HAX Design Library

Guideline 1

Make clear what the system can do

Help the user understand what the AI system is capable of doing.

Guideline 1 > Pattern 1A

Pattern 1A: Introductory blurb

Guideline 1 > Pattern 1A > Example

Copilot in PowerPoint | 1A: Introductory blurb

Guideline 1 > Pattern 1A > Example

Copilot in Word | 1A: Introductory blurb

Guideline 1 > Pattern 1A > Example

Microsoft Word | 1A: Introductory blurb

Guideline 1 > Pattern 1B

Pattern 1B: Use explanation (G11) patterns

Guideline 1 > Pattern 1C

Pattern 1C: Expose system controls

Guideline 1 > Pattern 1C > Example

Copilot in PowerPoint | 1C: Expose system controls

Guideline 1 > Pattern 1D

Pattern 1D: Demonstrate possible system inputs

Guideline 1 > Pattern 1D > Example

M365 Chat | 1D: Make clear what the system can do

Guideline 1 > Pattern 1D > Example

Copilot in PowerPoint | 1D: Demonstrate possible system inputs

Guideline 1 > Pattern 1D > Example

Business Chat | 1D: Demonstrate possible system inputs

Guideline 1 > Pattern 1E

Pattern 1E: Show a set of system outputs

Guideline 1 > Pattern 1E > Example

Word Online | 1E: Show a set of system outputs

Guideline 1 > Example

Notion | G1: Make clear what the system can do

Guideline 1 > Example

Business Chat | G1: Make clear what the system can do

Guideline 1 > Example

Copilot in Excel | G1: Make clear what the system can do

Guideline 2

Make clear how well the system can do what it can do

Help the user understand how often the AI system may make mistakes.

Guideline 2 > Pattern 2A

Pattern 2A: Match the level of precision in UI communication with the system performance – Language

Guideline 2 > Pattern 2B

Pattern 2B: Match the level of precision in UI communication with the system performance – Numbers

Guideline 2 > Pattern 2C

Pattern 2C: Report system performance information

Guideline 2 > Pattern 2D

Pattern 2D: Provide low performance alerts

Guideline 2 > Pattern 2D > Example

Business Chat | 2D: Provide low performance alerts

Guideline 2 > Pattern 2D > Example

Copilot in Outlook | 2D: Provide low performance alerts

Guideline 2 > Pattern 2D > Example

Word Online | 2D: Provide low performance alerts

Guideline 2 > Example

Google Gemini | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Google Gemini | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Visual uncertainty highlighting | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

AI Critiques | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Hugging Chat | G2: Make clear how well a system can do what it can do

Guideline 2 > Example

Copilot in Word | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Copilot in PowerPoint | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Google Bard | G2: Make clear how well a system can do what it can do

Guideline 2 > Example

Hugging Chat | G2: Make clear how well a system can do what it can do

Guideline 2 > Example

Business Chat | G2: Make clear how well the system can do what it can do

Guideline 2 > Example

Copilot in Outlook | G2: Make clear how well the system can do what it can do

Guideline 7

Support efficient invocation

Make it easy to invoke or request the AI system’s services when needed.

Guideline 7 > Example

Copilot in Word | G7: Support efficient invocation

Guideline 7 > Example

Copilot in Outlook | G7: Support efficient invocation

Guideline 8

Support efficient dismissal

Make it easy to dismiss or ignore undesired AI system services.

Guideline 8 > Example

Copilot in Outlook | G8: Support efficient dismissal

Guideline 8 > Example

Copilot in Word | G8: Support efficient dismissal

Guideline 9

Support efficient correction

Make it easy to edit, refine, or recover when the AI system is wrong.

Guideline 9 > Pattern 9A

Pattern 9A: Switch classification decisions

Guideline 9 > Pattern 9B

Pattern 9B: Rich and detailed edits

Guideline 9 > Pattern 9B > Example

Copilot in PowerPoint | 9B: Rich and detailed edits

Guideline 9 > Pattern 9B > Example

Notion | 9B: Rich and detailed edits

Guideline 9 > Pattern 9B > Example

Copilot in Outlook | 9B: Rich and detailed edits

Guideline 9 > Pattern 9B > Example

Pattern 9E: Batch-editing data

Guideline 9 > Example

Copilot in PowerPoint | G9: Support efficient correction

Guideline 10

Scope services when in doubt

Engage in disambiguation or gracefully degrade the AI system’s services when uncertain about a user’s goals.

Guideline 10 > Pattern 10A

Pattern 10A: Disambiguate before acting

Guideline 10 > Pattern 10A > Example

Copilot in Word | 10A: Disambiguate before acting

Guideline 10 > Pattern 10A > Example

iPhone | 10A: Disambiguate before acting

Guideline 10 > Pattern 10B

Pattern 10B: Avoid cold starts by eliciting user preferences

Guideline 10 > Pattern 10C

Pattern 10C: Fall back to other strategies

Guideline 10 > Example

Interactive collection of the 18 Guidelines for Human-AI Interaction, with design patterns for applying them and examples.

Guideline 1

Make clear what the system can do

Pattern 1A: Introductory blurb

Pattern 1B: Use explanation (G11) patterns

Pattern 1C: Expose system controls

Pattern 1D: Demonstrate possible system inputs

Pattern 1E: Show a set of system outputs

Guideline 2

Make clear how well the system can do what it can do

Pattern 2A: Match the level of precision in UI communication with the system performance – Language

Pattern 2B: Match the level of precision in UI communication with the system performance – Numbers

Pattern 2C: Report system performance information

Pattern 2D: Provide low performance alerts

Guideline 7

Support efficient invocation

Guideline 8

Support efficient dismissal

Guideline 9

Support efficient correction

Pattern 9A: Switch classification decisions

Pattern 9B: Rich and detailed edits

Pattern 9C: Undo automated actions

Pattern 9D: Do G9 through G15

Pattern 9E: Batch-editing data

Guideline 10

Scope services when in doubt

Pattern 10A: Disambiguate before acting

Pattern 10B: Avoid cold starts by eliciting user preferences

Pattern 10C: Fall back to other strategies