Data-driven GEO for Claude Code — measured on real AI answers, not guesses.
Open Profound is an open-source agent skill suite that measures how AI engines actually answer your buyers' questions — ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode — and turns the gaps into concrete page and content fixes.
Most GEO tools predict citability from page structure. Open Profound skips the prediction: it fires your prompt set at the real consumer surfaces through the querying.ai API and scores the answers that came back. One API key. No other vendors, no scraping setup, zero npm dependencies.
Open Profound is the open-source take on
Profound — the category-defining AI-visibility
platform. Commercial platforms (Profound, Peec, Otterly) do this well and
charge accordingly. Open-source trackers mostly query LLM APIs, which are not
the surfaces your buyers see — an API answer from gpt-4o is not the ChatGPT
answer with web search, and nothing from an API resembles a Google AI Overview.
Open Profound runs on real surfaces, inside the agent you already use.
/plugin marketplace add anymorph-ai/open-profound
/plugin install open-profound@anymorph-open-profound
Then set your key (get one at querying.ai):
export QUERYING_API_KEY=vt_...Use a per-user vt_... key — the monitor endpoints authenticate by key shape
and only accept these.
| Command | What happens |
|---|---|
/geo prompts acme.com |
Designs the 10–30 questions buyers actually ask AI in your category |
/geo audit acme.com |
Fires them across engines, reports mentions · citations · share of voice — with the actual answers as evidence |
/geo optimize |
Classifies each loss (content / citability / placement / locale gap) and writes the fix plan |
/geo monitor |
Registers a recurring server-side monitor; pull time-series reports any time |
The loop: prompts → audit → optimize → monitor. Changes get measured against a baseline instead of vibes.
- Which competitor is recommended instead of you, per prompt, per engine — quoted
- Which domains and pages engines cite for your category (your placement targets)
- Where you're strong in
USbut invisible inKR/JPanswers - Whether last month's content work moved the numbers (monitors keep the series)
The engines querying.ai serves to API customers:
CHATGPT · PERPLEXITY · GEMINI · GOOGLE_AIO (AI Overviews) ·
GOOGLE_AIMODE (AI Mode).
Live availability: node scripts/querying.mjs capacity.
Each prompt × engine is one querying.ai task, billed in credits. The skills show the task count before firing anything non-trivial, and monitor creation always shows per-run cost — recurring spend needs eyes-open consent.
MIT. Not affiliated with Profound (tryprofound.com); the name states the goal — an open alternative to that category of tool.