AI UX PlaygroundNewsletterJoin 2K+ AI designers and PMs on Substack. New teardowns, patterns, and prompts as they drop.

Chatbot

Conversation History Search

Find past chats and turns by keyword, meaning, or filters like date, project, or tag. Long archives become something you can retrieve, not something you scroll forever.

Interactive demo

Recent chats appear here when you search.

Overview

How might we design conversation history search so people can trust and act on AI output?

When to use

  • Perfect for power users, researchers, and professionals who need to reference past conversations and build upon previous AI interactions.

When to skip

  • Ephemeral single-session tools with no saved history.
  • Products that already store the same content in a dedicated docs or ticket system.
  • Regulated contexts where searching chat history would expose data that must stay siloed.

Rules

  • Search that only matches titles and ignores message bodies.

  • Results that jump into a chat with no highlight of the matching turn.

  • Semantic search with no keyword fallback when users know an exact phrase.

  • Exposing other users or workspace chats in personal search results.

Evidence

ProductImplementation
ChatGPTSearch across conversation titles and message content.
ClaudeHistory browsing with find-in-thread style retrieval.
Notion AISearch that spans AI chats and workspace pages.
ObsidianFull-text search across saved AI-assisted notes and chats.

FAQ

Why do AI products need conversation search?

Chats pile up fast. Search lets users recover decisions, prompts, and answers weeks later without scrolling a long sidebar.

Should search be keyword or semantic?

Offer both when you can. Keyword wins for exact errors and IDs; semantic helps when users remember the topic but not the words.

What should a result show?

Chat title, matching snippet, date, and enough context to know why it matched. Clicking should land on the highlighted turn.

How does this differ from memory management?

Search retrieves past conversations on demand. Memory management controls what the model automatically recalls next time.