Conversational search

Conversational search is an approach to information retrieval in which a person finds information through a back-and-forth, natural-language dialogue rather than a single keyword query. The system interprets each request in the context of the conversation so far, remembering previous turns, resolving follow-up questions, and refining results as the exchange continues. This allows people to search as though they were talking with a knowledgeable human.

In short: conversational search turns search into a dialogue, allowing people to ask, refine, and follow up in plain language, and the system keeps track of the thread.

How conversational search works

At essence, conversational search works by maintaining context across turns: the system interprets each new message in light of what came before, often reformulating your query behind the scenes (resolving references like “that one” or “the second option”) before retrieving results, then generating a natural-language response and folding your follow-ups back into the same evolving context. In practice it pairs natural-language understanding with retrieval, so the conversation itself becomes the interface for narrowing in on an answer.

The core difference from one-shot search is that meaning is carried across turns (a “turn” being a single back-and-forth exchange, meaning one message from you and the system’s response to that message), so the system has to track the conversation as well as the current request:

1. Natural-language understanding: The system reads intent from a conversational request instead of matching keywords.
2. Context tracking: It keeps the dialogue history so each new turn is interpreted in light of what came before.
3. Query reformulation: Follow-ups and references such as “what about the cheaper option?” are rewritten into self-contained queries by resolving pronouns and implied context.
4. Retrieval: The resolved query drives retrieval (keyword, semantic, or agentic) against an index or live sources.
5. Response generation: Results are synthesized into a natural-language answer, often with citations, rather than returned as a bare list of links.
6. Iteration: The person can refine, drill down, or change direction, and the loop repeats with the accumulated context intact.

The defining characteristic is that the conversation itself is both the interface and the memory: a query rarely stands alone, because its meaning depends on the turns (which, again, means a single question from you and the system’s response to it) around it.

Conversational search vs. keyword search, chatbots, and agentic search

• Keyword search takes one stateless query and returns a ranked list of links; conversational search carries context across turns and returns a synthesized answer you can refine by replying.
• A chatbot is the conversational interface itself; conversational search is what happens when that interface is backed by real information retrieval rather than scripted or purely generative replies.
• Agentic search describes a system autonomously planning and iterating its retrieval steps; conversational search describes the multi-turn, human-facing dialogue. The two often work together, with an agentic system powering a conversational front end.

Where conversational search is used

• AI assistants and answer engines: Conversational search is the primary interface for modern AI search tools, where users ask and refine in natural language.
• Customer support: Dialogue-based help resolves follow-up questions without making the user restart from scratch.
• Enterprise knowledge search: Employees query internal documents in plain language and narrow the results turn by turn.
• Voice assistants: Spoken queries are inherently conversational and rely on carrying context between turns.
• Product discovery: Shoppers narrow options through back-and-forth, such as asking for alternatives under a set price.

Conversational search can have tradeoffs, including context window limits, hallucinations, and privacy concerns with chat-history retention. System design and platform controls can reduce some risks, but they do not eliminate model limitations such as hallucinations or context-window constraints. (Note that many of these risks are mitigated with Brave Search and the Brave Search API.)

Conversational search is most valuable when intent is exploratory or evolving: when a single query cannot capture what someone needs and the path to the answer is itself a conversation.

Related terms

Agentic search, semantic search, retrieval-augmented generation (RAG), AI answer engine, chatbot, natural language understanding, query reformulation, multi-turn dialogue, context window.