Summary
AI tools like ChatGPT have exploded in popularity with good reason: they allow users to draft, summarize, and edit content with unprecedented speed. While these generic tools can generate any type of content or perform any type of content task, the user needs to craft an effective prompt to get high-quality output, and often needs to exchange multiple messages with additional guidance and requirements in order to improve results. When you’re building an AI-powered text generation feature, such as a product description or email writer, you typically can’t expect users to craft their own prompts. And unless you’re building a chat interface, you’re unlikely to offer the ability to iteratively improve the output. Instead, your feature needs a robust prompt skeleton that combines with user input to produce high-quality output in a single response. For the designer, this means building an interface that helps users provide the exact information that creates a successful prompt. This process is more complex than simple form design or a mad-lib prompt completion tool. The user input, often including free form text fields, might be required to fill in prompt variables, but it also could change the prompt structure itself, or even override base instructions. The effectiveness of the user input significantly influences the quality of the output, underscoring the need for designers to be deeply familiar with the backend prompt architecture so they can design the frontend. Drawing on recent text generation projects, I'll demonstrate how the interface design can respond to and evolve with the prompt architecture. I’ll talk about how to determine which prompt components to make invisible to the user, which to provide as predefined options, and which should be authored by the user in free-form text fields. Takeaways How prompt structure can impact user interface design and conversely, how design can impact prompt structure Techniques to provide effective user guidance within AI generation contexts to ensure consistently high-quality output Real-world examples and learnings from recent generative AI projects in an e-commerce software product
Key Insights
-
•
Designing AI text generation features requires predefining most of the prompt to ensure quality output, unlike general chat interfaces where users craft their own prompts.
-
•
Users should provide only essential variable inputs like product name, keywords, tone, and length while the rest of the prompt remains fixed and optimized by designers.
-
•
Free-text tone input confuses many users, making predefined tone options a better UX solution for brand-consistent content generation.
-
•
Providing six distinct, well-differentiated tones helps merchants easily select a voice that fits their brand and produce unique product descriptions.
-
•
Tone labels need to cover sufficiently different style zones, avoiding overlaps such as conversational vs friendly or witty vs humorous.
-
•
Detailed linguistic instructions behind each tone dropdown option improve output consistency, including vocabulary choice, pronouns, punctuation, and syntactic style.
-
•
Using AI to detect tone from existing website content is unreliable; the model gives inconsistent outputs with unjustified high confidence.
-
•
Balancing configurability and fixed prompt instructions is a core design challenge when building AI features with limited iterative prompting.
-
•
The UI must simplify complex prompt engineering by presenting users these configuration options in an intuitive and minimal way.
-
•
Generative AI models often default to a confident, positive tone, which designers need to explicitly adjust through persona prompts if a different tone is desired.
Notable Quotes
"If the user isn’t happy with the output, they might be able to change some of the inputs and try again, but the base instructions are predefined."
"When you’re building a specific AI feature, you don’t want to make your user write their own prompt."
"A prompt is a set of directions given to an LLM to generate text that meets criteria like format, length, tone, or topic."
"The tone of voice field turned out to be a much more complicated variable than it appears."
"Merchants often sell really similar products, so tone differentiated suggestions help them stand out with unique descriptions."
"People get stuck trying to articulate their brand voice in just one or two words during user testing."
"The LLM gave many different answers for the same passage and always reported it was 100% confident."
"Tones need to be distinct enough from each other so merchants can easily spot which one fits best."
"We had a quite a bit more detail in the prompt about vocabulary, pronouns, punctuation, and syntactic instructions for each tone."
"By default, generative AI has a very convincing, confident tone that never hedges, and you have to build other tones explicitly."
Or choose a question:
More Videos
"Share everything, own nothing but credit everyone."
Zariah CameronReDesigning Wellbeing for Equitable Care in the Workplace
September 23, 2024
"If anything’s on fire, go fix it is a perfectly valid top priority."
John Cutler Harry MaxPrioritization for designers and product managers (1st of 3 seminars) (Videoconference)
June 13, 2024
"Inviting people from other disciplines to experience user research helped Felipe see why user experience matters."
Nick CochranGrowing in Enterprise Design through Making Connections
June 3, 2019
"I often lock a process against changes for one to two months unless a major issue arises."
Deanna SmithLeading Change with Confidence: Strategies for Optimizing Your Process
September 23, 2024
"Our fellowships bring in private sector people to public service challenges, focusing more on the journey than on delivering final products."
Magdalena ZadaraZero Hour: How to Get Far Quickly When Starting Your Digital Service Unit Late
November 16, 2022
"Despite all the insecurities, most team members were willing to show up every week and challenge what they thought they knew."
Nova Wehman-BrownWe've Never Done This Before
June 4, 2019
"People don’t always get straight to their questions. They might need to do some investigation first before they figure out the question to ask."
Liwei DaiThe Heart and Brain of the AI Research
March 31, 2020
"The Digital Response Unit was a source of situational awareness and air traffic control within the flurry of daily COVID case counts and policy decisions."
Gordon Ross12 Months of COVID-19 Design and Digital Response with the British Columbia Government
December 8, 2021
"We use design thinking to make north stars for products; I help people find their north stars."
Tutti TaygerlyMake Space to Lead
June 12, 2021
Latest Books All books
Dig deeper with the Rosenbot
In what ways can Dick Buchanan’s concept of design as rhetoric change how we approach product design?
How can UX professionals take action to decarbonize digital services if they lack dedicated climate projects?
How can workshops linking software and hardware shed light on application-level carbon impacts?