Integrations

LLMMonitor connects to four major AI platforms — ChatGPT, Gemini, Claude, and Perplexity — using a combination of browser automation and official APIs to capture real user-facing responses.

Integration Overview

Platform	Method	Plan Required
ChatGPT	Selenium + API	Free
Gemini	Selenium + API	Free
Claude	Selenium	Lite+
Perplexity	API	Lite+

ChatGPT

LLMMonitor uses two methods to interact with ChatGPT:

Method 1: Web Scraping (Selenium)

Uses undetected-chromedriver to interact with chat.openai.com directly. This captures the exact responses users see, including web search results, source pills, and inline citations.

Auth: Cookie-based session persistence. LLMMonitor maintains authenticated sessions to avoid repeated logins.
Email verification: Auto-handles IMAP email verification code polling when re-authentication is needed.
Citations: Extracted from DOM source pills, <a> tags, and web search reference carousels.
Search queries: Captured from the web search step — shows what ChatGPT searched for before answering.

Method 2: API Mode

Uses the OpenAI Responses API with web_search tool enabled. Faster and more reliable than Selenium, with structured citation metadata.

Model: GPT-4o (configurable)
Citations: Extracted from web_search_call.action.queries and inline URL references
Rate limits: Subject to OpenAI API rate limits

Gemini

Dual approach for Gemini, similar to ChatGPT:

Web Scraping

Selenium-based interaction with gemini.google.com. Cookie-based session persistence.

API Mode

Uses the Gemini API with Google Search grounding enabled. This gives Gemini access to live web search results, producing responses that closely match what users see on the web interface.

Model: gemini-2.5-flash
Geo-targeting: Supported — you can specify which country to run searches from
Sources: Structured source metadata from API grounding results

Claude

Selenium-based tracking of Claude's web interface at claude.ai.

Auth: Cookie-based session management
Citations: Inline URLs and reference links parsed from the response text
Search queries: Not captured — Claude's web interface doesn't expose internal search queries in the same way ChatGPT and Gemini do
Availability: Lite plan and above

Claude limitations Claude's web interface has the most variance between sessions. Responses may differ significantly between runs due to Claude's more conservative web search behavior. Expect slightly lower citation counts compared to ChatGPT and Perplexity.

Perplexity

API-only integration using Perplexity's Sonar model.

Model: Sonar (Perplexity's search-augmented model)
Geo-targeting: Supported via Perplexity API parameters
Citations: Rich structured source data from API response — includes titles, URLs, and snippet text
Search queries: Captured from API metadata
Availability: Lite plan and above

Best citation data Perplexity provides the cleanest citation data of all four platforms. If source/citation tracking is your primary goal, prioritize Perplexity in your LLM selection.