Three audiences read the same HTML. Optimise once, win three channels.

OmniRank is not published to PyPI, npm or a container registry yet — install it from source into a virtual environment. Homebrew and other PEP 668-managed Python installs refuse a bare pip install, so activate a venv first:

git clone https://github.com/bemoshiur/OmniRank.git
cd OmniRank

python3 -m venv .venv && source .venv/bin/activate
make install

# Audit any live site — no configuration required
python3 -m omnirank.cli audit https://example.com

Output is a scored report plus a prioritised fix list, written to .omnirank/reports/<date>-audit.json.

For a configured repo, copy templates/omnirank.config.example.json to omnirank.config.json, fill it in, then:

python3 -m omnirank.cli audit --config omnirank.config.json --fail-on h1 canonical schema
python3 -m omnirank.cli geo   --config omnirank.config.json --out public

Exit codes: 0 clean · 1 a configured gate failed · 2 usage or config error. The non-zero exit makes it a drop-in CI check.

Prefer not to clone? Grab the packaged skill:

⬇ Download OmniRank skill (.zip)

Unzip it into your Claude Code plugins directory:

unzip omnirank-skill.zip -d ~/.claude/plugins/

Building it yourself is one command:

./scripts/build-skill-zip.sh          # -> dist/omnirank-skill-<version>.zip

Adapters for WordPress, Jekyll, Shopify, Astro, Vue and Svelte land at v1.0.

Real-only. No fabricated statistics, ratings, reviews or testimonials. facts.statistics carries only entries explicitly marked published; AggregateRating without a real ratingCount is reported as an error, not a warning.

Never claim an unevaluated gate passed. An unreachable URL is an error. A layer whose gates did not run is absent from the score map rather than scored 100.

Secrets are pointers. Config holds env:NAME only. A missing variable fails loudly — a skipped submission is otherwise indistinguishable from a successful one in logs.

Audit diagnoses; it never edits.

• It does not force rankings. Nothing can. It dominates the signals search engines use.
• It does not send cold email. Outreach is drafted for human review and manual send.
• It does not abuse the Google Indexing API. That API covers JobPosting and BroadcastEvent only; misuse earns a manual action.
• It does not generate AI slop. Every generated page expects a human edit pass.
• It does not ship thin programmatic pages. Matrices are pruned to real demand signals.

The wiki covers configuration, every gate, adapters, the report schema, and CI recipes. Questions and results belong in Discussions.

OmniRank implements and builds on public standards and published research:

• schema.org — structured-data vocabulary
• IndexNow — instant indexing protocol
• llms.txt — the proposal for AI-readable site indexes
• Sitemaps XML and the Robots Exclusion Protocol (RFC 9309)
• arXiv 2311.09735 — GEO: Generative Engine Optimization (Princeton, KDD 2024). Source of the finding that statistics, quotations and cited primary sources lift AI visibility.
• arXiv 2509.10762 — GEO-16, generative-engine optimisation criteria
• Ahrefs — AI-search and brand-mention correlation studies
• Seer Interactive — ChatGPT / Bing citation-overlap analysis

Correlations from that research are observational, drawn largely from English-language B2B datasets, and are treated as directional. Validate empirically per site.

Built and maintained by S M Moshiur Rahman at Public Pulse Agency, Dhaka — across publicpulse.com.bd, tenderpulse.com.bd and pulsetoday.com.bd.

S M Moshiur Rahman — Director of Business & Operations, Public Pulse Agency

For bugs and feature requests, please use Issues rather than direct message — it keeps the answer searchable for the next person. For consulting, partnerships, or anything project-specific, WhatsApp gets the quickest reply.

Code is MIT. Documentation and the content corpus exposed via llms.txt and llms-full.txt are CC BY 4.0.