An open-source GEO command center that runs on your machine

What it does

GeoLook is a Python-based, self-hosted dashboard for end-to-end Generative Engine Optimization (GEO). It samples answers from up to 17 engines—including ChatGPT, Perplexity, Gemini, DeepSeek, and a first-class matrix of Chinese engines like Doubao and Kimi—then quantifies mention rates, citation share, and rank. Beyond monitoring, it runs a four-layer site audit (Access → Orientation → Understanding → Quotability), generates structured implementation tickets with risk grades and acceptance criteria, provides a content workbench with fabrication-risk linting, and ships deployable assets like llms.txt and JSON-LD snippets. A built-in publishing layer covers channels from WordPress and GitHub to X and Reddit, while per-question before/after sampling and automatic regression reopening close the verification loop.

The interesting bit

Most GEO tools are SaaS dashboards that score you and keep your data; GeoLook keeps everything under a local work/ directory and treats GEO like an ops pipeline rather than a vanity metric. The site audit is unusually paranoid: it differentially probes WAF/CDN rules with real AI-crawler user agents, parses RFC 9309 robots.txt for wildcard specificity, and flags pages that lack independently quotable passages rather than just checking for keywords. It also anchors its scoring on an empirical citation corpus—602 prompts and over 21,000 citations—rather than proprietary heuristics.

Key highlights

• Samples 17 engines (10 via API, 7 manual) and separates Chinese and global metrics.
• Site audit uses a four-layer dependency chain; a failing Access layer makes downstream fixes invisible until resolved.
• Generates tickets with risk grades (quick win / observe / high-risk with rollback discipline) and auto-reopens them on regressions.
• Ships a Chrome extension and sandbox scripts for manual sampling hygiene, including an opt-in auto-run mode with documented ToS guardrails.
• Client-ready deliverables: diagnosis report, strategy, execution plan, ticket CSV, and acceptance sheet exportable in one click.
• Runs fully offline and local; zero mandatory API keys, though automated sampling costs extra.

Caveats

• Single-machine tool with no user accounts or team collaboration features.
• “Suspected negative” flags and some metrics require human review; the UI explicitly labels unmeasured data rather than fabricating scores.
• Publishing integrations deliberately omit platforms without usable official APIs (e.g., LinkedIn, Instagram, Xiaohongshu), offering webhooks instead.

Verdict

Agencies, consultants, and in-house operators who need to package GEO as accountable, repeatable work should look here. If you want a cloud-based, set-it-and-forget-it mention tracker, this is the wrong shape entirely.