To understand how UX professionals use generative AI in their day-to-day work, we conducted a large-scale survey with more than 800 respondents. 92% of the respondents claimed they had used at least one generative AI tool; among those who used these bots for work, 63% used them at least several times per week (if not daily). The most common type of activity is text-content generation and editing.
Our Study
We ran a survey study with 841 UX professionals. The study was conducted in July and August 2023.
We asked respondents 15 questions regarding:
- General AI usage: Participants were asked whether they had used generative-AI tools (like ChatGPT and MidJourney) for work or personal purposes and to list all generative-AI tools they had used. Those who said they had used these tools for work were also asked how often they used them and for what kinds of work-related activities.
- A specific use case of generative-AI tools: We asked people to recall a recent instance when they had used these tools to assist their work. (This type of question is known as the critical-incident technique.)
- Usage of UX tools with generative-AI capabilities: Respondents were asked if they had used generative-AI features in common UX tools (e.g., Dovetail’s summarization feature).
- Background information (optional): Respondents were asked to share their job titles and years of experience.
Widespread AI Usage in UX
92% of respondents reported that they had used at least one generative-AI tool (95% confidence interval: 90.3%-93.9%).
There’s an important limitation to this study, however — we recruited through email and social media and used the term “generative AI” in our recruitment text. It’s possible that people who were interested in AI were more likely to respond (selection bias). While we believe that the percentage of UX professionals who use AI tools is high, the exact number may not be as high as 92%.
On average, respondents used two different generative-AI tools. The two most mentioned tool types were text and multimedia generators:
- Text generators: ChatGPT (90%), Bard (17%), Bing Chat (7%)
- Multimedia generators: Midjourney (32%), Dall-E (15%), Adobe Firefly (5%)
Among those who had used these tools, most (78%, 95% confidence interval: 74.6%-80.5%) used AI tools for both work and personal purposes; 8% (95% confidence interval: 6.4%-10.4%) used AI only for personal use.
Many of those who used AI tools for work did so regularly: 63% (95% confidence interval: 59.3%-66.5%) used AI at least several times a week (24% reported using AI daily, while 39% used it several times a week).
4 Roles of Generative AI in UX Professionals’ Work
We identified 30 types of tasks that UX professionals used generative AI tools for in their work. We grouped these tasks under four roles: content editor, research assistant, ideation partner, or design assistant.
- Content editor: Generating and editing text, from microcopy to social media posts, based on specifications or copy given by UX practitioners
- Research assistant: Preparing desk-research summaries and specific UX-research documents (such as study plans, screeners, and interview guides)
- Design assistant: Generating and editing images, videos, and design deliverables such as personas, wireframes and prototypes, and journey maps
- Ideation partner: Exploring variations of content, including text and multimedia, at a diverse granularity ranges (from high-level feature generation to nitty-gritty design details), ideating solutions, and validating ideas
Text-Content Editor
More than 75% of respondents used generative AI for text-related tasks, specifically:
- To create or modify various types of text, including microcopy, navigation labels, social media posts, and user stories.
- To suggest alternative words and convey a particular tone of voice
AI support for content generation can save UX practitioners significant effort, especially if they do not have a professional-writing background or if they work in organizations with no dedicated UX writers.
