AI in UX design: what it changes, and what it does not

AI in UX design speeds up the parts of the work that are repetitive: summarising research, drafting copy and layouts, generating variants, building components and checking for errors. It does not replace the judgement that decides what a product should do, for whom, and why one screen works better than another. This guide covers where AI genuinely helps in a UX project, where it fails, how it changes the process, cost and timeline, and the questions to ask an agency that uses it.

Last updated: 21 October 2026.

What does AI in UX design actually mean?

The phrase covers two different things, and buyers often mix them up. The first is AI as a tool in the design process: software that helps designers research, write, sketch, prototype and test faster. The second is AI as a feature of the product itself: a chat assistant, a recommendation, a search that understands plain language. Both need UX work, but this guide is mainly about the first, with a section on the second.

In both cases the definition of UX does not change. User experience design is the work of making a product useful, understandable and pleasant for the people who use it, from the structure of the pages to the wording of a button. AI changes how quickly some of that work is produced, not what good work is.

In short: AI makes the production of UX faster and cheaper. It makes the thinking more important, not less, because a team that can produce ten layouts in an hour still has to know which one is right. The value of a UX partner now sits in research, decisions and craft. Speed is expected; judgement is what you pay for.

At Odduel, a creative and technology agency in Barcelona and São Paulo, AI carries the repetitive production in our UX and UI design work, so projects are delivered in weeks rather than months. Strategy, research interpretation, interaction design and the final review stay human, and that is where the budget goes.

Where does AI help in the UX process?

AI is most useful where the work is high in volume and low in ambiguity. The steps below follow a typical project and show where it earns its place.

1. Research preparation. Drafting interview guides, survey questions and screeners from a short brief, then tidying them by hand.
2. Research synthesis. Transcribing interviews and grouping notes into first themes. A researcher still reads the raw material, because the quote that matters is often the one a summary drops.
3. Content and microcopy. First drafts of labels, error messages, empty states and onboarding text, in several tones, for a writer to choose from and refine.
4. Early layouts. Rough wireframes and layout variants to compare, which shortens the blank-page stage of a project.
5. Design-system production. Generating component variants, states and documentation once the system’s rules are set.
6. Prototyping and front-end code. Turning approved screens into working prototypes or first-pass code that developers then review.
7. Checks. Spotting inconsistent spacing, missing states, contrast problems and broken flows before a human review.
8. First-pass translation. Drafting interface text in other languages, which a native speaker then corrects.

The pattern is the same in every step: AI produces a draft, a person decides. When that order is kept, the time saved is real.

Research synthesis without losing the user

Synthesis is where AI saves the most hours and carries the most risk. A model can group fifty interview notes into themes in minutes, but it tends to favour what is said often over what is said with feeling, and it cannot tell when a participant was being polite. Use it to sort and search, then go back to the recordings for every finding you plan to act on.

Prototypes in hours, not weeks

The largest change for buyers is how early you can see something that works. A clickable prototype, or even a coded one, can now exist within days of the first workshop. That makes testing with real users cheaper and earlier, which is the single best use of the time AI saves.

What can AI not do in UX design?

The limits matter more than the features, because they decide where your money should go.

• It cannot choose the problem. Deciding which user need the product should serve, and which it should ignore, is a business and design decision.
• It cannot observe people. A model has never watched someone hesitate over a checkout. Usability testing with real users stays irreplaceable.
• It averages. Generated layouts and copy drift towards what is common. Common is safe, but it rarely makes a brand recognisable.
• It does not know your context. Your regulations, your sales process, your support team’s constraints and your history with customers are not in the model.
• It can be confidently wrong. Generated research summaries, accessibility claims and code can look finished and still be wrong.
• It does not carry responsibility. Someone has to sign off a flow that handles money, health data or consent. That someone is a person.

A useful way to see it: AI is a fast junior who never gets tired and never asks why. The questions that start with “why” are the work.

AI-assisted UX vs traditional UX, side by side

The table compares how each stage of a project feels to the buyer with and without AI in the process. It describes the general direction we see in our own work, not measured figures.

Stage	Traditional process	AI-assisted process	What stays human
Research planning	Guides written from scratch	Drafted from the brief, then edited	Choosing who to talk to and what to ask
Synthesis	Days of manual tagging	First themes in hours	Deciding which findings are true and matter
Wireframes	Built screen by screen	Variants generated to compare	Choosing the structure and the flow
Visual design	Drawn by hand	Faster production inside set rules	Art direction, brand and detail
Prototype	Late in the project	Early, often coded	What to test and how to read the result
Copy	Written last, often rushed	Drafted alongside the layouts	Voice, clarity and the final words
Quality checks	Manual review	Automated first pass, then review	Sign-off and accessibility judgement

How does AI change UX cost and timelines?

