citeready

Can AI assistants read, trust and cite your business page? Pure stdlib, zero dependencies, one command.

When someone asks ChatGPT-class assistants for "a good bakery nearby", the answer is assembled from pages the assistant can actually read: structured data, consistent contact details, plain crawlable HTML. citeready audits one page the way an answer engine sees it and hands you a scored, explainable report. Every check is a plain deterministic rule you can read in the source — no model in the loop, no network calls beyond the page itself.

Install

pip install citeready

Usage

$ citeready https://vegabakery.example/
citeready — https://vegabakery.example/
score 96/100

Structured data
  ✓ JSON-LD structured data found (1 block)
  ✓ all JSON-LD blocks parse
  ✓ business node found (Bakery)
  ✓ name present
  ✓ address present
  ✓ telephone present
  ...
  ✓ phone matches the visible page

Page basics
  ✓ title: «Vega Bakery — sourdough & pastry in Valencia»
  ✓ meta description present
  ...

Site signals
  ✓ AI crawlers are not blocked
  ! no llms.txt (emerging, optional standard)
  ✓ sitemap.xml present

Also takes a local file (citeready page.html) or stdin (citeready -); --json for machine-readable output, --strict to exit non-zero when any check fails (handy in CI).

From Python:

from citeready import audit

report = audit("https://vegabakery.example/")   # or audit(html_string)
report.score                                    # 0–100
[(f.check, f.level, f.message) for f in report.findings]

What it checks

Structured data — the part assistants quote from:

• JSON-LD is present and parses (@graph handled)
• a business node exists (LocalBusiness/Organization/subtypes — Bakery, BeautySalon, LegalService, …)
• must-have properties: name, address, telephone, url (missing = fail)
• rich properties: description, geo, openingHours, sameAs, image, priceRange (missing = warn)
• consistency: the name, phone and street from structured data actually appear on the visible page — mismatched data reads as unreliable

Page basics — title (present, sane length), meta description, exactly one h1, lang attribute, canonical URL, no noindex, Open Graph pair.

Site signals (URL mode) — robots.txt does not block AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, …), llms.txt, sitemap.xml.

Scoring is the plain weighted share: pass = 1, warn = ½, fail = 0, ×100. No hidden weights, so two reports are always comparable.

Honest limits

Static HTML only — no JavaScript is executed, so a page that renders its content client-side will score poorly here. That is partly the point (most answer-engine crawlers read static HTML too), but a low score on a JS-heavy site deserves a manual look. Checks are heuristics with documented thresholds, not a certification; robots.txt parsing is deliberately naive (full-block detection, not path matching).

Scope is frozen

This package does one thing: it audits a page and reports. Bug reports (wrong counts, crashes on valid HTML, missed JSON-LD) are very welcome. Feature requests — crawling whole sites, fixing what it finds, tracking scores over time — are out of scope for now.

Built to power Laspi

citeready is the measuring half of the AI-visibility work inside Laspi — a marketing employee in a small-business owner's phone. Laspi keeps a living memory of the business (every fact confirmed by the owner), writes platform-native posts, images and short videos from it, and looks after the business's visibility in AI answers — in whatever language the business sells in. This audit is how that work starts: measure first, then fix.

The measuring is open source. The fixing is the product: laspi.pro.

License

MIT © 2026 GradeBuilder S.L.