Choosing the Right Tool

A Web Search Agent is the right tool when the answer is scattered across pages nobody has listed yet. It is the wrong tool for a page you already have, a fact one search returns, or a site that already has an Extract Template. Picking wrong costs minutes and dollars where seconds and cents would do.

Decision tree

Do you already know the URL?

One page: use Extract. It reads the page deterministically in seconds.A site with an Extract Template (Amazon, Google Maps, TikTok and hundreds more): run the template. It returns the site’s fields already structured.Many pages on one site: use Map to list them, then Extract each, or Crawl.

No URL. Is it one closed fact?

“Who is the CEO of Acme”, “when did X launch”, “what is Y’s HQ city”: use Search with include_answer. One call, one sentence, a source.

No URL, and you want a table, a filled record, or a brief

This is a Web Search Agent task. Pick the use case:

Rows you already have that need more columns: Enrichment
A list of entities that match criteria: Dataset Building
A sourced answer to an open question: Research

Side by side

	Search	Extract	Extract Template	Web Search Agent
You have	A question	A URL	A URL or ID on a supported site	A task, and optionally rows
You get	Ranked results, optional one-line answer	One page as markdown, JSON or HTML	The site’s fields, structured	A table, a completed record, or a cited brief
Latency	Seconds	Seconds	Seconds	Minutes: 1 to 3 at `medium`, 5 to 15 at `high`
Price	Per request	Per page	Per run	Per run by effort: $0.10 to $2.00
Best for	Lookups, monitoring, RAG grounding	Known pages	Known sites at scale	Discovery, multi-source synthesis, verification

Real asks, routed

Ask	Use	Why
”Extract price and stock from this product URL”	Extract, or the site’s Extract Template	Known page. An agent would search for what you already have
”Find the HQ city of these 50 companies”	Search with `include_answer`, one call per company	One closed fact each. A `low` agent run costs more and returns less
”List every dentist in Austin with rating and phone”	Google Maps Extract Template	The site is known and templated. An agent took 20 minutes for 10 rows on the same ask; the template returns 20 in seconds
”Which of these 200 vendors publish a pricing page, and what is the entry price?”	Web Search Agent, enrichment	A judgment per row from an unknown page on each vendor’s site
”Find Series A+ AI infrastructure startups in Europe funded since 2024”	Web Search Agent, dataset building	Nobody has published this list. It has to be assembled from many sources
”Due-diligence brief on a private company”	Web Search Agent, research	Synthesis across filings, press and the company’s own site
”Get every product in this manufacturer’s catalog”	Map + Extract, or an agent scoped to one section at a time	A whole catalog in one run truncates. See Running at Scale
”Pull posts from this Instagram profile”	Extract Template if one exists for the site	Login-walled social sites are not readable by an agent. See Controlling Sources

When in doubt, start with Search. If the first results contain the answer, you are done. If they only point at where the answer might be, that is what an agent is for.

Next Steps

Enrichment

Complete the rows you already have

Dataset Building

Build a list from criteria

Research

Get a cited brief

Getting Started

Search Tools

Agent

Extraction Tools

Extract Template

Extract Jobs

Guides

SDKs

Admin

Proxy

Choosing the Right Tool

Decision tree

Side by side

Real asks, routed

Next Steps

Enrichment

Dataset Building

Research

​Decision tree

​Side by side

​Real asks, routed

​Next Steps

Enrichment

Dataset Building

Research

Decision tree

Side by side

Real asks, routed

Next Steps