AI moves the budget rather than simply shrinking it. Production hours fall: fewer hours drawing variants, writing first drafts, building components and coding prototypes. Thinking hours stay the same or rise slightly, because more options reach the table and each needs a decision, and because earlier prototypes invite more testing.

For a buyer, the practical effects are three. The total for a defined scope is usually lower than it would have been, because the largest share of a traditional UX budget was production. The timeline is shorter, mostly at the start and in the build. And the proposal you receive should show proportionally more time on research, testing and review than an older one did. If it does not, ask why.

When you compare quotes, look at how the hours are split, not only at the total. A very low price with no research and no testing usually means generated screens with nobody checking whether they work. For a software product, our article on SaaS development cost explains how the same shift affects the build that follows the design, and a fixed quote always follows a short brief.

How do you design UX for AI features?

When the product itself contains AI, the UX work changes in kind, not only in speed. A search box that answers questions, an assistant that drafts a document or a recommendation that changes per user all behave less predictably than a form. Good design for them follows a few principles:

• Set expectations. Say what the feature can and cannot do, in plain words, before the user relies on it.
• Show the source. Where the answer comes from, so the user can check it.
• Make correction easy. Editing, regenerating or ignoring a result should take one step.
• Design the failure. The screen for “I do not know” or a wrong answer matters as much as the happy path.
• Keep a human route. A way to reach a person, or to finish the task without the AI, for anything important.
• Disclose clearly. People should know when they are dealing with an AI system.

That last point is also a legal matter in Europe. The EU Artificial Intelligence Act sets transparency obligations for some AI systems, including telling people when they interact with one, so disclosure belongs in the design from the first wireframe, not in a footnote. If you are still deciding what kind of AI your product needs, our explainer on agentic AI vs generative AI sets out the difference.

What are the risks for accessibility and privacy?

AI tools introduce risks that a careful UX process manages on purpose.

Accessibility

Generated layouts and components often look right and fail in use: low contrast, missing focus states, images without text alternatives, controls a screen reader cannot name. Automated checkers catch part of this, not all of it. Accessibility still needs a human review against the Web Content Accessibility Guidelines (WCAG) 2.2 and testing with a keyboard and a screen reader. For European products this is not optional for many sectors, and our guide to website accessibility covers what the rules require.

Privacy and confidentiality

Interview recordings, customer data and unreleased product plans should not go into a tool whose terms allow it to keep or train on them. Under the GDPR, personal data from research needs a lawful basis and care in how it is processed. Ask any partner which tools they use, under which terms, and what they never upload.

Sameness

The quiet risk is that every product starts to look and read alike, because the same tools suggest the same patterns. Strong art direction and a clear brand voice are the defence, and they are human work.

How do you choose a UX partner that uses AI well?

Most agencies now use AI somewhere. The question is whether they use it to spend more time on what matters or to cut it out. These questions separate the two:

1. Which steps of your process use AI, and which never do?
2. How many real users will we talk to or test with, and when?
3. Who reviews generated research summaries against the raw material?
4. How do you check accessibility beyond an automated tool?
5. Which AI tools touch our data, and under what terms?
6. How does the time AI saves show up in our proposal?
7. Who signs off the final design, and what are they checking?

Clear, specific answers are a good sign. “We use AI for everything” and “we never use AI” are both answers that deserve a follow-up.

What does a realistic AI-assisted UX workflow look like?

To make this concrete, here is how a website or app project runs when AI is used well. The order matters more than the tools.

1. Brief and goals. A workshop with you to agree the business goal, the users and how success is measured. Human throughout.
2. Research. Interviews or testing with real users, with AI helping to prepare guides and sort notes.
3. Structure. Information architecture and key flows decided by the designers, with generated variants used to compare options quickly.
4. Design system and screens. Art direction set by hand; components and states produced faster with AI inside those rules.
5. Prototype and test. An early working prototype, tested with users, then revised.
6. Build and review. Front-end code drafted with AI and reviewed by developers; accessibility and copy checked by people.
7. Launch and learn. Analytics and feedback reviewed after launch, with AI helping to summarise and people deciding what changes.

Questions people ask

If you are planning a website or app and want the speed AI brings without losing the research and craft that make a product work, our UX and UI design service explains how we run a project from brief to launch. Send us a few lines about your product and your users, and talk to us about your project.