Voiceflow Query Rewriting Tackles Ambiguous Knowledge Base Searches.

Voiceflow is giving businesses more control over how AI agents interpret customer questions before consulting a knowledge base. Its new Query Rewriting feature can transform a user’s latest message, guided by custom instructions, into a search query more closely aligned with the company’s content.

The feature, announced on LinkedIn, targets a persistent problem in conversational search: customers rarely phrase requests using the same terminology as product documentation, support articles or internal teams. Voiceflow cites the “vocabulary problem,” the idea that two people shown the same object will choose the same descriptive word less than 20% of the time.

That mismatch matters because even a well-maintained knowledge base can produce weak results when an agent searches it using the customer’s original wording. A user might describe an account restriction as being “locked out,” for example, while the documentation refers to “authentication failure” or “access recovery.” Query Rewriting introduces an interpretation step before retrieval rather than requiring businesses to anticipate every possible phrase in their content.

When the setting is enabled, Voiceflow’s model rewrites the user’s most recent message before running a knowledge base search. The rewrite is based on instructions supplied by the agent builder, allowing the business to define how customer language should be interpreted in its specific context.

The process can be summarized as:

1. A customer sends a conversational, incomplete or ambiguous message.
2. The model rewrites that message according to the builder’s instructions.
3. Voiceflow searches the knowledge base using the rewritten query.
4. The agent uses the retrieved material to formulate its response.

The original announcement does not provide a library of instruction templates or disclose which underlying model performs the rewrite. It does establish that the setting is optional, and that builders control the instructions used to interpret intent.

This is particularly relevant for follow-up questions. A message such as “What about the annual one?” may be understandable in the context of an ongoing conversation but poorly suited to a standalone knowledge search. Rewriting can potentially turn that fragment into a query that includes the relevant product or policy context, provided the instructions and conversation state support that interpretation.

For small businesses building support or sales agents, the practical benefit is reduced dependence on exact keyword alignment. Query Rewriting could help agents handle synonyms, customer shorthand, domain-specific language and vague follow-ups without requiring teams to duplicate articles under multiple phrasings.

The trade-off is that rewriting adds another model-generated step to the retrieval pipeline. An overly aggressive instruction could change the user’s meaning rather than clarify it. Builders should test direct questions, ambiguous requests and multi-turn follow-ups, then compare the rewritten search behavior with the previous results.

Voiceflow has also added knowledge base logs to evaluations, according to its official changelog. When enabled, those logs let an evaluation inspect the searches an agent performed, which should make it easier to determine whether Query Rewriting is finding the intended source material rather than merely producing a plausible final answer.

The release continues Voiceflow’s broader push to give teams finer control over production agents. It follows the company’s Markdown editor and reference resources for agent instructions and its addition of operational guardrails for AI agent operations. Together, these changes move more of an agent’s behavior into configurable, inspectable workflows rather than leaving interpretation entirely to a general-purpose model.

What is Voiceflow Query Rewriting?

Voiceflow Query Rewriting is an optional knowledge retrieval feature that rewrites a user’s latest message before searching. The rewrite follows instructions supplied by the builder, helping the agent translate conversational or ambiguous language into a query better matched to the business’s knowledge base.

Who is Query Rewriting intended for?

It is intended for teams whose customers use different words, shorthand or follow-up phrases than those found in their support and product content. It may be especially useful for agents working with specialized terminology or handling multi-turn conversations.

Which Voiceflow plan includes Query Rewriting?

Voiceflow’s announcement does not specify pricing, plan requirements or rollout restrictions for Query Rewriting. Teams should check their workspace and the current Voiceflow documentation for availability.