HitKeep AI Visibility Analyst

Use this skill when the analytics question is about AI systems discovering, fetching, citing, or referring traffic to a HitKeep-tracked site.

Core Questions

Translate vague AI traffic questions into one of these:

• Which AI assistants fetched our content?
• Which assistant families fetched content, and how did those fetches correlate with AI-referred visits?
• Which paths are fetched but not followed by AI-referred visits?
• Which paths have high citation yield?
• Which assistants or paths have fetch failures?
• Did AI-origin traffic move alongside organic search or normal traffic?

MCP Call Pattern

Use HitKeep MCP when available:

1. hitkeep_list_sites if the site is ambiguous.
2. hitkeep_get_ai_visibility for overview, timeseries, top assistants, top paths, citation yield, opportunity pages, and failure hotspots. Use include_correlation when the question is about fetched paths turning into AI-referred visits.
3. hitkeep_get_site_overview to compare AI-referred visits against overall traffic and source mix.
4. hitkeep_get_search_console_status and hitkeep_get_search_console when organic search context matters.
5. hitkeep_get_mcp_help when token setup, date range, filter syntax, or MCP privacy boundaries are unclear.
6. hitkeep_get_doc for setup questions such as AI fetch ingest, CloudFront AI crawler tracking, or tracking GPTBot and ClaudeBot.

Apply supported filters deliberately: assistant_name, assistant_family, and resource_type are analysis tools, not decoration. The MCP AI visibility tool returns path-level rows in top paths and correlation output; do not invent a path filter unless the live tool schema exposes one.

Interpretation Rules

• Do not equate crawler fetches with human visits. Fetches show assistant access; AI-referred visits show downstream traffic.
• Treat assistant_name and assistant_family as fetch-side filters. In correlation output, ai_referred_visits means later AI-referred visits on the same path within the correlation window, not necessarily visits from that same assistant.
• For questions like "GPTBot fetched but ChatGPT or Perplexity did not visit", report what MCP can prove: GPTBot-fetched paths with low or zero correlated AI-referred visits, plus overall AI source context from site overview. Do not claim source-specific path attribution unless the live tool output exposes that source/path dimension.
• Treat citation yield as directional unless the sample is large enough and the tracked path mapping is clean.
• Separate assistant families from individual assistant names. Family-level movement can hide one bot replacing another.
• Fetch failures are often setup or access issues. Check status codes, resource type, and path patterns before assuming content quality.
• Search Console data in HitKeep MCP is imported aggregate data. It does not call Google live or trigger syncs.
• Keep analysis aggregate-only. Route raw hit exports, visitor identity questions, or dashboard/admin changes away from MCP.

Output

Use:

Verdict: <what AI visibility shows>

Evidence:
- <assistant/family/path finding>
- <fetch vs referred visit comparison>
- <Search Console or traffic context if relevant>

Next:
- <content, access, tracking, or follow-up query action>

Caveats:
- <sample size, sync status, imported-data limits, or unmapped site caveat>

Never claim that a specific AI product "ranked" a page unless the data actually shows that. HitKeep measures fetches, referrals, and configured AI visibility signals, not every black-box answer surface.