AI Content Attribution Using Segment Matching and Selective Modification

Solution Overview

Problem

Generative models (GMs) often fail to ground responsive content in the underlying data on which they were trained, leading to unnecessary computational resource waste and latency due to large search spaces when matching segments of responsive content with user-provided content or training data.

Innovation Solution

Implement a system that normalizes and segments user-provided content and responsive content, comparing segments to determine matches, conserving resources by modifying only unmatched segments, and applying rules based on training dataset sources for attribution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the system searches through the entire training dataset to match segments of responsive content, then attribution accuracy is improved, but computational resources are wasted and latency increases

Engineering Contradiction:

Improveattribution accuracyVSAvoidcomputational resources

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system segments the training dataset into user-provided content and other content, then segments the responsive content into segments. It compares segments of responsive content against segments of user-provided content first, and only against other training data if no match is found. This segmentation approach reduces the search space while maintaining attribution accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary comparison against user-provided content before searching the broader training dataset. This preliminary action filters out matches from user content first, reducing the need to search through the entire training dataset and conserving computational resources.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If the system searches through the entire training dataset to match segments of responsive content, then attribution accuracy is improved, but processing time increases

Engineering Contradiction:

Improveattribution accuracyVSAvoidlatency

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system segments the training dataset into user-provided content and other content, then segments the responsive content into segments. It compares segments of responsive content against segments of user-provided content first, and only against other training data if no match is found. This segmentation approach reduces the search space while maintaining attribution accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary comparison against user-provided content before searching the broader training dataset. This preliminary action filters out matches from user content first, reducing the need to search through the entire training dataset and conserving computational resources.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If the system modifies all segments of responsive content, then attribution completeness is improved, but processing complexity increases

Engineering Contradiction:

Improveattribution completenessVSAvoidprocessing complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system applies local quality by modifying only those segments of responsive content that match segments of other training data (but not user-provided content). Segments that match user-provided content are not modified. This selective modification approach reduces processing complexity while maintaining attribution completeness for relevant segments.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

Instead of modifying all segments and then filtering, the system inverts the approach by identifying which segments should NOT be modified (those matching user content) and applying modifications only to the remaining segments. This inversion reduces processing complexity while maintaining completeness.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS20260064984A1Dynamic attribution and/or modification of ai-generated content based on contextual data

Publication Date: 2026.03.05 GOOGLE LLC

AI summary

Some implementations described herein relate to determining whether to modify segment(s) of responsive content that is generated using a generative model (GM) and based on processing user provided content (e.g., documents associated with a user) and a natural language prompt. Processor(s) of a system can: receive user input associated with a client device that includes the user provided content and the natural language prompt, generate the responsive content using the GM and based on processing the user provided content and the natural language prompt, and determine whether to modify the segment(s) of the responsive content based on whether the segment(s) of the responsive content match the user provided content. The processor(s) can refrain from modifying the segment(s) of the responsive content if there is a match to the user provided content, and can cause the responsive content to be rendered at the client device.