|
Content-Editor Tasks |
|
|
Tasks |
Examples |
|
Generating UX microcopy, including error messages, navigation labels, help text |
“I often use programs like ChatGPT to help me rewrite or get ideas for microcopy on things like modals and text within a page. I use it to help reword copy so that it is short and concise and easy for anyone to understand.” |
|
Create a specific type of content (e.g., press release, job description, objectives and key results –-OKR) |
“I recently used ChatGPT to write a press release for a new product the company is creating. The product is related to investments, something I’m not an expert in, so it was helpful in structuring the information in a way that made sense.” “As I was leading a OKRs working session with my team, I wanted to quickly find a good UX OKRs example. I stated the vision/mission statement (still related to my line of work) and ask ChatGPT to generate one objective, 4 key results, the related initiatives and the KPIs for each initiative.” |
|
Expand, trim, revise, or summarize the text based on limits or constraints |
“I used ChatGPT to create a summary of what UX Benchmarking studies are to include in my report introduction. I asked the AI to include details from my project that I provided to it. I was impressed with how good the paragraphs it provided me were.” “At a point I needed some descriptions with a limit of characters and then I went to a generative AI Tool in order to get suggestions of how these copies could be with the specified character limit. After getting the suggestions I made some changes upon these and was able to complete the task in a limited time scope.” |
|
Prepare the text for a specific audience |
“I used ChatGPT to generate some definitions of UX-related terms and concepts [for] a presentation I was going to give to some non-UX colleagues.” “Improving readability of content — rewriting for a certain grade level or FK score, asking it to “improve this with plain language.” |
|
Enforce a specific tone of voice |
“I use it to help me write intent copy for a native app that my team is creating. I have it follow a set of prompts that help us keep a similar tone and way of speaking, so that it stays consistent with the brand.” |
|
Rewording and proofreading |
“I am not a naturally gifted writer and normally spend hours on this type of writing, trying to find the right words and grammar - this tool has taken a lot of that kind of stress off my shoulders, I don't have to dwell and dwell to find the right word that's just on the tip of my tongue.” “I type up emails or communications I want to send in my own words, then I ask AI to reword it. It helps to make my thoughts clearer and more concise, but still contains the same messaging I personally wanted so it is still my voice.” |
Research Assistant
The second biggest category of activities that generative AI assisted with was research; Around half of the participants used these tools to perform tasks within this role.
The research activities were divided into two subcategories:
- Desk research: General information seeking to understand a specific problem or domain (for example, researching a common design pattern, or a new field for a consulting project)
- User research: Learning about a group of target users to inform product design
Desk Research
A huge efficiency benefit brought by generative AI is the aggregation of information in the information-foraging process. Instead of manually searching for relevant sources, extracting interesting information from each, then combining everything into a coherent result, professionals can ask the AI to do this work for them.
Sometimes respondents input content, such as a lengthy and abstruse article, into the AI and asked it to summarize key points for them.
UX professionals reported using generative AI for desk research in the following scenarios.
|
Desk-Research Tasks |
|
|
Tasks |
Examples |
|
Study a new topic |
“Answer questions about statistical analysis” “Quick research on UX Strategy concepts” |
|
Check best practices for specific types of pages, elements, workflows, or apps |
“To answer questions like ‘in bank app home screen, which information are most often displayed?’” “I used ChatGPT to ensure I covered all the steps needed to conduct a technical SEO check. The AI-generated response included a couple of steps I had not thought of, so I was able to take that and add more to my process steps and the project map.” |
|
Get resource recommendations for in-depth research |
“Recently I was learning about priming in UX; I used AI for getting the right sources to study about the topic. And it was very helpful.” “I used ChatGPT to get suggestions on where and which types of websites or sector I could find some specific UX and interaction patterns.” |
|
Acquire new domain knowledge for complex products |
“I was working on a B2B product that focuses on KYC processes. I used ChatGPT to understand the basic/foundation of KYC [Know Your Customer] as I had no background for it. I asked ChatGPT to generate why is it being used, how is it being used, what are the most common phases or process of KYC. From those data gathered, I use it to create assumptions/hypothesis for my product and revalidate them with actual users.” “Recently, I was researching on an accounting software and, as I’m not from that background, was having really difficult time on deciding what to include as a feature and which feature isn’t required for that specific system. When I started asking to generative AI, it helped me clear most of confusion and also helped me generate new feature ideas and opinions.” |
|
Summarize key points based on content provided by users |
“Generate summaries of AI/ML-related concepts for my notes” “Copy and paste large amounts of content to get a summary” |
UX Research Tasks
In addition to general information seeking, respondents reported using generative AI tools to assist throughout all phases of their UX research projects.
