AI Instruction Quality Tracking Through UI Checkpoint Interactions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems lack effective methods to evaluate the quality of AI-generated content, making it difficult for service providers to improve services that rely heavily on generative AI tools.

Innovation Solution

A method that evaluates AI-generated content by receiving instructions from a generative AI model, identifying checkpoint interactions using an interaction index, determining user interactions with the interface, and generating a quality alert based on a computed metric to assess the quality of the instructions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If AI-generated instructions are used to perform tasks, then productivity is improved, but quality control deteriorates due to lack of evaluation mechanisms

Engineering Contradiction:
Improvetask completion efficiencyVSAvoidcontent quality
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where user interactions with AI-generated instructions are collected and used to compute quality metrics. The system tracks whether users successfully complete tasks using the generated instructions and feeds this information back to evaluate and improve future AI content quality, creating a closed-loop quality control system.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent replaces manual quality review processes with an automated evaluation system that uses machine learning models to assess AI-generated content. The system automatically computes quality metrics based on user interaction data, substituting human reviewers with an automated mechanical evaluation process that can scale efficiently.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual review of AI-generated content is implemented, then quality is improved, but productivity deteriorates due to increased time consumption

Engineering Contradiction:
Improvecontent qualityVSAvoidcontent delivery speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent replaces manual quality review processes with an automated evaluation system that uses machine learning models to assess AI-generated content. The system automatically computes quality metrics based on user interaction data, substituting human reviewers with an automated mechanical evaluation process that can scale efficiently.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The AI-generated content evaluates itself through the automated quality metric system. The content's quality is determined by actual user performance data rather than external manual review, allowing the system to self-assess and improve without requiring dedicated human quality assurance resources.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If comprehensive user interaction tracking is implemented, then measurement precision is improved, but device complexity deteriorates

Engineering Contradiction:
Improvequality evaluation accuracyVSAvoidtracking system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential interaction data needed for quality evaluation, such as whether users completed tasks successfully. Rather than tracking all possible user actions, the system focuses on extracting specific metrics directly related to instruction effectiveness, simplifying the tracking requirements while maintaining evaluation accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4703907A1User interface action tracking for quality evaluation of ai-generated content
Publication Date: 2026.03.04 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4703907A1 patent drawingFigure 1
  • EP4703907A1 patent drawingFigure 2A
  • EP4703907A1 patent drawingFigure 2B

AI summary

A method of AI content evaluation includes receiving, from a generative artificial intelligence (AI) model, a set of AI-generated instructions that identifies steps for performing a task within an application, and selecting checkpoint interactions from an interaction index that define a plurality of interactions with a user interface. Each of the checkpoint interactions satisfies a similarity metric with a corresponding step in the set of AI-generated instructions. The method further includes determining, based on detected user interactions with the user interface, a subset of the checkpoint interactions completed by a user within an observation period, and evaluating a metric that to compute a quality score that quantifies user success with respect to performing the task associated with the AI-generated instructions. The metric depending at least in part on the subset of the checkpoint interactions completed by the user within the observation period. In response to determining that the quality score satisfies lowquality criteria, a remedial action is performed.