As designers, it is our mission to carefully craft our users’ experience and design for delight. But with AI entering our products, we’re losing a lot of control. We need a new set of design patterns to keep guiding and delighting our users. Luckily, AI, and especially generative AI, also comes bearing gifts. It empowers designers to approach their work with inclusivity in mind, requiring minimal effort and making accessibility the natural default.

Before AI entered the picture, most products were deterministic: what you clicked was exactly what you got. Designers worked out happy paths and guided users seamlessly through a product's experience. But when core features start relying on AI, this predictability changes. AI is rooted in probabilities and does not always produce the same output from identical inputs. Yes, it excels at handling nuanced, complex tasks where no single clear-cut outcome exists. But its behaviour can also be quite unpredictable. 

Make UX great again

This shift brings both opportunities and challenges for User Experience (UX). On the one hand, AI offers the potential to enrich and personalise interactions, but also carries the risk of making unexpected or suboptimal decisions on the other.

To address this tension and be able to craft the best possible AI experiences, we developed a brand new set of design patterns. We made sure that they both embrace AI’s possibilities and mitigate the risks it brings. And that they stand on the shoulders of giants, by building upon the existing wealth of knowledge about what makes UX great.

Come together, right now

You might be tempted to think that this is merely a design problem. No offense, but you’d be wrong. AI demands a fundamentally different mindset and approach, one that can only be effectively tackled through collaboration within a cross-functional team. Crafting outstanding, user-centric and safe AI products is only possible through a powerful team effort. That’s because there are a lot of pitfalls and trade-offs that emerge from the selection of the models, the adaptation of the prompts and the building of the architecture l. A close collaboration between business, design and engineering allows you to navigate those trade-offs carefully and thoughtfully. Reworking a prompt could give an equally good result for a fraction of the cost and with a shorter response time. These types of evaluations can only be done when you have all perspectives around the table.

Solutions for recurring problems

In the 70s, the renowned architect Christopher Alexander proposed a set of solutions or so-called “patterns” for recurring architectural problems  in his book ‘A Pattern Language’. For instance, the pattern "Streetside Café" provides a solution to the common issue of deserted, lifeless residential city areas during the day by creating a space where people can relax, be visible, and observe the world around them.

The idea of patterns caught on and was later used in software development as well as in design. Design patterns, for instance, are user interface elements that solve a recurring problem. Do you need a way to easily enter a date? A date or calendar picker can be the perfect pattern for that.

Design patterns are typically quite stable, but they can evolve very quickly in new areas. AI products are a prime example of this. As designers, we need to carefully listen and test our products to stay up-to-date with what’s expected. User expectations and patterns will develop along with the field of AI-based products.

We’ve listed a selection of the AI design patterns we currently use to help users understand how to interact with the product and ensure the best possible experience. Our AI guidelines and patterns are a combination of our own work, the wonderful Shape of AI and Google’s PAIR guide.

Be transparent about AI

Good fences make good neighbours. AI hallucinates and makes mistakes. Just like us humans, it is not infallible. Own up to that. Clearly communicate the limitations of your AI product with an “AI notice”. Adding a "Preview" or "Experiment" badge with consistent styling can effectively signal to users that they need to approach the results with caution.

Provide suggestions

Inspire your users with suggested prompts or actions that clearly showcase how your product can be used. Help them overcome any prompt-writing hurdles by providing practical, hands-on examples that guide them forward. People just don’t have the time to figure out your product on their own—set them up for success with a head start.

Use templates

Simplify the user experience by using templates or intuitive form elements that guide users toward their desired results without requiring them to type out each word. AI models thrive on context and templates make it easier for users to provide the necessary information faster. Wherever possible, reduce the need for lengthy, complex prompts by turning frequent inputs into convenient shortcuts that appear when they’re needed.