|
UX Research Tasks |
|
|
Tasks |
Examples |
|
Draft study protocols, scripts, interview guides |
“I used ChatGPT to generate interview guides. The goal was to come up with questions to validate several aspects of a new feature idea. The AI did remarkably well, it suggested a couple of questions I didn’t even think of!” “Today I used it to write an email to participants taking part in a study, I then used it to write a script including question[s] I would ask during a usability study to improve an app.” |
|
Finetune task wording |
“I used it to generate types of questions to ask users without bias.” |
|
Generate recruitment text like communication emails, study-introduction documents |
“I was creating a survey and needed to draft a few sentence descriptions that participants would see before beginning. I had writers block and was having trouble writing the description. I used ChatGPT to draft something for me that I edited. ” |
|
Summarize insights and extract themes from research sessions |
“Content analysis. I ran a survey and had two open ended questions and wanted to see how useful ChatGPT can be and used it. It took a lot of trial and error to find the best way together what I wanted and I would say it did a decent job. 70% accuracy.” |
|
Generate research reports and presentation outlines or slides |
“Used GPT-4 with plugins to generate summaries of concepts related to a study and suggest a format for a report. Inputted rough notes from lit review into GPT-4 and got a first draft summary write up.” |
Participants also used generative AI for high-level ResearchOps planning. One participant shared:
“We have been considering starting a UX community of practice in our organization. I was unsure of where to start so I used ChatGPT to help me figure out where to begin and it gave me clear steps: ‘Define the purpose and objectives of the community; Identify the target audience; Establish rules and guidelines; Plan and schedule activities and Promote the community.’ This outline gave me a place to start, so I took those suggestions and ran with them.
Then I went back and asked it to give me some examples of rules and guidelines for a community of practice, just to get a sense of what that could look like. While I didn't use the exact things it suggested, it gave me enough of an understanding to tailor these to our organization.
Once I'd drafted all of this, I decided I wanted to create guiding principles based on our purpose and objectives. I used ChatGPT to improve the labeling and optimize the summaries for these. Next, I used it to generate some ideas for activities and was very pleased with the results and used most of them. I was able to create an entire proposal for our UX Community of Practice in less than 2 hours — a task that normally would have dragged on over several days.”
Ideation Partner
The third most common usage of generative-AI tools was for inspiration and ideas. 31% of the respondents mentioned they used the bot to generate something to start with and overcome blank-page syndrome. They worked collaboratively with the AI and built upon each other’s output to move forward.
In this role, respondents treated these tools like coworkers: they shared background information and explored potential solutions to a given challenge together. A participant pointed out:
“I find these tools help you overcome personal bias by pointing out things you've overlooked and giving you another perspective to consider.”
|
Ideation-Partner Tasks |
|
|
Tasks |
Examples |
|
Explore content variations at different stages of the creation process (from general outlines or concept visualizations to specific words or color palettes) |
“I asked ChatGPT how to explain what is UX research to nondesigners. Also I asked for content structure for PPT presentation. I use it because it always takes me time to start with this kind of initial ideation. ChatGPT reduces this time to 30 seconds, and I can move forward.” “I used ChatGPT-4 to explore fresh ideas and design concepts. The AI-generated responses provided design ideas, ranging from innovative navigation concepts to intuitive gestures and animations.” |
|
Ideate scenarios and solutions |
“Suggestions of ice breaker activities for design thinking workshops.” “I used chat GPT to generate ideas about a complex form for an HR software.” “Recently I used ChatGPT to concept a feature by letting it generate related user stories. Due to the broad output provided by ChatGPT, I was able to think towards multiple directions.” “Get ideas on different ways to handle work scenarios.” “In product design work, it helps me find better ideas and cover the edge cases in the app.” |
|
Role playing: Asking the AI to play a specific role (e.g., a stakeholder) and provide an evaluation or suggestions for a given task |
“I assign roles to ChatGPT and then I inquire to provide instructions for a specific task, or evaluation of learning a new subject. “ |
|
Validate ideas and identify gaps, by having the AI inspect a specific plan or sequence of actions |
“Feed information about our product to quickly identify potential features and requirements that may have been looked over.” “Sometimes I ask ChatGPT to review my ideas for things like user flows.” “I used ChatGPT to ensure I covered all the steps needed to conduct a technical SEO check. The AI generated response included a couple of steps I had not through of, so I was able to take that and add more to my process steps and the project map.” |
Design Assistant
24% of our survey respondents mentioned using generative-AI tools for design-related tasks. These included generating and modifying:
- Multimedia like images and videos
- Design deliverables such as personas, prototypes, and journey maps
|
Design-Assistant Tasks |
|
|
Tasks |
Examples |
|
Generate and edit images and videos |
“Generate and match illustrations as per the existing design language.” “I recently prepared a mood board to display the visual aspects of a digital product to the client. To accomplish this, I relied on Midjourney for generating images, referring to other images, and exporting mockups that I could easily edit on Photoshop for the final touchups.” |
|
Create personas (including text and visuals, such as persona photos) |
“I described our user groups and gave some important characteristics about [our] user in ChatGPT, and I got personas, which, with some changes, I could use for my project.” “I created personas for a project and used ChatGPT. I wanted a diverse selection of personas and was hoping AI might consider angles that I did it. It worked well and AI suggested pain points, goals, frustrations that I did not come up with.” “I used Adobe Firefly for personas face generation.” |
|
Create storyboards and user flows |
“[I] used ChatGPT to create a storyboard for a performance-management tool’s feature; it gave out bland responses at first but, with adequate detailing, it was able to generate specific captions. I used these prompts to create a storyboard with illustrations.” “The AI-generated content inspired new user-flow possibilities that I didn’t consider initially. GPT-4 suggested alternative ways users can navigate through the app and interact with its features.” |
|
Generate wireframes and prototypes |
“I wanted to design an order-list page; I wanted the details that it should have, so I asked ChatGPT.” “I wanted to create a goods card, so I ran Automator (a Figma plugin). […] After several tries, I created a card with image, title, and subtitle.” |
|
Provide color-palette recommendations |
“I had 3 values per each color in my design system. I needed more of them, so I decided to ask ChatGPT to generate the corresponding ones.” “I have just started my personal project, so I use the text-to-image tool to generate my branding guidelines such as logo ideas, colors, and more photos for illustration.” |
|
Generate specific UX maps such as journey maps and empathy maps |
“Creating a draft of a customer-journey map for a client where research was not possible and time was very limited — serves as a discussion point and as a way to get a user-centered mindset into the implementation teams.” |
Combine Roles for More Powerful Problem Solving
In many of our participants’ responses, the AI was played more than one of the four roles at a time.
For example, content-generation tasks were often accompanied by ideation. Or a design-assistant activity included a content-editor component. Combining roles increased the AI’s problem-solving abilities and usefulness.
Tasks that used the AI in multiple roles were often more complicated and required more back-and-forth communication: this was not only because of the articulation barrier, but also because users were learning more about the problem space and narrowing down their final choices.
For instance, one respondent shared their process of error-message ideation with us. In this case, ChatGPT was both a content editor and an ideation partner:
“I needed some error messages so I described the use case to ChatGPT to get back some ideas of how it could be. I had to refine it to accommodate some edge cases/restrictions that the AI could not know and then implemented it in my design.”
Similarly, combining the roles of research assistant and ideation partner allowed UX professionals to first learn about a topic and then make informed decisions, as another respondent shared:
“Recently, I was researching accounting software and, as I'm not from that background, was having really difficult time on deciding what to include as a feature and which feature isn't required for that specific system. When I started asking to generative AI, it helped me clear most of confusion and also helped me generate new feature ideas and opinions.”
When more roles generative-AI tools are combined, the AI can support even more complex tasks. One respondent described how they used ChatGPT and Midjourney to ideate user journeys and interface copies, and further inspire design.
“I designed an experience for an app marketplace, using only generative AI tools. We had the business knowledge and the identified need for this product, and we used ChatGPT and Midjourney to generate journeys, give us hints of how the layout should be structured, color palettes, copy, error and success messages. And we took everything to Figma and built an experience in 2 days. [It] was really impressive to see how fast we reached 3 different options, that can now be tested and validated with our users.”
Content Editor: The Most Common Role
In this survey, UX professionals were far more likely to use generative AI for text generation than for multimedia generation. The difference was statistically significant (p < 0.001). Using AI to help with design-focused tasks (like creating storyboards, journey maps, or color palettes) was far less commonly mentioned by UX professionals, even compared to research-focused tasks (p < 0.001) or ideation tasks (p = 0.012).
We believe this current preference for text generation occurs because:
- UX professionals do lots of writing.
- Dedicated AI support for UX is (currently) limited.
- Many UX professionals have not yet learned how to use AI to perform tasks that are specific to their profession.
