Self-audit #1 — 5-notebook set
Why 5 notebooks and not 1 long PDF? Each notebook is a 5-minute read, can be shared with a different stakeholder (CMO / content lead / dev lead / legal / SEO), and is updated independently when the underlying data changes. The 5-notebook structure is the contract we make with every Standard-tier customer. Showing it on ourselves is the dogfooding promise of the v3 plan.
The 5 notebooks
| # | Notebook | What it answers | Audience | Read time | Link |
|---|---|---|---|---|---|
| 1 | Index | "What is my AI-search score right now, in one page?" | CEO, CMO | 2 min | nb1-index.html |
| 2 | Intent | "Which query categories am I winning, tying, or losing on?" | Content lead, SEO | 5 min | nb2-intent.html |
| 3 | Content | "Which publishers and aggregators do the LLMs trust in my space?" | PR, partnerships | 5 min | nb3-content.html |
| 4 | Quotables | "What exact phrases are LLMs saying about my category — and where am I absent?" | Copywriters, PR | 5 min | nb4-quotables.html |
| 5 | Strategy | "What are the 5–10 prioritized actions to move the needle, and what lift should I expect?" | Head of growth, dev lead | 8 min | nb5-strategy.html |
How to read this set
- Start with Notebook 1 (Index) to see the overall score and the citation-source landscape.
- Then go to Notebook 2 (Intent) to see which prompt categories are bleeding the most citations.
- Notebook 3 (Content) tells you which publishers to pitch first — because LLMs are already citing them in your category.
- Notebook 4 (Quotables) gives copywriters the verbatim LLM language to pattern-match.
- Notebook 5 (Strategy) turns all of the above into a prioritized backlog with effort estimates and expected citation-rate lift.
Reproduction
All 270 raw datapoints (9 engines × 30 prompts) are in the public agentgeek-geo/audit-logs repo. Each notebook points to the specific JSON files it cites, with a direct URL.
Limitations and stochasticity
LLM responses are stochastic. A re-run of the same 270 prompts within 24 hours will vary by ±10% on (Claude, Gemini, DeepSeek) and ±0% on (Doubao, ERNIE). All numbers in these notebooks are exact counts from a single run on 2026-06-11 09:00–09:08 UTC+8; they are not projected averages. The next self-audit snapshot is scheduled for 2026-07-01.
Start with Notebook 1 → ← Back to 1-page summary Download raw JSON