Tracking AI

Tracking AI is a free monitoring platform that systematically quizzes major AI chatbots on IQ tests and political compass questions every week, publishing the results publicly. Created by Maxim Lott (executive producer of Stossel TV), the site tracks both verbal and vision AI models including GPT, Claude, Gemini, Grok, DeepSeek, and more. It provides interactive charts showing AI IQ scores over time and political bias positions, helping users understand the ideological and cognitive tendencies of the AI tools they use. The site uses an offline Mensa-member-created IQ test not in any AI training data, ensuring uncontaminated benchmarking.

• Political Compass Testing — All tracked AIs are asked standardized political compass questions weekly, with results plotted on an economic vs. social axis to reveal ideological leanings.
• IQ Benchmarking — AIs are tested using both the public Mensa Norway test and an exclusive offline test never published on the internet, preventing training data contamination.
• Vision Model Support — Vision-capable AI models are tested using actual test images rather than verbalized prompts, enabling separate tracking of multimodal performance.
• Historical Score Tracking — Interactive time-series charts show how each AI's IQ and political scores have changed over time, with daily and moving-average views.
• Per-Question Answer Viewer — Users can browse every AI's exact answer to every test question, including reasoning and refusal tracking.
• Searchable Database — A searchable database of all AI responses is available for deeper research and comparison.
• Refusal Tracking — The site records when AIs refuse to answer questions, using the most recent prior answer for scoring and flagging the refusal.
• Model Info Pages — Each tracked AI has a dedicated info page linking to where users can access that chatbot directly.