We think that, while there may be some truth to all of these, the last two explanations have implications that shed light on the future development of these generative AI tools.
UX Professionals Do Lots of Writing
Writing is an essential part of UX practitioners’ work, because UX is a communication-focused discipline. We draft research plans, reports, microcopy, UI copy, feature documentation, and planning documents — on top of emails and Slack messages we’re writing every day.
Text-generation tools streamline the workflow of writing. They can create or refine drafts. As a UX professional for whom English is a second language, I can also attest that text-generation tools greatly help nonnative speakers.
Dedicated AI Support for UX Is Limited
Beyond the need for writing in UX, AI-based text-generation tools are also far more developed than multimedia-generation tools, or UX design and research apps with specific AI features. Their reliable performance has gained them popularity and has made them more likely to be used by all professionals, including UX professionals.
Currently, general-purpose AI tools like ChatGPT, Bard, or Bing Chat provide, at best, only modest support for UX tasks like heuristic evaluation, expert review, or even thematic analysis. (And our analysis of AI-powered features in UX tools shows that such features are, at the time of writing, quite rudimentary.)
In fact, most (if not all) UX activities require a deep understanding of specific products, users, and domains. In order for general-purpose generative AI tools to help with UX tasks, a lot of background knowledge and contextual information needs to be input into them. The more complicated the tasks are, the more dependencies they will have, and the more information the AI will require to make informed suggestions. This might explain why ideation tasks were mentioned less often by our respondents.
The need to input lots of information into the AI system can make the interaction cost too high and may prevent practitioners from even attempting UX tasks with AI.
For instance, while 6% of respondents reported having the AI generate proto personas, only 0.5% used it to create user journeys — likely because the latter activity requires that professionals import many more details before the AI can yield satisfying results.
Plus, data privacy and security constraints prevent many organizations from sharing information with a general-purpose AI tool. Several respondents mentioned this issue:
“I couldn't be very specific because there was confidential info in my project, so I had to do a lot of workshopping with what Bard gave me.”
“I wish it did not use the information entered for training. I would be more likely to use if it was confidential.”
Additionally, generative-AI tools can produce incorrect answers that sound plausible. A few respondents complained that, even when they succeeded in using AI for data analysis, the results weren’t entirely right.
Limited AI Proficiency
Creating prompts that produce helpful results is not easy. It can take many attempts and finetuning. Many UX professionals do not have the expertise to bridge the articulation barrier yet. Respondents in our survey typically had to try multiple times to get the bot to produce the content they wanted. Some of them ended up giving up altogether.
“A big complaint — I was writing a protocol and wanted to succinctly edit my questions prior to stakeholder reviews. No! ChatGPT did not meet my expectations and I wrote the protocol on my own. It was too formal. Even when included in the prompt, the language is still too formal.”
How Can Generative AI Help UX Professionals?
Here are a few ideas for making generative UI more useful to UX professionals. Some companies are already working on some of these, and we look forward to seeing their impact.
- Integrate AI into existing UX tools. Instead of forcing users to go to an AI tool for UX-specific tasks, generative AI should be seamlessly integrated into existing UX tools to streamline workflows. For example, many professionals use Figma AI plugins for their work; more such integrated AI features would save them context-switching effort and would increase efficiency.
- Make it easy to import relevant resources into AI tools. Currently, to get a helpful answer, lots of details and background knowledge need to be shared with the AI in the prompt. Supporting different input formats can make this communication process more straightforward. ChatGPT recently added the ability to accept image input in addition to text, which is a considerable improvement. We can imagine a specialized data-analysis tool (like Dovetail) automatically using the context of a project and its data files as part of the input for a specific prompt.
- Set transparent data-sharing policies and ensure that users can choose not to use confidential project information to train generative AI models and can control how their input data is used.
- Increase the AI literacy of UX professionals. Crafting effective prompts is hard; the AI tools should expedite this process. Create analysis templates for top research and design activities and request corresponding data proactively to help people generate good results quickly. (This is something that Nielsen Norman Group intends to help with in future articles and videos.)
We are still at the very beginning of the AI era. The UX field needs guidance about the best ways in which AI can be used efficiently, accurately, and responsibly.