Keep users engaged

Depending on the model and your architecture, AI responses can take a longer time to appear. Give your users something to read or keep them engaged with status updates - like smooth and continuous animations - while the response is loading. Be transparent about what is happening behind the scenes, as with Midjourney’s fuzzy, loading image states. ChatGPT shows a continuous circle animation to help users know that the answer isn't complete yet. These approaches manage the user’s expectations, if the response takes a while.  Also, always try to avoid jargon or technical terms.

Make uncertainty explicit

AI makes mistakes and can reach undesired outcomes. That’s the nature of the game. If your product is dealing with a lot of uncertainty, be transparent about it through the user interface (UI). Look at human interaction for inspiration: we use tone of voice and specific wording to indicate we may not be certain about something. Google’s NotebookLM is one of the few products to currently nail this and admit when it can’t provide a good answer to a question.

Think carefully about icons & mascots

Sparkles, magic wands and orbs are most often associated with AI features. Robots are most associated with co-pilots or assistants. But this is changing fast and has become a bit of a faux pas lately. Notion AI really popularised the sparks icon but has since moved to a scribbly face, a mascot of sorts.

A (non-human) cartoon-like mascot can make your product more fun to use and can make users more forgiving and understanding when things don’t go as planned. Ideally, it’s also gender-ambiguous and ethnicity-ambiguous, in order not to leave anyone feeling excluded. In general, though, it’s better to avoid human-like avatars since AI is not human.

Inclusive design, turbocharged by AI

Inclusive design often gets overlooked, as the extra effort it requires tends to push it to the back burner. Fortunately, that’s about to change. In June 2025, the European Accessibility Act takes effect, requiring many organisations to ensure their products and services are accessible. While AI can add complexity in some places, it also brings exciting opportunities to supercharge inclusive design. Here are six ways AI is already helping us create more inclusive experiences:

Reimagining Subtitles

AI-powered features like automatic video subtitles can offer a solid alternative for people with hearing disabilities. While better than nothing, they often lack punctuation, background audio captions, or sound effects, making it hard to follow the full context. Advanced AI transcripts can address this by analysing audio, creating time-coded transcripts with punctuation and sound descriptions.

Visual Descriptions everywhere, all at once

People with vision disabilities face similar challenges, with the heaps of image and video content created today remaining inaccessible to them. Videos often contain embedded text that screen readers can’t detect, leaving users unaware of the content. AI can help by generating video as well as image descriptions and creating audio description tracks using realistic text-to-speech technology.

Code generators are inclusive by default

Despite our best efforts, humans can overlook accessibility when developing digital products. AI-powered code assistants could address this by default, ensuring essential accessibility code and features aren’t missed. 

Annotate and fix design

In the inclusive design community, it’s often noted that 67% of accessibility issues stem from design. Failing to design accessibly upfront increases future costs and delays. Auditing tools can flag issues early, annotate designs for correct implementation, and ensure design systems include contrast-validated colors, readable fonts, and scalable components that adapt to accessibility settings.

Visual intelligence to understand the world around you

AI-powered tools like Apple’s Visual Intelligence, Google Lens, and ChatGPT’s video features enhance how we interact with the world. By pointing your phone’s camera, these tools can describe objects, solve math problems, or locate items like books on a shelf—offering significant benefits for people with disabilities. As AR devices become more common, these technologies could form the foundation for viewing the world through smart glasses or lenses.

Everyone gets the interface they need

Today, we mostly interact with smart interfaces through personalized feeds, tags, or categories powered by machine learning. In the future, entire interfaces - not just content - could be AI-driven, tailoring every aspect to individual needs. For example, colorblind users could have optimised color themes, those hard of hearing might prefer transcribed articles over podcasts, and users with motor challenges could benefit from larger buttons. Intelligent interfaces could make these customisations seamless, creating truly inclusive experiences.

Creating AI products can feel daunting at first, but remember—you can build on tried-and-true patterns and stand on the shoulders of giants. AI is here to stay, so let’s embrace its challenges and seize the opportunities it offers to design better, more inclusive products.

‍