missing scenarios,
- compare alternative solutions,
- prepare an initial version of documentation,
- explain a decision to different audiences.
The first tells you what you could design.
The second helps you think more carefully about how to approach the problem.
Neither of them does the design work for you.
Before You Use a Prompt, Give It Context
Most weak AI responses begin with a weak instruction:
Improve this UX.
ChatGPT does not know what product you are designing, who its users are, what constraints exist, or what “better UX” means in that particular context. It may therefore generate an answer that is linguistically correct but practically useless.
In The Designer’s AI Playbook, I propose a four-stage workflow:
Lead → Expand → Refine → Elevate
Lead — Set the Direction
Provide:
- context,
- intent,
- constraints,
- your current perspective.
Expand — Broaden the Possibilities
Ask for:
- alternatives,
- scenarios,
- risks,
- edge cases,
- different points of view.
Refine — Narrow and Improve
Remove:
- unsupported assumptions,
- repetition,
- generic advice,
- solutions that do not fit the product.
Elevate — Add Human Judgement
Complete the result with:
- your own experience,
- knowledge of the system,
- real constraints,
- the final decision and its rationale,
- responsibility for the outcome.
AI can expand and organise. You still set the direction, choose the solution, and take responsibility for its consequences.
The prompts below are starting points. Adapt them using your own data, constraints, and project context.
1. Turn a Messy Idea into a Clear Design Problem
Prompt:
I am a Product Designer working on [type of product]. Users are experiencing [observed problem]. Turn the notes below into a clear problem statement. Separate facts from assumptions and identify what information is missing. Do not propose a solution yet.
Notes: [paste the material]
When to Use It
At the beginning of a project, when you have conversations, comments, and broad expectations, but it is still unclear what exactly needs to be solved.
The most important part of this prompt is:
Do not propose a solution yet.
AI tends to jump from an unclear problem directly to a list of features. Design, however, begins with understanding the situation, not with drawing components from a hat.
2. Identify Assumptions You May Not Notice
Prompt:
Analyse the concept below and identify assumptions I may be making without sufficient evidence. Divide them into assumptions about users, business, technology, and process. Do not present your conclusions as facts.
Concept: [describe the solution]
When to Use It
When a solution starts to feel suspiciously obvious.
The most dangerous assumptions rarely look like assumptions. They often sound like this:
- the user will know what to do,
- the manager needs all the available data,
- more options mean greater flexibility,
- an additional step will improve security,
- the user will read the instructions.
AI can help name those assumptions. It cannot confirm whether they are true.
3. Find the Smallest Valuable Version of a Feature
Prompt:
Analyse this feature and propose the smallest version that still solves the user’s core problem. Separate essential elements from optional ones. Explain what value would be lost by removing each element.
Feature: [description]
When to Use It
When the solution starts growing faster than weeds in the backlog.
This prompt should not lead to automatic scope reduction. Its purpose is to help distinguish the elements that solve the main problem from those that entered the project because they “might be useful”.
4. Compare Two Design Decisions
Prompt:
Compare the two solutions below in terms of:
- usability,
- cognitive load,
- risk of error,
- technical complexity,
- long-term maintainability.
Do not choose a winner. Identify the trade-offs and the conditions under which each solution may be appropriate.
Option A: [description]
Option B: [description]
When to Use It
When a team discussion has stalled at:
I just prefer version A.
AI should not make the decision. It can, however, organise the criteria, reveal costs, and show that the “better” solution depends on context.
5. Find Missing Edge Cases
Prompt:
Analyse the flow below and identify possible edge cases. Divide them by risk level: critical, significant, and low priority. Include issues related to data, permissions, interrupted processes, system delays, and actions performed in the wrong order.
Flow: [description]
When to Use It
For forms, multi-step processes, status changes, permission systems, and enterprise products.
The AI response should later be verified with developers, analysts, and people who understand the actual system. A model can suggest a possible scenario, but it does not know whether that dependency really exists.
