Design Guidelines
Twelve NNG-derived guidelines for AI chatbot usability, each with a do/don't example.
Design Guidelines
Synthesised from Nielsen Norman Group research on AI chatbot usability. Each guideline includes a concrete do/don't example drawn from the Adres property context.
Disclose the AI upfront
Always identify that users are talking to an AI, not a human. Disclose limitations in the opening message to set accurate expectations.
Do
Don't
Show what the AI can do
Use the opening message and context-aware suggested prompts to communicate capability. Don't make users guess what to type.
Do
Don't
Offer prompts as buttons
Suggested questions should be clickable chips, not plain text instructions. Reduce typing effort and guide users toward successful interactions.
Do
Don't
Keep chat accessible everywhere
Users expect the chatbot to persist across pages, especially during multi-step flows like property search and enquiry.
Do
Don't
Include images and rich content
Display property photos, floor plans, maps, and comparison tables alongside text. Visuals make AI recommendations significantly more useful.
Do
Don't
Use progressive disclosure
Allow users to expand additional detail rather than triggering a new chat response. Prevents excessively long conversations and loss of context.
Do
Don't
Don't autoscroll responses
Never automatically scroll users to the end of a long response during streaming. Users must control their own reading position.
Do
Don't
Allow resize and expand
Default chat windows are too small for rich content. Provide a maximize button and let users resize, particularly when viewing property details.
Do
Don't
Save and share conversations
Users want to reference AI-generated comparisons, shortlists, and guides later. Provide copy, export, and share options throughout the conversation.
Do
Don't
Support voice input
Offer voice-to-text for accessibility and convenience. Critical for users with motor impairments or during property viewings when hands are occupied.
Do
Don't
Handle ambiguity gracefully
Tolerate typos, vague phrasing, and mid-flow topic changes. Infer intent and confirm when uncertain rather than returning a hard error.
Do
Don't
Offer a clear path to a human
When AI fails, be honest and immediately offer an escape hatch — a real agent, a phone number, or a different support channel. Never leave users stuck.