Writing with AI
This section outlines the principles to follow whenever AI is used in the design process.
When to use AI
Use AI-generated copy when the context is clear, the outcome is low-risk, and users stay in control of what's produced or displayed. AI works best for tasks that are repetitive, scalable, or informational rather than judgment-based.
- Personalization at scale
Examples: auto-generated subject lines or greetings based on user input ("Hi Alex, here's your lead summary") - Functional assistance
Examples: autocomplete or predictive text in a search bar, conversational inquiry in a chatbot - Summaries
Examples: AI condensing a chat history, transaction log, or long article for quick reference - Low-stakes automation
Examples: SEO content, placeholder text for mockups, bulk generation of test data, or internal-only prototypes
When to avoid AI
Use human-created copy when accuracy, empathy, or trust are essential. If an error, omission, or misinterpretation could confuse users, break compliance, or harm credibility, rely on people—not automation.
- Policy and compliance
Examples: terms of service, privacy notices, consent flows, safety disclaimers - Brand and tone-critical content
Example: product value propositions - Edge cases where empathy matters
Examples: error states involving data loss, accessibility guidance, or communications after a failure - Trust-critical UI text
Examples: text that explains how AI itself is working, language that describes our privacy policy
How to craft AI-generated content
AI-generated content should follow our voice and tone guidelines. Use conversational, plain language. Keep it clear and concise. Unless it's being used in a chatbot-type experience, avoid personifying the AI by using first-person language (like "we" or "I"). First person may be used in a chatbot-type experience, but default to using plural ("we") rather than singular ("I").
How to label AI-generated content
Disclosing AI use helps us meet legal requirements and build user trust.
Disclosures and disclaimers
The way we treat our AI use depends on how we're using user inputs and how much control the user has over the experience. Dynamic content should always include clear language like "Powered by AI." For more information on how to treat different use cases, visit our AI design guidelines.
Contextualizing limitations
Because AI output is often incomplete or inaccurate, it's often important to educate users on what data is used to generate the content. Disclaimers that provide this information help empower users to spot biases, inaccuracies, etc. — and to make changes for a more complete and accurate experience.
Principles for using AI
- Be intentional. Consider when and why to use AI — it's a tool, not a default.
- Be consistent. Keep human standards for voice, tone, and accuracy.
- Be diligent. Different use cases may require different disclosures.
- Be accountable. People stay responsible for truth, clarity, and impact.