Generates task-shaped result surfaces (book a flight, send a message, schedule a meeting, set a reminder) directly from queries, so search becomes an action launcher rather than just a document retriever.
Patent Overview
- Inventor
- Ramanathan V. Guha
- Assignee
- Google LLC
- Filed
- 2012-12-17
- Granted
- 2014-06-19 (published application)
- Application Number
- US 13/717,236
The Challenge
The Challenge
Many search queries express not an information need but an action intent: 'book a flight to Tokyo', 'remind me at 3pm', 'send a message to Mom'. The traditional SERP returns documents about these actions rather than executing them. The system needed to detect action intent and offer direct task execution surfaces.
- Document Results Miss Action Intent — When the user wants to book a flight, returning travel-information pages forces unnecessary navigation. The query implies an action; the system could let them execute directly.
- Tasks Need Structured Input Surfaces — Each task type (booking, messaging, scheduling) needs a specific input form. The system must offer structured input surfaces, not just text fields.
- Action Intent Detection Must Be Reliable — Wrong action detection routes the user to a task surface they did not want. Detection must clear precision thresholds.
- Tasks Span Multiple Services — Booking a flight involves an airline; messaging involves a messaging app; scheduling involves a calendar. Tasks need integration with external services.
- Privacy And Authentication Matter — Tasks involve user accounts and data. Authentication, consent, and data-handling controls must be first-class.
Innovation
How The System Works
The patent classifies queries for action intent, identifies the task type and required parameters, surfaces a structured task input form alongside or instead of search results, routes the completed task to the appropriate external service, and confirms execution back to the user.
- Classify Query Action Intent — Action-intent classifier outputs the probability that the query implies an action. Above-threshold cases trigger task pipeline; below-threshold cases route to standard search.
- Identify Task Type — Per action intent, identify the specific task type: flight booking, messaging, calendar scheduling, etc. Each task type has its own input schema.
- Extract Task Parameters From Query — Query parsing extracts available task parameters: destination, time, contact, message content. Missing parameters become input form fields.
- Surface Task Form — The task UI surfaces in the SERP or as a dedicated screen. Pre-filled with extracted parameters; missing fields prompt user input.
- Authenticate And Authorize — Task execution requires user authentication with the relevant service. Authorization flows handle account linking and consent.
- Execute Task Through Service Integration — The system routes the completed task to the external service via API. Booking, sending, scheduling all execute through service-specific integrations.
- Confirm And Surface Result — Task completion confirmation surfaces back to the user. Subsequent related searches can reference the completed task.
Search As Action Launcher
The patent's load-bearing idea is to extend search from document retrieval to task execution. Action-intent queries get direct task surfaces; the SERP becomes a multi-modal interface where information and action coexist.
Detect Action, Execute Directly
Many queries express action, not information. Detecting action intent and offering direct execution removes the navigation friction between query and outcome.
- Action Intent Classification — Learned classifier reads queries for action intent. Above-threshold cases enter the task pipeline.
- Task-Type Specific Forms — Each task type has its own input schema. Surfaces are tuned per task: booking forms for flights, message composers for messaging.
- Service Integration — Tasks execute through external service APIs. The system routes completed task data to airlines, messaging apps, calendar services.
Technical Foundation
Technical Foundation
The patent specifies the action-intent classifier, the task-type identifier, the parameter extractor, the task UI compositor, the authentication and authorization layer, the service integration framework, and the confirmation channel.
- Action Intent Classifier — Learned model classifies queries for action intent. Multi-label output supports queries with multiple action interpretations.
- Task Type Identifier — Per action intent, identifies specific task type. Each task type maps to a schema and a corresponding service integration.
- Parameter Extractor — Parses queries for available task parameters. Missing parameters become form fields the user fills.
- Task UI Compositor — Per task type, composes the input surface. Forms include pre-filled fields, validation rules, and execution affordances.
- Auth And Authorization Layer — Handles user authentication with external services. Authorization flows respect consent and data-sharing boundaries.
- Service Integration Framework — Per task type, integration framework routes data to external service APIs. Integrations are versioned and tested.
The Process
The Process
The task pipeline runs in the query path with optional UI handoff for user input. Action-intent queries can complete tasks in a few interactions.
