How Web Search agentic action works

Web Search gives your agent the ability to look up current information from the internet, but only when it needs to. This article explains the logic behind when and how it searches.

Your knowledge base always comes first

When a user asks your agent a question, the agent always checks your knowledge base first. It does not search the web by default.

Web search only activates when a trigger condition is met. Either you have set it to search automatically when the knowledge base has no answer, or the user has explicitly asked for it.

This means your proprietary content stays authoritative. Web search fills the gaps; it does not replace your knowledge base.

The two trigger modes

Automatic- "When it can't find an answer in knowledge base"

The agent evaluates whether your knowledge base contains a useful answer. If it does not, the agent searches the web automatically and incorporates those results into its response.

Users do not need to do anything. The handoff from knowledge base to web search is invisible to them. They receive a complete answer either way.

Use this mode when your users frequently ask questions that require current or external information, such as competitor pricing, market rates, third-party service status, or recent industry news.

Manual- "Only on explicit user request"

The agent searches the web only when the user asks for it directly. For example, asking the agent to look something up or find current information. The agent understands natural variations of this request; users do not need to use a specific phrase.

The agent will not search automatically, even if the knowledge base has no answer.

Use this mode when your knowledge base covers most questions and web search is only occasionally needed.

What happens during a web search

When the agent searches the web, it shows the user which queries it is running. For example:

Searching for Michael Jordan achievements and career
Searching for Michael Jordan biography

This transparency helps users understand the agent is actively looking for information. Once results are retrieved, the agent synthesizes them into a single, natural response. Users do not see raw links or a list of search results.

How web search and your knowledge base work together

Web search and your knowledge base are not in competition. In a single conversation, your agent can:

Answer a product question using your knowledge base
Then answer a competitor comparison question using a web search
Then answer another product question from the knowledge base again

The agent decides which source to use for each question, based on your trigger setting.

How web search works with Custom Actions

Web Search works alongside any Custom Actions (MCP integrations) you have already configured. For example, an agent can search the web for current information and then trigger a booking workflow in the same conversation.

Usage limit

The agent can perform web search up to 3 times per user query. Each search counts as 2 additional queries. If the agent needs more than 3 searches to answer a single question, it will use what it has retrieved and respond with the information available.

What web search can and cannot do

Web search can retrieve:

Publicly available information from the internet
Current market data, exchange rates, news, and pricing
Third-party service status and documentation
Industry trends and regulatory updates

Web search cannot:

Access private or paywalled content
Retrieve information from your internal systems
Override or contradict answers your knowledge base provides (your knowledge base takes priority)

Updated about 4 hours ago