6. Check What Happens When the User Behaves Unpredictably
Prompt:
Analyse this process from the perspective of a user who:
- stops halfway through,
- returns after a long break,
- completes actions in a different order,
- does not understand the terminology,
- repeats the same mistake several times,
- works with outdated data.
Identify the points where the system may stop being understandable or safe.
Process: [description]
When to Use It
When the happy path already looks beautiful and everyone begins behaving as though users have signed an agreement never to deviate from it.
They have not.
7. Turn an Interview Transcript into Structured Observations
Prompt:
Analyse the transcript below and prepare:
- five main observations,
- evidence supporting each observation,
- contradictions or ambiguities,
- questions that require further investigation.
Do not add information that is not present in the material. Separate participant statements from interpretation.
Transcript: [paste the material]
When to Use It
After conversations with users, clients, or domain experts.
AI can shorten the material and help reveal patterns. It cannot replace contact with users or turn a single statement into a representative insight simply because it has been neatly formatted.
8. Find Patterns and Contradictions in Qualitative Data
Prompt:
Identify recurring themes, behaviours, and problems in the notes below. For each pattern, indicate:
- which fragments support it,
- which fragments contradict it,
- how strong the evidence is,
- what cannot be concluded from it.
Notes: [paste the data]
When to Use It
During research synthesis, especially when you have a large amount of material and it becomes easy to notice only what confirms your initial hypothesis.
Good research is not about finding five statements that agree with the designer. It is also about noticing the sixth one that ruins the entire convenient theory.
9. Turn Findings into Recommendations Without Pretending to Be Certain
Prompt:
Based on the findings below, prepare possible design recommendations. For each recommendation, provide:
- the evidence it is based on,
- the level of confidence,
- the risk of misinterpretation,
- a method for further validation.
Do not present the recommendations as confirmed facts.
Findings: [paste the material]
When to Use It
When you need to move from research to decisions, but the available evidence does not offer complete certainty.
In product work, certainty rarely comes in the deluxe version. We usually work with evidence of varying quality. It is better to state that clearly than to cover it with the tone of an all-knowing consultant.
10. Turn Chaotic Notes into a Design Rationale
Prompt:
Turn the notes below into a structured design rationale. Include:
- the problem,
- the context,
- alternatives considered,
- constraints,
- the chosen decision,
- risks,
- questions that remain open.
Use a neutral and specific tone. Do not add metrics, requirements, or outcomes that are not present in the material.
Notes: [paste the content]
When to Use It
After a design discussion, before handoff, or when you know that three months from now nobody will remember why something was designed that way.
Documentation is not only for people who join the project later. It is also for the team that will discover, with some surprise, that everyone remembers the same conversation differently.
11. Explain a Flow to a Non-Technical Audience
Prompt:
Explain the flow below in plain language for a non-technical audience. Describe:
- the purpose of the process,
- the main path,
- the two most important failure scenarios,
- decisions made by the user,
- actions performed by the system.
Do not simplify terminology in a way that changes its meaning.
Flow: [description]
When to Use It
During presentations for business teams, leadership, clients, and stakeholders who do not need a technical specification but must understand the consequences of the solution.
Simplification should not mean distortion. The goal is understanding, not a marketing fairy tale about a process that “just works”.
12. Turn a Requirement into Material Ready for a Developer Discussion
Prompt:
Analyse the requirement below and prepare:
- expected system behaviour,
- initial conditions,
- roles and permissions,
- dependencies,
- success and failure scenarios,
- questions that require clarification.
Do not fill gaps with your own assumptions.
Requirement: [paste the content]
When to Use It
When the requirement sounds something like:
Users should be able to manage their data easily.
Everything and nothing is clear at the same time.
AI can help break a broad expectation into specific questions. The answers must still come from the right people on the team.
13. Turn a Project into a Case Study Structure
Prompt:
Based on the description below, prepare a portfolio case study structure. Focus on:
- the problem,
- my role,
- constraints,
- key decisions,
- alternatives considered,
- trade-offs,
- the final solution,
- real outcomes or learnings.
Do not invent research, metrics, or business impact.
