Most conversations about AI in design swing between two extremes: either the tools are about to replace the profession, or they are toys that produce generic mush. Working designers know the truth sits somewhere less dramatic. AI has become a set of fast, tireless assistants that are very good at some stages of a project and quietly risky at others. The useful question is not “should I use AI?” but “at which step, for what, and who checks the result?”
This guide walks through a typical project, from first brief to handoff, and marks where AI earns its place and where your judgement remains the product the client is paying for.
Stage 1: research and moodboards
Moodboarding is where AI feels most natural. A designer who used to spend an afternoon pulling references can now generate dozens of visual directions in minutes: color moods, lighting styles, textures, compositions. That speed is genuinely valuable at the start of a project, when the goal is to open up options rather than close them.
What AI does well here:
- Volume. Twenty quick directions to react to, instead of three you had time to assemble.
- Unblocking. A rough generated image can help a client articulate what they mean by “premium but approachable” faster than a paragraph of adjectives.
- Combining constraints. Asking for “editorial layout, warm palette, lots of negative space” produces a starting point that would take longer to find in a stock library.
What it does badly: AI moodboards drift toward the statistically average. The same polished, slightly cinematic look turns up again and again, which is exactly why so many AI-assisted brands start to resemble each other. Use generated boards to provoke a conversation, then curate hard. The curation, deciding which direction fits this brand and this audience, is the design work.
Stage 2: image generation in production
Moving generated images from the moodboard into the final deliverable is a much bigger step, and it deserves a deliberate decision rather than a default.
Three questions to ask before a generated image ships:
- Does it hold up at full size? Hands, text inside images, product details and architectural lines are still frequent failure points. Zoom in before the client does.
- Is it consistent? A single hero image is easy. A set of twelve images that share a character, a lighting setup and a style across a campaign is still hard, and inconsistency reads as low quality.
- What are the usage terms? Each tool has its own terms of service, and they differ on commercial use, ownership claims and indemnification. Some vendors, Adobe with Firefly for example, market their models as trained on licensed content specifically to reassure commercial users. Read the terms of the tool you actually use, and tell the client which tool produced which asset.
There is also a legal dimension that clients increasingly ask about. In the United States, the Copyright Office has taken the position that copyright protects human authorship, and that material generated entirely by a machine from a prompt is not protected on its own. In practice, that means a brand built mostly on raw generated imagery may have weaker protection over those visuals than one built on commissioned work. This is general information, not legal advice; for anything that will become a core brand asset, involve the client’s counsel.
Stage 3: copy, microcopy and content structure
Designers write more than they admit: button labels, error messages, onboarding hints, placeholder text that somehow ships. Language models are useful drafting partners here.
| Task | Where AI helps | What you still own |
|---|---|---|
| Placeholder content | Realistic text lengths instead of lorem ipsum, so layouts are tested against real copy | Making sure placeholders never reach production |
| Microcopy variants | Ten options for a CTA or empty state in seconds | Choosing the one that matches the brand voice and the user’s state of mind |
| Error messages | A clear first draft that explains what happened | Accuracy: the message must match what the system actually does |
| Content audits | Summarizing long pages, spotting duplicated sections | Deciding what to cut, merge or rewrite |
| Accessibility text | Draft alt text for images | Checking it describes purpose, not just appearance |
The pattern is the same in every row: AI produces options fast, and a person chooses, verifies and takes responsibility. Generated microcopy is fluent but often generic, and fluency can hide errors. An error message that confidently says “your payment failed” when the real issue is an expired session is worse than no message at all.
Stage 4: production chores
This is the least glamorous and probably the most profitable use of AI for a working designer. Renaming layers, generating size variants, removing backgrounds, writing first-pass specs, converting rough notes into a tidy handoff document: these tasks eat hours and require little creative judgement.
If you are deciding which assistants deserve a slot in your stack, a comparison resource such as this top 10 AI tools ranking is a sensible starting point to shortlist general-purpose assistants before you test them on your own files. Run any shortlisted tool on a real past project, not a demo, and time the result against your usual method. If the saving is not obvious within a week, the tool is not earning its subscription.
For a closer look at specialist design tools, from UI generators to design-to-code bridges, see our guide to the AI design tools actually worth using.
Where human judgement stays essential
Some decisions should never be delegated, however good the tools become.
Strategy and positioning. AI can summarize a market, but it cannot sit in a room with a founder and notice that the stated goal and the real goal are different. Briefs are negotiated, not generated.
Taste and coherence. A design system is a long chain of small, connected choices. AI suggests each link in isolation; a designer makes sure the chain holds. This is the reason AI-only identity work tends to fall apart at the edges, a point we explored in AI logo generators versus a designer.
Usability. Generated interfaces often look plausible and test poorly. Research from organizations such as the Nielsen Norman Group, which maintains a study guide on using AI for UX work, consistently frames AI as a tool to support researchers and designers, not to replace observation of real users. Nothing generated replaces watching someone struggle with your checkout flow.
Ethics and accountability. Who appears in the imagery, what the copy promises, whether a pattern nudges people against their interest: these are judgement calls with consequences, and the client’s name is on the result.
A simple operating rule for your studio
If you want one principle to write on the wall, use this: AI drafts, designers decide. In practice that means:
- Use AI freely for exploration, variants and production chores.
- Treat every generated asset that reaches a client as unverified until a person has checked it at full size and in context.
- Keep a short log of which tool produced which deliverable, along with the terms you relied on.
- Be transparent with clients about where AI was used. Most do not mind; they mind finding out later.
The takeaway
AI has not made design judgement less important; it has made it more visible. When anyone can generate a hundred options, the value moves to knowing which one is right, why, and what could go wrong with it. Designers who build their workflow around that shift spend less time on chores and more time on the decisions clients actually remember.
