# AI-Assisted Design

AI-assisted design uses generative or predictive systems to support design work while keeping human judgment, verification, accountability, and user rights central.

- Type: practice
- Domain: Leadership & teams
- Level: Advanced
- Last reviewed: 2026-09-21
- Original: [Soroush’s Product Wiki](<https://wiki.soroush.uk/topic/ai-assisted-design/>)

## What it means

AI-assisted design is the use of artificial intelligence to help with research synthesis, ideation, content, prototyping, evaluation, production, or design operations. The system may accelerate parts of the workflow, but the design team remains responsible for problem framing, source quality, validation, accessibility, privacy, intellectual property, risk decisions, and the consequences of what ships.

## Why it matters

AI can expand options and reduce repetitive work, yet it can also fabricate evidence, reproduce bias, expose confidential data, flatten original thinking, and produce plausible but inaccessible interfaces at scale. A responsible practice treats AI output as untrusted material to examine, not authority, and reserves meaningful decisions for accountable people.

## When to use it

Use AI where assistance is proportionate, inputs are permitted, outputs can be checked, and failure is recoverable. Apply stricter controls—or avoid it—for sensitive research, personal data, safety-critical decisions, regulated content, identity inference, high-impact personalization, or work where provenance and consent cannot be established.

## How it works

1. Define the task, expected benefit, affected people, failure modes, prohibited inputs, human decision owner, and evidence needed to verify output.
2. Select tools using privacy, security, accessibility, data-governance, provenance, model-limit, vendor, and retention criteria—not novelty alone.
3. Minimize and protect inputs; never upload participant data, confidential strategy, or copyrighted material without appropriate authority and controls.
4. Review outputs for factuality, bias, representational harm, accessibility, originality, and fit with research; trace important claims back to primary evidence.
5. Document material AI use, test the resulting experience with affected people, provide oversight and recourse, monitor outcomes, and stop use when risks exceed value.

## A practical example

A design team uses a language model to generate alternative help-center outlines from de-identified themes. Researchers compare every theme with source notes, content designers verify policy and tone, accessibility specialists review structure, and no participant transcript enters the tool. The team records AI involvement and rejects invented explanations before publishing.

## Where it breaks

Model behavior, training data, and vendor practices may be opaque or change without notice. Human review can become a rubber stamp, especially under time pressure, and reviewers may miss fluent errors or harms outside their experience. AI cannot grant consent, establish truth, represent absent users, or absorb accountability for a shipped decision.

## Common mistakes

- Uploading raw research or confidential material to a tool without authorization, minimization, or retention controls.
- Presenting generated personas, quotes, usability findings, or market evidence as if they came from real people.
- Measuring time saved while ignoring correction work, sameness, accessibility, bias, rights, and downstream user harm.

## Visualization

[View the visualization on the website](<https://wiki.soroush.uk/topic/ai-assisted-design/#visualization-ai-assisted-design>)

## Sources

1. [Artificial Intelligence Risk Management Framework \(AI RMF 1.0\)](<https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10>) — National Institute of Standards and Technology · Government framework
2. [People + AI Guidebook](<https://pair.withgoogle.com/guidebook/>) — Google People + AI Research · Design guide
3. [OECD Principles on Artificial Intelligence](<https://oecd.org/going-digital/ai/principles>) — OECD · Intergovernmental principles

## Related topics

- [DesignOps](<https://wiki.soroush.uk/topic/design-ops/>)
- [Product Discovery](<https://wiki.soroush.uk/topic/product-discovery/>)
- [Accessibility](<https://wiki.soroush.uk/topic/accessibility/>)