Project description: [paste the material]
When to Use It
When the project is finished, but its story still exists as 46 Figma screens, seven documents, and the memory of one particularly long Wednesday.
AI can organise the narrative. It cannot invent a process that never happened.
14. Shift the Focus from Screens to Decisions
Prompt:
Analyse this case study section and rewrite it so that it emphasises design decisions, constraints, and consequences rather than only the appearance of the interface. Preserve the facts and my reasoning. Do not add strategic language that is not supported by the material.
Section: [paste the text]
When to Use It
When the project description consists mainly of:
- then I designed the screen,
- then I added filters,
- I changed the button colour,
- the stakeholders liked it.
A recruiter does not need a guided tour of every Figma frame. They need to see whether you understand why the product works the way it does.
15. Prepare a Two-Minute Project Presentation
Prompt:
Prepare a two-minute presentation of this project for a portfolio review or job interview. Include:
- product context,
- the most important problem,
- my responsibility,
- one key decision,
- the most important trade-off,
- the outcome or learning.
Use a specific, professional tone. Avoid exaggerated language and do not add information I have not provided.
Project: [paste the description]
When to Use It
Before an interview, portfolio presentation, or meeting where you need to explain a six-month project before someone has time to ask whether you have any questions for them.
How to Use a Prompt Properly: A Complete Example
Imagine that you are designing a CSV upload process for an enterprise analytics platform.
Instead of writing:
Find edge cases for CSV upload.
conduct the conversation in four stages.
Lead
I am designing a CSV import process for an enterprise platform. The user uploads data, the system checks the file structure, validates the records, and allows the user to continue only after critical errors have been corrected. I want to identify missing edge cases. Do not assume the system repairs data automatically.
Expand
Generate possible edge cases related to file format, encoding, missing values, duplicate records, file size, interrupted connections, and user permissions. Group them by stage of the process.
Refine
Remove cases that require features not described in the context. Merge repetitive scenarios. Divide the remaining cases into critical errors, recoverable issues, and warnings.
Elevate
This is where the actual design work begins.
You review the result with a developer and analyst, compare it with the behaviour of the real system, establish priorities, define messages, and decide when the user may continue.
ChatGPT did not design the process.
It helped you analyse it more broadly.
What Not to Delegate to ChatGPT
Do Not Ask It to Choose the “Best UX”
Without criteria, evidence, and context, “best” usually means the most common pattern.
Do Not Treat Generated Personas as Research
AI can organise real data. It cannot produce evidence about users nobody has spoken to.
Do Not Let It Invent Project Outcomes
A case study without data does not become more credible after adding:
The redesign increased efficiency by 40%.
It does, however, gain a very specific ethical problem.
Do Not Accept an Answer You Cannot Defend
If you do not understand why a recommendation makes sense, it is not yet your decision.
Do Not Share Confidential Data Without Permission and Appropriate Safeguards
Research notes, client data, internal documents, and information covered by an NDA do not become less sensitive simply because the prompt field looks harmless.
A Good Prompt Does Not Replace the Designer
The greatest value of ChatGPT is not generating screens or producing answers at a speed that makes humans suspect they have been doing everything dramatically slowly.
Its value lies in reducing friction:
- it helps begin an analysis,
- organises chaotic material,
- presents alternative perspectives,
- reveals possible gaps,
- prepares initial versions of documentation.
The designer remains responsible for:
- context,
- interpretation,
- priorities,
- trade-offs,
- the decision,
- quality,
- consequences.
AI can expand your thinking. It should not take control of its direction.
Do You Need a Complete System for Working with AI?
The prompts above are only starting points.
The Designer’s AI Playbook includes a library of 50 UX prompts, the Lead → Expand → Refine → Elevate framework, and practical workflows and templates supporting:
- research and analysis,
- design decisions,
- UX documentation,
- writing and microcopy,
- collaboration with PMs, developers, and stakeholders,
- portfolios and project presentations.
It is a book for designers who want to use AI to organise their thinking and improve their workflow without delegating their professional judgement.
