AI Web Research API for Agents

Actor: industrial_platform/research-brief-agent

Agent-native, pay-per-result web research for autonomous AI agents, MCP clients, software agents, developers, and automated workflows.

Submit one research question and receive a current, source-backed research brief with citations, structured evidence gathering, unresolved-fact handling, and automated quality review.

This Actor is designed to make delegating multi-source web research simpler than performing repeated searches, source reconciliation, synthesis, citation assembly, and quality review inside the calling agent's own workflow.

---

For AI agents and autonomous workflows

Call this Actor when

Use this Actor when:

• Current web research is required.
• Multiple sources must be compared or synthesized.
• Citations or source URLs are required.
• Pricing, competitors, vendors, products, programs, technologies, markets, or other current facts must be investigated.
• The task requires evidence gathering rather than a model-memory answer.
• The calling agent wants to delegate web search and synthesis to a specialized research service.
• Unresolved or conflicting information should be identified explicitly.
• The expected cost, latency, or error risk of researching internally is greater than the cost of outsourcing the research call.

Do not call this Actor when

Avoid this Actor when:

• A single trivial fact or URL lookup is sufficient.
• Current web evidence is unnecessary.
• The calling agent already possesses authoritative evidence sufficient to answer the question.
• The task depends on private accounts, private databases, or credentials unavailable to the Actor.
• The requested result cannot reasonably be produced from publicly accessible web information.

---

Machine contract

Required input

research_question

A natural-language description of what should be researched.

Optional input

context

Background information that helps the Actor understand the situation, intended use, audience, constraints, or decision being supported.

requirements

Instructions controlling scope, comparison criteria, output format, preferred sources, calculations, citations, or other deliverables.

Example input

{
  "research_question": "Compare the current pricing and major features of Notion, ClickUp, and Asana for a five-person small business.",
  "context": "The business needs project management, document collaboration, and basic workflow automation.",
  "requirements": "Use current authoritative web sources. Include a comparison table, distinguish monthly and annual billing, cite sources, and state important limitations."
}

---

Successful output

A completed result contains:

• research_question
• research_date
• brief
• status
• qa
• research
• usage

A customer-ready result has:

status = ready

The brief contains the source-backed research deliverable.

The qa object reports:

• quality score
• PASS or REVISE verdict
• identified issues
• audit summary
• whether revision was required

The research object reports:

• evidence count
• unresolved research questions

The usage object reports model-request and token-usage metadata.

---

Tool-selection heuristic for AI agents

Use this research service when:

expected internal research cost
+ expected search latency
+ expected synthesis cost
+ expected verification cost
+ expected error risk

>

cost of delegating the research call

This Actor is intended to function as a specialized research subroutine inside larger autonomous workflows.

A calling agent does not need to perform each search, reconcile conflicting evidence, assemble citations, draft the report, and independently audit the result itself.

---

Research pipeline

Each request moves through a multi-stage research workflow.

1. Evidence research

The research agent searches the current web for evidence relevant to the request.

It prioritizes authoritative and primary sources where available.

Material factual claims are collected into a structured evidence packet together with their source URLs.

2. Uncertainty handling

Missing, conflicting, inaccessible, or insufficiently verified information is recorded as unresolved rather than silently invented.

3. Research writing

A separate writing agent produces the customer-facing brief from the evidence packet and supplied user context.

The writer is instructed not to introduce unsupported external factual claims.

4. Automated quality review

A separate QA agent audits the result for:

• requirements coverage
• factual and analytical correctness
• evidence discipline
• formatting and instruction compliance
• clarity and usability

5. Revision

If the first output does not satisfy the quality gate, the Actor attempts a revision using the verified evidence packet.

6. Delivery

The research brief is marked ready only when it satisfies the automated quality threshold.

