The Google Duplex automated-calling patent. Places autonomous phone calls on the user's behalf with conversational voice and status updates — restaurant reservations, hair appointments, business hours, all handled by Assistant calling the business directly.
Patent Overview
- Inventor
- Yossi Matias, Yaniv Leviathan, others
- Assignee
- Google LLC
- Filed
- 2018
- Granted
- 2020-11-03
The Challenge
The Challenge
Many tasks (restaurant reservations, appointment bookings, business hours inquiry) still require phone calls. The user must call, navigate menus, talk to staff. Automated calling on the user's behalf removes the friction — but requires conversational voice synthesis, real-time dialog handling, and status reporting back to the user.
- Phone-Required Tasks Persist — Many businesses still require phone contact. Online booking isn't universal.
- Conversational Voice Is Required — Robotic-sounding calls fail. Conversational, natural-cadence voice is required for business uptake.
- Real-Time Dialog Handling Is Hard — Phone conversations branch unpredictably. Real-time NLU and response generation required.
- User Status Must Be Updated — User isn't on the call. Status updates (calling, in-progress, complete, failed) keep them informed.
- Disclosure Required — Ethical requirement: callee must know they're talking to an automated system. Disclosure built into the call flow.
Innovation
How The System Works
The system receives user task requests, places autonomous calls with conversational voice synthesis, runs real-time NLU on the callee's responses, generates appropriate replies, discloses automation status, and reports task outcome and status updates to the user.
- Receive User Task — User requests task (e.g., 'book me a haircut appointment at 4pm Saturday').
- Place Outbound Call — System places call to business. Disclosure of automation status included.
- Conversational Voice Synthesis — Per turn, voice synthesis produces natural-cadence speech.
- Real-Time NLU On Callee — Per callee utterance, NLU processes in real time.
- Generate Response — Per callee utterance, response generated to advance task.
- Report Status To User — Per call phase, status update sent to user (calling, in-progress, complete, failed).
- Confirm Outcome — Final task outcome (confirmed, declined, requires user attention) reported to user.
Conversational Automation At Phone Scale
The patent's load-bearing idea is that automated calls require conversational handling — natural voice, real-time NLU, response generation, status reporting. The architecture combines speech synthesis, dialog, and reporting into a complete automated-caller stack.
Real-Time Dialog Plus Disclosure
Per call, real-time dialog handling enables task completion. Per call, automation disclosure preserves ethical and legal status. The two requirements combine into the architectural design.
- Conversational Voice Synthesis — Per turn, natural-cadence speech synthesized.
- Real-Time NLU — Per callee utterance, NLU processes in real time.
- User-Status Reporting — Per call phase, status update sent to user.
Technical Foundation
Technical Foundation
The patent specifies the task receiver, call placer, voice synthesizer, NLU processor, response generator, status reporter, and outcome confirmer.
- Task Receiver — Receives user task requests with constraints.
- Call Placer — Places outbound calls with disclosure.
- Voice Synthesizer — Per turn, natural-cadence speech synthesis.
- NLU Processor — Per callee utterance, real-time NLU.
- Response Generator — Per turn, task-advancing response generated.
- Status Reporter — Per call phase, status updates user.
The Process
The Process
Per user task, the Duplex pipeline places call and reports outcomes.
- Receive Task — User task captured.
- Place Call With Disclosure — Call placed; automation disclosed.
- Begin Conversation — Voice synthesis begins; greeting.
- NLU On Callee — Per callee response, NLU runs in real time.
- Generate Response — Task-advancing response generated.
- Update User — Status updates flow to user.
- Confirm Outcome — Final outcome reported.
Quality Control
Quality Control
Failed calls or wrong outcomes damage user trust. The patent specifies safeguards.
- Disclosure Enforcement — Per call, automation disclosure enforced. Compliance with state laws.
- Voice-Quality Validation — Per turn, voice synthesis quality monitored.
- NLU Confidence Threshold — Per callee utterance, NLU confidence required. Low-confidence triggers user handoff.
- User-Handoff Trigger — Complex situations trigger handoff to user. Automation doesn't force-complete uncertain tasks.
- Continuous Recalibration — Voice, NLU, response generation models recalibrate against fresh data.
Real-World Application
Google Duplex is one of the highest-profile Assistant automation patents. The conversational-call architecture has expanded to business-hour queries, appointment confirmations, and Google's 'Hold for Me' feature.
- Conversational Voice Quality — Natural-cadence voice synthesis produces phone-uptake-worthy speech.
- Real-time NLU Dialog Handling — Per callee utterance, NLU processes in real time.
- Status-reported User Experience — Per call phase, status updates flow to user.
Why Phone-Listed Businesses Should Maintain Online Booking
Duplex calls businesses that haven't shifted to online booking. Businesses with online booking systems remove Duplex friction entirely, serving Assistant users directly through structured booking interfaces.
Why Business-Hour Accuracy Matters For Assistant Era
Duplex and Assistant query business hours, availability, and booking details. Accurate, up-to-date business information across Google Business Profile and structured-data markup directly serves the Assistant query layer.
<\/section>What This Means for SEO
What This Means for SEO
This Google Duplex patent places autonomous, conversational phone calls on a user's behalf for tasks like reservations and hours inquiries. SEO implication: accurate, up-to-date business information and online booking serve the Assistant query layer directly and remove call friction.
- Keep Business Hours And Details Accurate — Duplex and Assistant query business hours, availability, and booking details. Accurate, current information across Google Business Profile and structured-data markup directly serves the Assistant layer that initiates these tasks.
- Offer Online Booking To Remove Friction — Duplex calls businesses that have not moved to online booking. Providing a structured online booking interface serves Assistant users directly and removes the friction Duplex exists to handle.
- Structured Data Feeds The Assistant Layer — The system relies on knowing a business's bookable services, hours, and availability. Exposing this through structured data makes you machine-readable to Assistant-era query handling.
- Stale Information Breaks The Task — An automated call against outdated hours or availability produces a failed or wrong outcome. Information accuracy is not just a UX nicety here, it determines whether the automated task succeeds at all.
- The Assistant Era Reads Your Profile Programmatically — Hold for Me, appointment confirmations, and business-hour queries all consume the same business data. Maintaining one accurate, complete profile serves a growing set of automation surfaces.
- Disclosure And Handoff Are Built In — Calls disclose automation and hand off complex situations to the user. Businesses that make common tasks self-serve online reduce the need for these calls and the chance of a handoff or failure.
- Booking Interfaces Beat Phone-Only — Phone-required tasks persist precisely where online booking is absent. Closing that gap with structured booking is the way to serve Assistant users on your own terms rather than via an automated call.