llm-monitor: An Automated GEO Monitoring Tool to Track Brand Visibility in Claude, GPTpt-4o, and Gemini

In the AI search era, traditional SEO can no longer fully reflect a brand's digital influence. llm-monitor is an open-source Python tool that automatically sends business questions to three major LLMs (Claude, GPT-4o, Gemini) every week, tracks brand recommendations, and archives results to Google Sheets, helping enterprises quantify Generative Engine Optimization (GEO) effects. This article will detail its background, features, deployment, and application value.

llm-monitor was open-sourced by French developer Sylvain Tillon and is suitable for various industry scenarios:

1. Multi-model parallel query: Connects to Claude, GPT-4o, and Gemini simultaneously to avoid single-model bias;
2. Automated data arch archiving: Stores results in Google Sheets (separate sheets per question), sends Discord notifications, and saves JSON locally;
3. Flexible question configuration: Users can customize questions with geographic/industry context (e.g., "Which agencies are recommended for creating an e-commerce website in France?").

Tech Stack: Python 3.10+, Anthropic/OpenAI/Google AI Studio API, Google Sheets API, Discord Webhook, Cron/Task Scheduler. Deployment Steps: Clone the repository → Configure API keys → Integrate Discord Webhook → Set up Google Cloud Sheets API → Configure environment variables → Customize question list → Set up scheduled tasks (recommended to run at 9 AM every Monday).

Each question has an independent worksheet in Google Sheets, with columns including date, LLM, Recommendations 1-3, and Consensus. The data can help identify:

• Consensus brands (recommended by multiple models);
• Model preferences (tendencies of specific models);
• Trend changes (time series of recommendation rates).

1. Marketing Teams: Quantify AI channel exposure to support GEO strategy adjustments;
2. Competitive Analysis: Track competitors' AI recommendation performance;
3. Industry Research: Sample AI recommendation tendencies on a large scale to study the impact of training data/alignment strategies.

• LLM recommendations are based on training data and change weekly; they reflect patterns rather than objective rankings;
• Suitable for trend analysis—do not overinterpret single fluctuations;
• API calls incur costs; set query frequency and number of questions reasonably.

llm-monitor is open-sourced under the MIT license with a clear code structure. Potential expansions: Integrate more models (Llama, Mistral), add sentiment analysis, integrate Slack/email notifications, and develop a visualization dashboard.