Intro

AI is reshaping both the words we ship and the way we write them. This page gives a quick tour of two areas: making product content work for people and AI agents, and using generative tools in your content workflow.

AI-Ready Product Content

SAP is an AI-first company. In the future, AI agents like Joule will act independently within SAP products: navigating the UI, triggering actions, and interpreting status information autonomously. This means that every UI text potentially has two audiences: the human reading the screen, and the agent operating it.

Writing for agent readiness isn't a separate discipline. It's a natural extension of writing clearly. The same precision that helps a busy user scan a screen quickly is what lets Joule navigate it reliably.

Apply these checks to every screen:

• Give every interactive element a unique, unambiguous label. If two buttons on the same page share the same label (two Edit buttons, two Save buttons), then an agent can't tell them apart. Use specific labels (Edit Address, Edit Contact) or add an ARIA label that provides the missing context.
• Make action labels verb-first and self-explanatory. An agent reads button text in isolation, without seeing what's around it. Save is unambiguous. Confirm raises the question: confirm what? OK says almost nothing. Use specific verbs that name the action and its object: Save Address, Submit Request, Approve Invoice.
• Don't let icons or position carry meaning on their own. Every interactive element that has no visible text label must have an ARIA label that names the action explicitly. Without it, an agent has no way to identify or invoke the element.
• Use controlled, consistent vocabulary for status and state. In Progress, Approved, and Pending Review are machine-readable. Long, free-form prose is not. Keep your status vocabulary finite and consistent across the product.
• Write labels that stand alone. An agent processes a label without the surrounding section heading, card color, or icon that gives it meaning on screen. Details is ambiguous. Shipment Details and Contact Details are not. If a label only makes sense in context, it needs to be more specific.

Using AI to Generate Text

Generative AI can be a great help in producing product content that aligns with our priorities and standards. With the right context and prompts, AI can help you to achieve and maintain the SAP personality of the approachable expert. AI works best for straightforward, low-risk scenarios like drafting standard patterns, checking for linguistic correctness, and enforcing style guide rules.

Never use AI-generated UI text or system messages without careful human review.

Use cases and advantages

• Brainstorming: AI excels at generating ideas and can help you explore options you might not have considered. This support can break through creative blocks and surface unexpected solutions for UI text.
• Drafting: Use AI to quickly generate draft text as a starting point. Provide the LLM with as much context and information as possible to ensure accurate results.
• Multivariate testing: The speed of GenAI makes it easy to generate numerous variations of a UI string or message. Use AI to explore different approaches, rapidly create a range of prototypes for comparison, or conduct usability testing with internal or external users.
• Shortening text: When UI space is limited, AI can efficiently shorten text to comply with the available character count.
• Style guide adherence: AI helps you maintain SAP writing standards and guidelines. Check the UAi Booster for current prompts for UI text, such as error messages.
• Tone adjustment: The most recent LLM models have an extremely nuanced grasp of English and can help adjust tone to a specific audience, context, or use case. Ask the LLM to make your content more professional or more conversational; more engaging or more formal; more empathic or more task-focused.
• Terminology consistency: Use AI to review terminology consistency in i18n files using GitHub Copilot or by uploading the entire i18n file to an LLM for review.
• Localization support: Ask AI to identify potential localization issues, such as text strings that may expand problematically in certain languages. It can also flag messages or UI strings that work well in one language or culture but not in others.
• Accessibility and inclusion: Use AI to check for common issues that can compromise the accessibility or inclusivity of your text.

Responsible use

The current limitations of using AI to generate UI text include:

• Brand voice erosion: AI mimics common patterns that it finds on the internet or in its training data. As a result, when we use AI to generate content, we risk compromising the voice of the approachable expert that is unique to SAP. If you use GenAI to draft UI content, be rigorous in analyzing whether the suggested text truly embodies the SAP product voice.
• Tone: You may find that AI suggestions for interface text are overly chatty or too casual. Keep in mind that the SAP brand voice is professional, but not overly colloquial. Adjust the AI content suggestions as needed.
• Lack of empathy: The ability to empathize with users is at the heart of truly excellent UX writing. Review suggestions from AI and consider whether they really meet the needs of users.
• Terminology: When you write AI prompts, specify exactly which terms and patterns must appear in the output. Likewise, when you review AI-generated content, pay careful attention to language and terminology. Make sure that the text adheres to the product standard UXC-015.

Related Links