Engine hub · Claude
Claude cites reluctantly. Give it no reason to hedge.
Claude reasons carefully over long context — which means thin evidence becomes a hedge and strong evidence becomes a confident recommendation. Structure and transparency decide which one you get.
- Claude-SearchBot retrieval
- ClaudeBot is training-only
- Extractable passages win
- MCP-native surface
Short answer
How does Claude decide?
Why does Claude hedge about some brands?
Because hedging is the honest output when evidence is thin: vague pricing, merged entities, claims without corroboration. Claude would rather qualify than invent — which means a hedge is a diagnosis, not an insult. Fix the evidence and the hedge becomes a recommendation on the next cycle.
What is the Claude bot split?
Anthropic runs separate fleets: ClaudeBot for training, Claude-SearchBot for index building, Claude-User for live fetch, plus anthropic-ai for agentic crawling. Retrieval answers depend on the search and user bots. The audit reports each token separately with longest-prefix matching so training opt-outs never masquerade as retrieval blocks.
Why does MCP matter more here?
Claude-native tooling — Claude Desktop, Claude Code, Cursor — speaks Model Context Protocol natively. Six brand-scoped tools expose citation reports, prompts, sources and the action queue where engineers already work, so visibility data joins the build loop instead of living in another dashboard tab.
The Claude playbook
Four moves from hedge to recommendation
Configuration, extractability, transparency, integration — in that order.
- 01
Split the Anthropic fleets correctly
Allow Claude-SearchBot and Claude-User for retrieval answers. ClaudeBot is training collection — blocking it does not block answers, but blocking search does. Report each verdict separately so a single bad edit cannot silence the surface.
- 02
Write answers Claude can lift verbatim
Question-shaped headings with a forty-to-sixty-word self-contained passage immediately beneath, FAQPage or HowTo markup, no dependency on surrounding context. Long context does not mean long answers win — extractable ones do.
- 03
Expose transparent offers
Product or SoftwareApplication schema with real price or priceRange. Claude penalizes vague pricing more harshly than most surfaces; transparent offers are a stance upgrades, not just schema points.
- 04
Pull numbers into the IDE
Six brand-scoped MCP tools over Streamable HTTP let Claude Code and Cursor ask what changed this week without leaving the editor. Same numbers as the dashboard, bearer-token isolated per brand.
get_citation_reportCurrent and previous snapshot, per-engine breakdown, delta.list_promptsTracked buyer queries with stance, sentiment and position per engine.list_sourcesSource domains, brand presence, competitor presence, opportunities.list_actionsPending remediation queue with draft copy and priority.mark_action_doneLocks the current snapshot as the impact baseline.dismiss_actionRemoves an action from the queue.
- 40–60w
- Answer passage
- 2+
- sameAs links
- 6
- MCP tools
- 60
- Checks / week
Self-contained, quotable
Stops namesake merging
Cursor + Claude-native
Including Claude
Claude questions
Asked by engineering teams
Structured for verbatim quoting.
Do you query the AI engines automatically?
Run generation, response extraction, scoring, drift detection and remediation drafting are automated. Answer collection runs on a weekly cycle rather than continuously, so this is drift detection — not live monitoring, and not a real-time alerting system.
Why track six surfaces instead of one?
Each assistant reads different sources and answers in its own way. A brand can be the top pick in one and absent in another on the same afternoon. Ten prompts across six surfaces produce sixty checks per cycle, which is what makes week-over-week comparison meaningful.
Are Google AI Overviews and AI Mode different providers?
No. They are two distinct Google Search experiences that share Googlebot, so the platform reports them separately while making clear they are not independent model providers.
What actually moves an engine answer?
Retrieval access first, then entity clarity, machine-readable structure and third-party corroboration. The audit scores crawler access, llms.txt, Schema.org and semantic structure out of one hundred, and the remediation queue fires only when your own data calls for it.
Win this surface
See what Claude tells your buyers this week
Long-context answers with hedging diagnosed, structured fixes drafted, agents reading the same numbers. Fourteen days free.