Off-Site & AI Referral Analyzer
Audit backlink quality and measure traffic and conversions from AI assistants.
The evidence shows least-privilege intent through a WebFetch-only declaration, user confirmation before saving, measured/user-provided/estimated labeling, untrusted-content handling, and robots.txt/TOS safeguards. However, GA4/GSC/server logs may contain sensitive data, external flows through Tavily/GDELT/Cloudflare and the actual memory-write boundary are not fully specified in this skill, and backlink remediation can have consequential effects; 8 points are deducted.
The evidence shows a two-mode contract, input requirements, stop conditions, abnormal-input handling, metric labeling, and output templates. Static review cannot reproduce the key paths; connector, ledger, and memory-write references may not be available under the only declared WebFetch tool, and there are no skill-specific tests. The score is therefore capped by static calibration and reduced by 11 points.
The evidence clearly defines the audience, trigger phrases, backlinks and ai-referrals modes, inputs, outputs, and exclusions, with Chinese trigger terms included. Boundaries around data formats, host capabilities, and reachability of some external services are incomplete, and mainland-China availability is unverified; 4 points are deducted.
The evidence provides strong progressive structure with Quick Start, contract, scope guard, references, templates, security guidance, Apache-2.0 licensing, and version metadata. Publisher identity is unverified; maintenance ownership and update paths are not clear enough, and SECURITY.md identifies 17.x as current while the skill and README state 18.0.0, so 5 points are deducted.
The evidence defines concrete workflows, done conditions, reusable report templates, source labels, and handoff paths for both modes. No real execution output or third-party validation is included, and useful results depend on complete user exports or functioning connectors, so the static cap applies and 8 points are withheld.
The evidence includes per-metric source labels, a measurement protocol, a link-quality rubric, benchmarks, and change-tracking procedures. There is no skill-specific test suite, execution log, or independent third-party reproduction evidence, so the static-review ceiling of 5 applies.
- Before using server logs, GA4, or GSC exports, define redaction, data minimization, and third-party transfer boundaries.
- Before any disavow or other link-remediation action, preserve the raw export, manually review candidates, and obtain explicit user approval; candidates must not be treated as confirmed risk.
- Reconcile the 18.0.0 skill/README version with SECURITY.md's 17.x supported-current-line statement, and verify that the host actually provides the referenced connectors, scripts, and memory-writing capability.
- Tavily, GDELT, and Cloudflare Radar may be unreachable from mainland-China networks; prepare user-provided exports or accessible alternative data sources.
What does this skill do, and when should you use it?
Off-Site Signal Analyzer is an SEO/GEO evaluation skill in the Aaron Marketing Skills repository. It has two modes: backlinks for link-profile analysis and ai-referrals for traffic sent by assistants such as ChatGPT and Perplexity. It reads user-provided backlink exports, GA4, Search Console, or server-log data and labels figures as Measured, User-provided, or Estimated. It fits teams investigating link risk or AI-referred traffic, but it does not cover internal linking, keyword rankings, or full stakeholder reporting.
In backlinks mode, it analyzes referring domains, anchor-text distribution, link types, quality and toxicity indicators, competitor link gaps, link changes, and link-building or disavow actions. In ai-referrals mode, it adapts an AI-source regex to observed values, measures AI-assistant sessions and trends, lists top landing pages, and compares AI engagement and conversion with organic traffic for the same window. Inputs can include pasted CSVs, GA4 exports, Google Search Console data, or server-log slices; the documentation also describes WebFetch, optional connectors, GDELT, Tavily, Cloudflare Radar, and a local Python ledger workflow. Outputs include a user-facing report, a standard handoff summary, and—after confirmation—an optional file under memory/monitoring/.
- An SEO team has a backlink export and needs toxic-link review, a labeled toxicity ratio, and disavow priorities.
- A growth team wants to compare its referring domains with competitors and identify link-building opportunities.
- A brand wants to know whether ChatGPT, Perplexity, Gemini, Copilot, or Claude is sending site traffic.
- An analyst needs an apples-to-apples comparison of AI referrals and organic traffic, including conversion performance.
- A marketing lead wants to hand off off-site findings to a domain-authority audit or a broader performance report.
What are this skill's strengths and limitations?
- Keeps backlink and AI-referral datasets separate instead of collapsing them into one metric.
- Requires source labels for metrics and uses N/A when required evidence is unavailable.
- Covers toxicity, competitor gaps, AI-source matching, landing pages, trends, and AI-versus-organic conversion.
- Supports a keyless workflow based on pasted data, with optional connectors and ledger-based tracking.
- Core metrics cannot be measured without backlink exports, GA4, Search Console, or server-log data.
- AI referrals show that an AI answer sent a click, not how prominently or authoritatively the site was cited.
- The skill does not calculate CITE scores, analyze internal links, track keyword positions, or assemble stakeholder reports.
- The source provides no independent test-suite results or concrete configuration evidence for each optional platform connector.
How do you install this skill?
The repository supports Claude Code, Agent Skills-compatible hosts, and a plain git clone. For Claude Code, run /plugin marketplace add aaron-he-zhu/aaron-marketing-skills, then /plugin install aaron-marketing@aaron. For a generic host, run npx skills add aaron-he-zhu/aaron-marketing-skills -s offsite-signal-analyzer, or clone it with git clone https://github.com/aaron-he-zhu/aaron-marketing-skills. The source does not document a dedicated destination directory for the single-skill installation.
How do you use this skill?
Example prompts include Analyze backlink profile for example.com, Find link-building opportunities by analyzing competitor1.com, competitor2.com (--mode backlinks), and Track AI referral traffic for example.com over the last 90 days (--mode ai-referrals). You can also specify --mode backlinks or --mode ai-referrals explicitly. Without a mode, link, anchor, toxic, or referring-domain language selects backlinks; AI-assistant, ChatGPT, Perplexity, or GA4-referral language selects ai-referrals. If both intents are genuinely present, the skill asks which mode to run first. Without the required data or a connected tool, it does not estimate core metrics from the domain alone.
How does this skill compare with similar options?
Unlike site-structure-optimizer, this skill analyzes external links rather than internal link structure. Unlike rank-tracker, it does not report keyword positions. Unlike performance-monitor, it does not assemble multi-metric stakeholder reports. Backlink toxicity or authority concerns can be handed to domain-authority-auditor for formal CITE scoring.