Paweł Chłodnicki — Content Designer at DocPlanner
Content Design Toolkit. Ways of working with product copy - review and ideas
September 4, 2023
12:00
UX Writing
🇬🇧 English
UX Writing
AI-Powered UX Content Design: Good Product Content for Everyone!
About
This talk isn't about adding to any speculation about AI in our space. We’re here to candidly share what we’ve practically learned from integrating AI into our work as content designers. All of us working in design know about the impact of words and language in our products, but despite content design being a growing discipline, many of us still have to work with no or limited content design resources. That's why I want to share how my team and I use AI as a tool to help ""do the words"" in the design process and improve the overall UX.
I'll share concrete tried-and-tested techniques, prompt frameworks for writing superb microcopy without a writer, and inspiration for how you can use AI to supercharge your own content design skills and help you become a well-rounded full-stack UX designer.
Watch the full talk
Watch this WaysConf session, then continue with related talks or explore the current programme.
From the recording
Talk in brief
At WaysConf 2024, Ben Davies-Romano, Head of Content Design at Klarna, examines how product teams can apply structured content design principles to generative AI prompting. Drawing on his experience as a product manager and design leader, Davies-Romano demonstrates how shallow prompting leads to low-quality interface copy and introduces four core prompt engineering techniques: the Message Context Goal framework for supplying context, acceptance criteria for guiding copy constraints and automated QA, few-shot prompting for brand tone consistency, and Chain of Thought prompting for step-by-step reasoning. Through practical interface examples, he illustrates how non-writers and product creators can build reusable prompt frameworks to consistently produce clear, user-focused microcopy.
Key takeaways
- 01
Surface-Level Prompts Produce Low-Quality Product Copy
Requesting interface text without underlying user context, screen location, or business constraints forces AI to guess, resulting in generic or misleading microcopy.
Watch from 12:40 - 02
Structuring Inputs with Message Context Goal
Breaking prompts into message, context, and goal ensures the AI receives complete information about screen placement, user state, and required actions.
Watch from 15:09 - 03
Validating Microcopy with Acceptance Criteria
Defining clear constraints beforehand enables creators to build structured prompts and run automated quality assurance checks on generated options.
Watch from 20:34 - 04
Maintaining Tone of Voice via Few-Shot Examples
Providing three to five real-world microcopy samples within a prompt establishes style, voice, and formatting more effectively than abstract adjectives.
Watch from 26:36 - 05
Guiding Reasoning Through Chain of Thought Prompts
Demonstrating step-by-step design reasoning using a question-answer-question structure guides AI models to create clear, single-action button labels.
Watch from 31:46
Read edited transcript highlights
These concise notes were edited from automatic captions and checked against the talk structure. They are not a verbatim transcript.
Flaws in Superficial Roleplay and Prompt Shortcuts
Directing AI models to act as experienced writers without giving structural context rarely improves user experience copy. In practice, models simply select more elaborate phrasing that decreases readability and raises cognitive load. Extreme persona instructions can even cause hallucinations, such as fabricating historical company achievements. Interface text generation fails when prompts focus strictly on surface wording while omitting audience state, technical limits, and screen location.
Watch from 9:43Informing AI Inputs with Product Design Thinking
Effective product copy follows the same design thinking process used across research and product design. Before writing interface text, a designer considers feature scope, user intent, progressive disclosure, and overall screen cognitive load. Generative tools struggle when prompts skip these underlying considerations. When inputs reflect core structural choices rather than simple requests for words, generated product strings improve significantly.
Watch from 12:40Structuring Content Inputs with Message, Context, and Goal
The Message Context Goal framework provides a clear structure for generating user interface text. The message defines the core information needing to be conveyed. Context forms the main portion of the prompt, capturing screen position, previous user actions, legal constraints, and character limits. Finally, the goal specifies what the user should understand, feel, or do after reading. Structuring prompts with these three elements provides generative tools with essential design context.
Watch from 15:09Automating Microcopy QA via Acceptance Criteria
Establishing acceptance criteria aligns team members on what interface text must achieve prior to drafting copy. When these criteria are built into a prompt, language models produce text tailored to specific constraints like character counts and required formatting. Designers can follow up by asking the AI to evaluate its output against those criteria, creating an automated quality check that highlights missing elements and guides manual refinements.
Watch from 20:34Establishing Voice and Tone through Few-Shot Examples
Abstract adjectives like conversational or striking often fail to convey brand tone accurately to AI tools. Few-shot prompting resolves this by incorporating three to five real microcopy samples into the prompt. These examples do not need to share the same topic; they simply illustrate desired formatting, voice, and style. By pulling microcopy from existing products or industry benchmarks, teams guide language models toward predictable, brand-aligned text.
Watch from 26:36Guiding Button Copy Logic with Chain of Thought
Chain of Thought prompting utilizes a Question-Answer-Question structure to guide language models through human design reasoning. By showing a similar sample problem alongside a narrative of how a designer moves from draft to final copy, the model learns to mirror that analytical path. Applied to components like buttons, this method avoids common errors like multi-action confusion, producing clear, action-oriented microcopy.
Watch from 31:46
Explore WaysConf 2026
