CITED Score

CITED Score is a free, open-source AI SEO tool - a crawler and auditor for GEO (generative engine optimisation) and AEO (answer engine optimisation). It crawls your entire site, renders every page with headless Chrome the way a non-JS AI crawler does, and scores how citable and extractable each page is for ChatGPT, Perplexity, Claude, Gemini, Copilot and Google AI Overviews. It reports one honest headline number and, more importantly, ends in a ranked action plan.

Think of it as Screaming Frog for AI search - built for AEO, GEO and LLM SEO. It is the free companion auditor to the book CITED by Chris McCarron (GoGoChimp).

The book: gogochimp.com/cited  ·  This tool: gogochimp.com/cited-score  ·  Built by Chris McCarron at GoGoChimp, the AI-SEO / AI-CRO agency.

Scores are given three ways:

• Overall CITED Score (0-100).
• By pillar - the three questions an engine asks (CITED ch3): Known (do I know you?), Findable (can I find your answer?), Trusted (do I trust you enough to name you?).
• Per engine - ChatGPT, Perplexity, Google AI Overviews, Gemini, Copilot, Claude, since each weights different signals.

No API keys, no LLM. Uses your installed Chrome or Edge to render (essential: curl/requests miss JS-injected schema).

Download (Windows)

Grab the latest CITED-Score.exe from the Releases page. Double-click it; no Python needed. It still needs Google Chrome or Microsoft Edge installed (used to render pages) and warns if neither is found. The app checks for newer releases on launch and shows an update banner when one is available.

Why Windows shows a warning (and why it's safe)

On first run, Windows Defender SmartScreen may say "Windows protected your PC" with "unknown publisher". Click More info -> Run anyway. This is not a virus warning.

It appears for any app that is not code-signed. CITED Score is unsigned by choice: as of 2026, signing no longer earns an instant clean download (Microsoft removed that benefit from EV certificates in 2024, so every newly-signed app has to build reputation through downloads anyway), so signing a fast-moving free tool adds cost without removing the first-run prompt.

Your real safeguards instead: the full source is in this repo (inspect it, or build the exe yourself with the steps below); the app makes no network calls except the GitHub update check; and it renders pages with your own installed Chrome/Edge, not a bundled browser. Nothing leaves your machine.

Run from source

Requirements: Python 3.10+, Google Chrome or Microsoft Edge, then pip install -r requirements.txt.

App (no command line):

python app.py

Opens a local web app at http://127.0.0.1:5000. Enter any website, click Run, watch the live crawl, open the report. Reports are saved per domain under reports/ and listed for revisiting; re-running a site shows the since-last-crawl diff.

Command line:

python aiseo_audit.py --url https://www.example.com --out report

• --max-pages 0 (default) crawls the entire site; a number caps it.
• --workers 6 renders N pages in parallel (isolated Chrome profile each).

Outputs report.html (branded, tabbed), report.json, report.csv. Re-running appends to report-history.jsonl and the next report shows a since-last-crawl diff.

The report (tabs)

• Overview - CITED Score, three pillar rings, six engine rings, "do these first", totals, page types, and the diff vs the last crawl.
• Action Plan - every fix ranked by projected score gain (the tool re-scores the whole site with each fix simulated), each showing the +overall and per-engine gain, pillar, chapter, effort, and the affected-page count. Ends in a 30/60/90-day roadmap.
• Issues - grouped by pillar; each carries a why (evidence line), a fix, and the expandable affected-URL list.
• Pages - sortable, searchable table: score, pillar scores, per-engine columns, response time, errors. CSV export.
• Per engine (6 tabs) - that engine's readiness, the signals it weights (with site pass rate), and its worst pages.
• Site structure - by section and crawl depth. Response times - server + render, slowest pages.
• Print / PDF and Export CSV buttons for client delivery.

Ruleset obeys CITED, not GEO folklore

A data-led author's tool must not prescribe what his own book calls a myth:

• llms.txt is informational, not scored - SE Ranking found no correlation with citations (ch5). Shown for reference only.
• Sections are judged "self-contained, no walls of text", not a word-count band - one liftable answer per block, never "chunk artificially" (ch5).
• Answer capsule target is 40-60 words (ch5).
• Every check carries an evidence line naming its source (engine documentation, first-party citation data, or a CITED chapter), and the footer states the honest ceiling: the score estimates citability; it does not measure citations.

Calibrate against real citations (optional)

python aiseo_audit.py --calibrate citations.csv --report report.json

citations.csv is url,citations - export per-URL citation counts from Bing Webmaster Tools > AI Performance. The tool correlates (Spearman) each pillar, engine and check against real citations, and flags which signals to up-weight. (This repo does not ship a citations file; bring your own.)

Desktop packaging

desktop.py wraps the local server in a native window (pywebview / Edge WebView2 on Windows); app.py also runs standalone in a browser.

pip install pyinstaller pywebview
pyinstaller CITED-Score.spec

Ships dist/CITED-Score.exe. Attach it to a GitHub Release tagged with the app version (see APP_VERSION in app.py) so the in-app update check can find it. Mac builds must be done on a Mac.

Brand

Green-on-ink (CITED Score V2.0): ink #0F1410 field, verdict green #42D848 (deep green #1E8A2C for green type on light grounds), paper #EDF0EB type. The mark is a monoline C (the crawl) closing on a green check inside it (the verdict). Display and UI type is Archivo; metrics, labels and code are IBM Plex Mono. The report is built to be screenshotted into LinkedIn posts and client emails.

Notes

• The ruleset encodes CITED's playbook (answer-first, self-contained passages, evidence density, entity/schema clarity, AI-crawler access), sourced.
• Rendering uses isolated Chrome profiles so it never clashes with a running Chrome.