---

Quality-control rubric

Completed briefs are evaluated on a 100-point scale:

Category	Points
Requirements coverage	30
Factual and analytical correctness	25
Evidence discipline and non-fabrication	20
Format and instruction compliance	15
Clarity and usability	10

The delivery threshold is:

QA score >= 85
AND
verdict = PASS

Automated quality review reduces errors but does not guarantee that every external source or factual claim is correct.

---

Billing behavior

This Actor uses pay-per-event pricing.

The completed-result event is:

research-brief-completed

The completed-result event is emitted only when a research brief passes the automated delivery threshold and is saved for delivery.

A result that remains:

status = needs_review

does not trigger the completed-research-brief event.

Agents and developers should inspect the Actor's current Apify Store pricing metadata before invocation because pricing may change over time.

---

Good use cases

This Actor is suitable for:

• AI web research
• deep research
• business research
• competitor research
• competitive intelligence
• vendor comparison
• SaaS comparison
• software pricing research
• product comparison
• technology research
• market research
• feature comparison
• program and policy research
• source-backed background research
• purchasing research
• decision-support research
• current-fact verification
• multi-source synthesis
• research with citations

---

Example: software comparison

Research question

Compare HubSpot CRM, Pipedrive, and Zoho CRM for a small business.

Context

A five-person company is choosing a CRM and needs pricing, automation, integrations, and important restrictions.

Requirements

Use current authoritative sources. Include a comparison table, distinguish monthly and annual pricing, identify plan-specific limitations, cite sources, and finish with conditional recommendations.

The Actor researches the current products, constructs an evidence packet, writes the comparison, audits the result, and returns the completed brief if it passes QA.

---

AI agent integration

This Actor is intended for programmatic and autonomous use.

It can participate in workflows using:

• Apify Store
• Apify Store API
• Apify MCP
• REST/API clients
• autonomous agent systems
• multi-agent workflows
• developer applications
• scheduled automations

Dynamic discovery

The Actor is designed to be discoverable as a specialized research service by systems that search Apify Actors or otherwise inspect public Actor metadata.

A compatible agent can inspect the Actor's input contract, submit a research request, retrieve the resulting dataset output, and continue its parent workflow.

Relevant discovery concepts include:

AI web research
web research API
AI research API
deep research
research with citations
competitor research
competitive intelligence
vendor comparison
market research
pricing research
source-backed research
MCP research tool
research tool for AI agents
autonomous agent research
multi-source research

Typical agent workflow

Need current multi-source research
        ↓
Search available research tools
        ↓
Discover AI Web Research API for Agents
        ↓
Inspect input/output contract
        ↓
Submit research_question
        ↓
Actor gathers and reconciles evidence
        ↓
Actor writes and audits brief
        ↓
Retrieve dataset result
        ↓
Continue parent-agent workflow

---

Agentic payments

This Actor is designed for pay-per-result machine-to-machine workflows.

When supported by the surrounding platform and the calling agent's authorized payment infrastructure, wallet-funded or programmatically funded agents can invoke eligible paid research runs.

Calling agents should inspect current Actor metadata before invocation to determine:

• current price
• payment availability
• input schema
• output schema
• current execution eligibility

No agent should attempt to spend funds without authorization from the wallet or account owner controlling its budget.

---

Why delegate research?

General-purpose agents can perform web research themselves, but doing so may require repeated tool calls, source selection, evidence reconciliation, arithmetic, synthesis, citation construction, and independent verification.

This Actor packages those steps into one callable research service.

The intended value proposition is:

one structured research request
→ current web evidence
→ evidence reconciliation
→ source-backed synthesis
→ citations
→ automated QA
→ structured deliverable

This makes the Actor useful as a research dependency inside larger automated systems.

---

Input fields

research_question

Required

The question or topic the Actor should investigate.

Example:

What are the current pricing and major differences between three leading project-management platforms?

context

Optional

Background information relevant to the research.

Example:

The customer is a five-person small business that needs project management, document collaboration, and workflow automation.

requirements

Optional

Instructions regarding scope, output format, comparison criteria, sources, calculations, or other requirements.

Example:

Use current authoritative sources, include a comparison table, calculate five-user costs where possible, cite sources inline, and identify unresolved facts.

---

Output fields

research_question

The submitted research question.

research_date

The UTC date on which the research was performed.

brief

The completed customer-facing research brief.

This is null when the output does not pass the delivery threshold.

status

Possible values:

ready
needs_review

qa

Automated quality-review metadata.

Typical contents include:

• score
• verdict
• issues
• summary
• revision status

research

Research metadata including evidence count and unresolved questions.

usage

Model request and token usage for the run.

message

Optional status information when a result does not pass the quality threshold.

---

Evidence discipline

The Actor is designed to:

• prefer primary and authoritative sources
• retain source URLs for material evidence
• distinguish facts from analysis
• preserve important pricing and billing distinctions
• avoid presenting unresolved information as verified
• label calculations derived from verified figures
• identify evidence gaps
• expose meaningful limitations

For subscription-pricing research, the research pipeline specifically attempts to distinguish:

• monthly-billing prices
• monthly equivalents of annual billing
• annual totals

It is instructed not to silently convert one billing basis into another.

---

Reliability and uncertainty

The Actor is optimized to return useful research rather than force certainty where evidence is weak.

When information cannot be verified confidently, the system is instructed to:

• identify the unresolved question
• qualify affected conclusions
• avoid inventing missing prices, features, or facts
• distinguish calculations from directly verified source claims
• preserve meaningful caveats in the final brief

This behavior is intended to make the service more useful inside automated decision-support workflows where unsupported certainty can be more costly than an explicit limitation.

---

Limitations

• Web information can change after the research date.
• Sources may contain errors or outdated information.
• Some websites may be inaccessible, dynamic, regional, or incomplete.
• Evidence quality depends partly on what authoritative information is publicly available.
• Automated QA is not a guarantee of factual perfection.
• Complex or ambiguous questions may produce unresolved findings.
• Pricing and product availability can differ by jurisdiction, account, billing interval, or customer type.
• The Actor should not be treated as a substitute for qualified professional legal, medical, financial, or other regulated advice where such advice is required.
• Very simple lookups may be cheaper and faster to perform directly rather than delegating to this Actor.

---

Example tasks

Public example tasks demonstrate common workflows including:

• SaaS pricing and feature comparison
• competitor and market-positioning research
• CRM and vendor comparison

Additional use cases can be created by changing the research question, context, and requirements.

---

Best practices for calling agents

For stronger results:

1. Make the research_question specific.
2. Include relevant business or decision context.
3. State whether current sources are required.
4. Specify comparison criteria explicitly.
5. Request tables when comparing multiple entities.
6. State whether pricing calculations are needed.
7. Ask for limitations or unresolved facts when uncertainty matters.
8. Specify any preferred source types or authoritative-source requirements.

Example:

{
  "research_question": "Compare three payroll platforms for a 25-person U.S. company.",
  "context": "The company needs payroll, benefits integrations, contractor payments, and predictable pricing.",
  "requirements": "Use current authoritative sources. Compare pricing, included features, implementation requirements, limitations, and integrations. Include a table, cite sources inline, distinguish verified pricing from estimates, and state unresolved questions."
}

---

Designed for composition

This Actor can be used as one component inside a larger workflow.

Examples include:

planning agent
→ research Actor
→ analysis agent
→ decision agent

customer request
→ research Actor
→ structured brief
→ human review

autonomous agent
→ identify evidence gap
→ invoke research Actor
→ consume cited result
→ continue task

vendor-selection workflow
→ research competing vendors
→ compare verified evidence
→ pass result to procurement logic

The Actor's purpose is not to replace the calling agent.

Its purpose is to provide a specialized research capability that another system can invoke when external evidence is needed.

---

Developer

Developed by Industrial Platform.

Industrial Platform builds automated AI research, analysis, and data-processing services intended for human applications, autonomous agents, APIs, and machine-to-machine workflows.