- User Issues Query — Query arrives. Action-intent classifier runs alongside standard ranking.
- Classify And Identify Task — Action intent classified; task type identified. Below-threshold cases route to standard search.
- Extract Parameters — Parser extracts available task parameters from query. Output is partial task specification.
- Surface Task UI — Task UI renders with pre-filled fields. Missing fields prompt user input.
- User Completes Form — User fills missing fields and confirms. Validation runs before execution.
- Authenticate And Execute — Auth flow runs if needed. Service integration executes the task via external API.
- Confirm Back To User — Completion confirmation surfaces. Related searches can reference the completed task.
Quality Control
Quality Control
Wrong tasks executed cause real damage. The patent specifies safeguards.
- High Confidence Threshold — Action-intent detection threshold is conservative. False positives execute wrong tasks; false negatives miss opportunities. Conservatism favors the former safety.
- Pre-Execution Confirmation — Tasks confirm before execution. Users review filled parameters before the system routes to external services.
- Auth Boundary Enforcement — Authentication and authorization are mandatory. Tasks involving accounts or data cannot execute without explicit user authorization.
- Reversal Mechanisms — Where possible, tasks support reversal (cancel booking, recall message). Users can undo if execution went wrong.
- Sensitive Task Restrictions — Sensitive tasks (financial transactions, health bookings) face stricter confirmation and auth flows. Caution scales with task sensitivity.
Real-World Application
Task generation primitives underpin Google Assistant action execution, search-app task surfaces, and the agentic-task layers in modern AI assistant products. The patent's vision of search-as-action-launcher has become foundational to assistant design.
- Action-classified Trigger — Action-intent classifier gates task surfacing. Document queries skip the task pipeline.
- Service-integrated Execution Model — Tasks execute through external service APIs. Integration framework handles service-specific routing.
- Confirmation-gated Safety Layer — Tasks confirm before execution. Wrong tasks are avoided through user confirmation.
Why Assistants Inherit Task Primitives
Google Assistant, voice agents, and modern AI assistant products all build on action-intent detection plus task execution. The patent's primitives generalize from search-as-action-launcher to assistant-as-task-completer.
Why Action-Aware Content Earns Visibility
Pages serving action-intent users (booking sites, app-launching surfaces, transaction flows) win when the action-intent classifier routes to them. Pure information content cannot compete in action-intent slots.
<\/section>What This Means for SEO
What This Means for SEO
The patent detects action intent in queries and surfaces a direct task-execution surface (book, schedule, remind, message) instead of mere documents. SEO implication: for action-intent queries the SERP becomes a task launcher, so transactional and action-serving content competes for execution slots that informational content cannot reach.
- Action Intent Routes Past Information Content — The classifier separates action intent from information need. Queries that express doing rather than learning route to task surfaces. Pure information pages cannot win these slots, so identify which of your target queries are action-shaped and serve them with transactional pages.
- Structured Task Parameters Matter — The system identifies task type and required parameters to build the input form. Pages that expose the parameters of an action (dates, locations, options, prices) in structured form are better candidates for routing than pages that describe the action abstractly.
- Assistants Inherit These Primitives — Voice agents and AI assistants build on action-intent detection plus execution. Optimizing for direct task completion (clean booking flows, actionable structured data) positions you for assistant-driven traffic, not just classic search.
- Reduce Friction Between Query And Outcome — The patent's value is removing navigation steps between intent and action. Pages that let a user complete the action quickly, with minimal steps, align with the task surface's purpose and are favored over pages that bury the action behind clicks.
- Service Integrations Are The Endpoint — Completed tasks route to external services. Being a connected, reliable execution endpoint for a task type (a bookable service, a transactable product) is what earns you the routed action, beyond just describing the service.
- Match Content Type To Intent Type — Action-intent queries and information-intent queries deserve different pages. Splitting a topic into an informational guide and a transactional action page lets each compete in its own slot rather than one page underperforming on both.
- Confirm-And-Convert Beats Inform-And-Hope — Because the SERP can execute, users with action intent expect to finish there. Content engineered to convert the action (clear CTAs, complete parameter capture) captures action-intent traffic that informational depth alone would lose.