Skip to content

Repository files navigation

LLMScout

Find out if AI search can quote you. One page, one score, one prioritized to-do list — based on what ChatGPT, Claude and Perplexity actually reward.

npx llmscout https://example.com/page

What it is

LLMScout scores a page's AI answer-readiness from 0–100 across nine checks: AI crawler access, answer-first structure, JSON-LD schema, heading hierarchy, author/E-E-A-T signals, extractable Q&A, named-statistic density, content-to-markup ratio, and llms.txt. Every finding comes with a concrete fix, ranked by impact × ease.

What it is NOT

Not a rank tracker, not a monitoring platform, not a full SEO audit. It scores one thing: whether an AI engine that fetches your page can extract and cite an answer from it.

Usage

llmscout <url>                     # human-readable report
llmscout <url> --json              # stable JSON (CI/programmatic)
llmscout <url> --min-score 75      # exit 1 below threshold (CI gate)
llmscout --explain                 # every check, weight and rationale
llmscout <url> --only json-ld,eeat-signals
llmscout <url> --fixes 5 --verbose
llmscout serve --port 3000         # local web report UI

Exit codes: 0 ok · 1 below --min-score · 2 fetch/URL error.

One-click local web UI

Double-click scripts/start-llmscout.command (macOS). It installs dependencies on first run, builds, starts the server, and opens your browser at http://127.0.0.1:3000 automatically. Or from the terminal:

npm start

Add your LLM API key (optional, BYOK)

The upcoming AI narrative feature ("why you're not cited", v1.1) uses your own LLM key. Setup is two steps:

cp .env.example .env     # 1. create your local env file
# 2. open .env and paste your key:
#    LLMSCOUT_LLM_API_KEY=sk-...

Anthropic and OpenAI keys also work via their standard variables (ANTHROPIC_API_KEY / OPENAI_API_KEY), either in .env or exported in your shell. Verify with:

llmscout key   # → ✓ LLM key found (sk-a…f3k2) via .env:LLMSCOUT_LLM_API_KEY

.env is gitignored — the key stays on your machine and is only ever sent to your own LLM provider.

Methodology — honest by design

Weights live in one auditable file (src/core/weights.ts) and are evidence-informed (v0.2, July 2026): each weight is tied to published research and to our own live citation tests. The full rationale, evidence table, and citation list are in METHODOLOGY.md. The short version of how weights are set:

  • Controlled experiments outrank correlations. Three exist publicly: the GEO paper (KDD 2024 — adding statistics/citations/quotes lifted generative visibility 30–41%), Ahrefs' schema intervention test (May 2026 — newly added JSON-LD produced no citation uplift, so we weight schema at just 8), and the Vercel/MERJ crawler study (Dec 2024 — GPTBot/ClaudeBot/PerplexityBot execute zero JavaScript, so server-rendered content weighs 12).
  • We test our own tool against reality. We run real queries through Perplexity, ChatGPT and Gemini, capture what they actually cite, and scan cited vs non-cited pages. Crawler access was the only signal positive in both test rounds — hence its top weight (25).
  • Nulls get weighted like nulls. llms.txt sits at 2/100: Google says no AI system uses it, and two independent large-N studies (SE Ranking 300K domains, Trakkr 38K) found zero citation effect.
  • We publish what the score can't do. Engines cite high-authority domains even when their on-page readiness scores an F (we watched Perplexity cite a bank page we score 31/100). LLMScout predicts citations among comparable-authority sites competing on the same topic — the on-page half you control.

Run llmscout --explain or see /methodology in the web UI for the same information.

Development

npm install
npm test         # vitest
npm run build    # tsc → dist/
npm run dev -- https://example.com

MIT © Anthony Guidry / SCOAEONGEO

About

LLMScout scores a page's AI answer-readiness from 0–100 across nine checks: AI crawler access, answer-first structure, JSON-LD schema, heading hierarchy, author/E-E-A-T signals, extractable Q&A, named-statistic density, content-to-markup ratio, and llms.txt. Every finding comes with a concrete fix, ranked by impact × ease.

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages