AI tools are already part of many designers' daily work. They can save hours of routine, but they can just as easily replace thinking with template solutions. This guide is about using AI as an assistant rather than an author: where it genuinely helps, where it gets in the way, and how to keep your own judgment.
These tools change fast: features, limits and terms of use are updated all the time. So here we talk about types of tasks and approaches, not specific versions or pricing. Always check a service's current terms before you use it.
Where AI really helps
1. Research synthesis. Language models such as ChatGPT or Claude are good at a first pass of grouping large amounts of text: interview notes, open-ended survey answers, user reviews. Ask the model to group statements by theme and quote the original words for each theme. The quotes are key: they let you check that the model hasn't invented a pattern that isn't there.
2. Copy drafts. Microcopy — button labels, error messages, empty states, onboarding — is often written at the last minute. AI can quickly give you ten options to choose from and refine. Give it context: who the user is, what just happened, the product's tone of voice, the maximum line length.
3. Image generation. Services like Midjourney are useful for mood boards, exploring a visual direction, placeholder illustrations and early concepts. They help you quickly show the team “roughly this mood” without spending a day hunting for references.
4. UI generation and Figma plugins. There are tools and plugins that generate interface drafts from a text description, fill layouts with realistic data, rename layers or help with routine operations. For example, in April 2026 we hosted a charity workshop on the AI tool Figmosha, led by its author Denys Osadchyi, about how such tools fit into a design workflow. Details are on the workshop page.
5. Technical odds and ends. Explaining an unfamiliar CSS property, writing a simple spreadsheet formula for survey results, translating an interface draft into another language for a first check: AI does all of these quickly and well enough to get started.
The quality of the answer almost always depends on the request. A good prompt for a design task includes four things:
- Role and context: what the product is, who the user is, what stage the work is at.
- A specific task: not “help with onboarding” but “suggest five headline options for the first screen, up to 40 characters”.
- Constraints: tone of voice, language, length, what must not appear.
- Answer format: a list, a table, options with a short explanation for each.
Where AI doesn't help, or does harm
- Understanding real people. A model can't replace a conversation with a user. “Synthetic personas” or “answers on behalf of users” are generalisations from training data, not your people in their context.
- Product decisions. AI doesn't know your constraints, business goals or what your team has already tried. It suggests the average, not what's right for your situation.
- Facts and figures. Language models can confidently invent statistics, sources and quotes. Check any number you plan to show someone against the original source.
- Original visual style. Generated interfaces often look alike. Accept them unchanged and your product will look like everyone else's.
- Accessibility and state details. Generated layouts often ignore contrast, focus, errors, long text and empty states. You'll have to finish these by hand.
Ethics and copyright
The legal landscape around generated content is still taking shape and differs from country to country. So it's worth taking a cautious approach:
- Don't upload confidential data to public AI services: NDA work, users' personal data, internal documents. Check how the service uses your input and whether you can turn that off.
- Don't ask for work “in the style of” a specific living artist for commercial use: it is ethically questionable and may have legal consequences.
- Before using generated images in a real product, read the service's terms on rights to the output.
- Be open about where you used AI: in your portfolio, in a competition, in work for your team. It's a matter of trust.
- Check AI-generated logos and brand elements for similarity to existing marks.
How to keep your own judgment
The main risk of AI for designers isn't that it will “take the job” but that it will quietly train you out of thinking. A few practices that help avoid this:
- Thought first, prompt second. Before asking AI, write down your own hypothesis or make a quick sketch. Then compare it with what the model suggests.
- Generate options, not an answer. Ask for five different approaches and pick one yourself, explaining to yourself why that one.
- Check everything that leaves your desk. Facts, quotes, numbers and interface copy are your responsibility, not the model's.
- Keep your core skills in shape. Now and then, do a task entirely without AI: typography, grid, illustration. This shows you where the tool helps and where you've started to lose the skill yourself.
- Test with people. A generated layout is as much a hypothesis as a hand-drawn one. It still needs to be shown to users.
Practice: the community AI challenge
The best way to understand a tool's limits is to try it on a task with a deadline. From 15 to 28 May 2025, Spaceberry Community ran an AI challenge: participants built a landing page whose logo, copy and visuals all had to be AI-generated. Four entries came in, and the winner was chosen by the community in an anonymous vote.
This kind of format shows the main point well: AI generates material, but assembling it into a coherent, clear and convincing product is still the designer's job. Selection, composition, hierarchy and editing stay with the human.
Checklist: is AI a good fit here?
- I understand the task and have my own hypothesis before I prompt.
- I'm not passing confidential or personal data to the service.
- I know how I'll verify the result: an original source, a test with people, a colleague's review.
- I'm ready to rework the result substantially rather than use it as is.
- I know the terms for using the output and have the right to use it.
- I'll be open about using AI where it matters.
AI is a powerful tool, but design starts with understanding people and ends with responsibility for decisions. For now, both of those are entirely on you. You can discuss tools and share your experience in Spaceberry Community.