Marketing
Fact-Check AI-Generated Content
This n8n workflow fact-checks AI-generated content. It uses a Code node to split text into sentences, then uses an Ollama Chat Model (bespoke-minicheck) via Langchain to verify each sentence against provided facts. Incorrect statements are filtered, and a separate Ollama Model…
Built · ~4 hours saved per week
Fact-Check AI-Generated Content
Workflow Overview This workflow automates the process of fact-checking an article or piece of text against a provided source document. It systematically breaks down the article into individual claims (sentences), verifies each claim against the source material using an AI model, filters out the incorrect claims, and then uses a second AI model to generate a structured, human-readable summary of the errors found.
Automation & Value This workflow provides immense value by automating a tedious and error-prone manual task. Instead of a human editor carefully comparing two documents line by line, this automation does it in seconds. It ensures consistency, reduces human bias, and provides a clear, actionable report for content editors to quickly identify and fix factual inaccuracies. This significantly accelerates the content review and publishing lifecycle.
Technologies Used - n8n: The core automation platform coordinating the entire process. - Ollama: A service for running local Large Language Models (LLMs), enabling private and cost-effective AI processing. - JavaScript: Used for initial text processing and data preparation. - AI Models: - bespoke-minicheck: A specialized, lightweight AI model designed specifically for the binary task of fact-checking a claim against a document. - qwen2.5: A general-purpose AI model used for understanding context and generating the final formatted summary report.
AI Integration This workflow employs a sophisticated two-stage AI process for maximum efficiency and accuracy:
1. Sentence-Level Fact-Checking: A highly specialized and efficient AI model (bespoke-minicheck) is used to perform a focused, high-speed analysis on each individual sentence. It provides a simple "Yes" or "No" output on whether the sentence is factually supported by the source document. This分而治之 (divide and conquer) approach is highly scalable. 2. Intelligent Summarization: After isolating the incorrect statements, a more general AI model (qwen2.5) is prompted with a detailed set of instructions. It takes the list of errors and synthesizes them into a structured, professional report.
Tags: SEO, Content Management, Data Transformation, AI Automation, Multi-Agent Chain, Subflow
Integrations: Code, Ollama