Back to Glossary
Retrieval UX
Synonyms: search UX, findability design, RAG UX, answer engine UX, information retrieval interface design
Do not index
Definition
Retrieval UX is the design of how people ask a product for information and get the right piece back. It covers search bars, filters, autocomplete, and empty states, plus how AI products pull documents and show where an answer came from. It exists because users only act on answers they can find and check.
Use cases
When retrieval fails, users blame the whole product. They retype the query twice, then email support.
- Zero-results pages: A shopper types "charger for iPhone 15." The catalog says "USB-C power adapter," so search returns nothing.
- Answers from old docs: An internal AI assistant pulls the 2022 refund policy instead of the 2025 one. No one notices until a customer gets the wrong refund.
How it's used in practice
- Sort real queries first: Pull 200 queries from search logs. Tag them by type (exact name, symptom, compatibility, use case) and design for the biggest groups.
- Design the empty state early: Show the closest matches and a way to reach a person.
- Cite each claim: Link every sentence to its source passage. Claude's citations API returns the exact quote for hover previews.
- Ask engineering for the failure rate: Anthropic cut failed top-20 retrievals from 5.7% to 1.9% with contextual indexing plus reranking. Get your number before any UI work.
Put an "Updated" date next to every cited source.
Search ranking matches words and ignores which policy is current. A date lets users catch stale answers.
Challenges & limitations
- Ranking sits with engineering: A clean UI can't fix bad results.
- Citation clutter: Sentence-level sources crowd a 390px phone screen.
- Mockups hide failures: Figma screens show perfect results. Real failures only show up with real data.
Free resources
- How AI Succeeds (and Fails) to Help People Find Information (NN/g) — Kate Moran on where AI tools beat keyword search and where they fall short
- Information Foraging with Generative AI: A Study of 3 Chatbots (NN/g) — 2-week diary study showing users rarely check chatbot answers
- 9 Best Practices for Autocomplete Suggestions (Baymard) — research-backed autocomplete patterns for desktop and mobile
- 5 Search Scope Features (Baymard) — ways to let users narrow a search to one category
