GEO Brand Mentions Scanner
Scans brand presence across AI-cited platforms (YouTube, Reddit, Wikipedia, LinkedIn) and produces a 0-100 Brand Authority Score with platform-specific recommendations.
Evidence shows: skill uses only Read/Grep/Glob/Bash/WebFetch/Write; the embedded Python snippet queries public Wikipedia/Wikidata APIs with a custom User-Agent; data flow is largely transparent; output is written to a working-directory file; no credential access, no covert exfiltration, no destructive defaults — a low-risk read-only analysis skill. Deductions: Bash permissions unscoped, no user-confirmation mechanism, no rollback notes, publisher unverified, so not full marks.
Evidence shows: SKILL.md provides a structured procedure, per-platform scoring rubrics, and a deliberate dual Wikipedia verification (API-first). Deductions: core steps rely on 'site:' search operators via WebFetch, which is unreliable without a search backend and has no documented fallback; repo tests cover only fetch_page.py SSR heuristics, not the brand-scan key paths; no script-side error handling or failure feedback for abnormal input — capped below 10.
Evidence shows: clear target scenario (brand AI-visibility audit), inputs, check items and 0-100 rubrics per platform, and a semantically precise trigger description. Deductions: no declared non-fit boundaries (personal brands, non-English brands); entirely dependent on YouTube/Reddit/LinkedIn/Wikipedia, which are restricted or unreachable from mainland China; no Chinese-language support — a material environment penalty for FollowSkills users.
Evidence shows: well-layered docs (insight → platform rubrics → formula → procedure → output template → reference data), a complete Markdown report template, MIT license, and repo-level uninstall instructions. Deductions: no per-skill versioning/changelog, no FAQ, no known-limitation disclosure (search API limits, estimation uncertainty), maintenance responsibility visible only at repo level.
Evidence shows: produces a structured Markdown report (score table, platform detail, staged recommendations, competitor comparison) that is complete and directly usable, with real marginal value over manual research. Deductions: platform scores rest heavily on subjective estimation; the central claims (Ahrefs Dec 2025 study of 75K brands, 0.737 correlation) are author-reported and unverifiable, with the Dec 2025 citation date itself suspect; static-review cap of 7 applies.
Evidence shows: named sources (Ahrefs, Profound, Terakeet) with some coefficients, but all second-hand; the repo's real test suite covers fetch_page.py only, unrelated to this skill's key paths; no reproducible sample outputs or third-party verification. Deductions: key quantitative claims untraceable, fact and inference not separated (correlation treated as causation), hence a low score.
- Core data claims (e.g., 'Ahrefs Dec 2025 study of 75K brands, correlation 0.737') are author-reported only; independently verify before relying on them.
- Many checks depend on 'site:' search operators plus WebFetch; real execution may fail or yield estimates — treat report scores as rough indicators, not precise conclusions.
- The skill depends entirely on YouTube/Reddit/LinkedIn/Wikipedia, which may be unreachable from mainland-China networks, and offers no Chinese-language support.
- Bash permissions are unscoped; review the embedded Python snippet's network targets before running; publisher identity is unverified.
- Recommendations include promotional advice (e.g., hiring consultants) consistent with the repo's Skool-community marketing; exercise judgment.
What does this skill do, and when should you use it?
This is a sub-skill of the geo-seo-claude repository (skills/geo-brand-mentions/SKILL.md), focused on brand authority signals for GEO (Generative Engine Optimization). It is grounded in an Ahrefs study from December 2025 covering 75,000 brands, which found unlinked brand mentions correlate roughly 3x more strongly with AI citations than backlinks. The skill scans each platform, scores it against explicit rubrics, and computes a weighted composite score (YouTube 25%, Reddit 25%, Wikipedia 20%, LinkedIn 15%, other platforms 15%). It then outputs a GEO-BRAND-MENTIONS.md markdown report containing per-platform breakdowns, sentiment analysis, and time-phased recommendations.
Collects brand name, founder, domain, industry, and competitor info; uses WebFetch with site: searches to check official channels/accounts, third-party mention counts, recommendation contexts, and sentiment; verifies Wikipedia via a Python API check against en.wikipedia.org and wikidata.org, explicitly requiring the API before search-based methods to avoid false negatives; scores each platform 0-100 using detailed rubric tables and applies the weighted formula to produce a Brand Authority Score (0-100, banded Dominant/Strong/Moderate/Weak/Minimal); optionally runs a competitor comparison; writes everything to GEO-BRAND-MENTIONS.md with immediate/short-term/long-term action lists.
- A GEO/SEO agency auditing a client's AI visibility needs a quantified view of brand presence on AI-cited platforms plus a deliverable report
- A SaaS company wants to know why ChatGPT/Perplexity recommends competitors and which platforms are its weakest signals
- A marketing team, before shaping a Reddit or YouTube strategy, wants current mention volume, sentiment, and recommendation positioning
- A new brand or local business checks whether Wikipedia/Wikidata entity signals exist so AI systems treat it as a 'real' entity
- An agency uses the competitive comparison table to calibrate scores and justify improvement priorities to clients
What are this skill's strengths and limitations?
- Explicit, reproducible scoring: per-platform rubric tables plus a transparent weighted formula
- Multi-method Wikipedia/Wikidata verification (API-first plus direct URL checks) reduces false negatives
- Complete report structure (platform breakdowns, sentiment, competitor comparison, phased recommendations) ready to deliver to clients
- Weights are grounded in published research (Ahrefs Dec 2025, Profound, Terakeet) rather than arbitrary choices
- Platform scanning relies on site: searches and WebFetch; mention counts and subscriber figures are estimates, not verified data
- Rubric thresholds are hand-authored; there is no test suite or accuracy validation
- Ahrefs data is correlational, not causal, and exact coefficients for some platforms (e.g., Reddit) are unpublished — 'High' is qualitative
- The 'other platforms' bucket (news, podcasts, forums) only gets coarse presence checks
- Results fluctuate with search-tool behavior and time, so the same brand can score differently across runs
How do you install this skill?
Install the collection (this skill ships with it): on macOS/Linux run curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/geo-seo-claude/main/install.sh | bash; on Windows use install-win.sh from Git Bash. Requires Python 3.8+, Claude Code CLI, and Git; dependencies are installed into an isolated venv at ~/.claude/skills/geo/.venv/. The source does not document installing this sub-skill standalone.
How do you use this skill?
After installing, run /geo brands <url> in Claude Code, or simply ask it to scan a brand's cross-platform presence. The skill gathers brand info, scans and scores each platform (Wikipedia verified via the Python API first), computes the weighted score, and generates GEO-BRAND-MENTIONS.md. Providing the brand name, founder, and main competitors up front yields more accurate results and enables the competitive comparison.