Optimize Search Visibility

Optimize Search Visibility

Score and improve a website's visibility in both ranking search (SEO) and citation-based AI answer engines (GEO/AEO). Report what was actually observed versus inferred; never invent a ranking or citation.

Attribution

Ported from two prior open-source skills. See NOTICE.md, LICENSE-geo-seo-claude, and LICENSE-seo-geo before redistributing this skill on its own.

Inputs

Require:

• target URL(s) and the business type or product;
• target audience, key questions the page should answer, and locale;
• named competitors, if any;
• observation date;
• DATAFORSEO_LOGIN / DATAFORSEO_PASSWORD for paid SEO features — proceed with the free path and say so explicitly when absent;
• OPENAI_API_KEY / PERPLEXITY_API_KEY for live citation measurement — skip that check explicitly when absent, never silently.

Never request or persist a raw credential in a document, commit, or chat message. Read keys only from environment variables via scripts/credential.py.

Routing

• Single check ("citability score", "is our crawler access blocked", "keyword research for X") → run the one matching script or reference below, report that category alone.
• Full audit ("audit our GEO", "full SEO+GEO review") → run the Full Audit flow.

Capabilities

SEO (requires DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD; seo_audit.py is free)

GEO / AEO

Citability scoring weights are grounded in Princeton GEO (KDD'24), AutoGEO (ICLR'26), and C-SEO Bench (2025) — see references/citability-research.md. Re-verify against current research before reuse; source-dated 2026-06-05.

Distributional scoring

Report AI-visibility scores as a range with a sample size ("approx. 55, range 48-61, N=5"), never a bare single number — that is false precision. Run:

python3 skills/optimize-search-visibility/scripts/distribution.py <script_name> <url> --runs N

The local citability heuristic is deterministic (fixed HTML in, same score out every run) — that is not evidence of stability, it means no live model was queried. Only live_citation.py measures real answer-engine variance. Cite the deterministic heuristic as a proxy, not as measured citation behavior.

Full Audit flow

For a full-site request, run these steps yourself in order — do not assume a named subagent exists on the installing system; every check here is one script or reference read, sequenced by this skill.

0. Preflight: check for requests/beautifulsoup4/lxml (see requirements.txt), DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD, and OPENAI_API_KEY/PERPLEXITY_API_KEY. Report each missing piece as "skipped — no credential", not as a zero score.
1. Discovery: fetch the target URL with scripts/fetch_page.py and identify business type and key pages.
2. Category checks, each independent and each producing its own findings:
   • AI citability — scripts/citability_scorer.py
   • Brand authority — scripts/brand_scanner.py
   • Content E-E-A-T — references/geo-research.md checklist against the fetched page
   • Technical GEO (crawler access, llms.txt) — robots.txt/meta checks + scripts/llmstxt_generator.py
   • Structured data — schema/*.json templates against the fetched page
   • Platform optimization — references/platform-algorithms.md
3. Synthesis: combine into one GEO score (0-100): AI Citability 25%, Brand Authority 20%, Content E-E-A-T 20%, Technical GEO 15%, Structured Data 10%, Platform Optimization 10%.

Outputs

Create only the reports the requested operation needs:

• geo-citability-score.md
• geo-crawler-access.md
• geo-brand-mentions.md
• geo-platform-optimization.md
• geo-schema-report.md
• geo-audit-report.md — full-audit output: combined score, per-category breakdown, issue severity (Critical/High/Medium/Low), a 30-day action plan, and an appendix of pages analyzed.
• llms.txt — generated file for the target site, not this repository.

Read references/output-contract.md before writing them.

DataForSEO credential branch

• Credential present → full SEO feature set.
• Credential absent → run scripts/seo_audit.py (free) plus WebFetch-based checks; all GEO features still work unaffected. For a paid SEO feature with no credential, explain how to set one and stop — do not fail silently or fabricate a result.

Boundary

• Does not publish, deploy, or modify a live site — every output is a report or a standalone file (like llms.txt) for the user to review and apply themselves.
• Does not report a ranking, citation, or score the underlying script or fetch did not actually observe.
• Not for questions unrelated to a website's search or AI-answer visibility.

References

• Methodology: references/geo-research.md, references/platform-algorithms.md, references/seo-checklist.md, references/schema-templates.md, references/tools-and-apis.md
• Scoring basis: references/citability-research.md, references/platform-citation-factors.md (staleness warning included)
• Attribution: NOTICE.md, LICENSE-geo-seo-claude, LICENSE-seo-geo