Skip to main content
WP Bulk Publishing

AI models → Claude

Claude standards

Anthropic operates the most citation-conservative of the frontier labs. Claude prefers to name sources by domain and to quote sparingly; it favours structured, well-sourced pages over rhetorically confident ones. That aligns almost perfectly with what makes a page good — which makes Claude a useful north star for programmatic publishing.
TL;DR: Claude cites pages that look like references — clear sourcing, calm prose, unambiguous claims. WBP publishes with that shape by default.

1. The ClaudeBot crawler

User-agent: ClaudeBot

Respects robots.txt. Anthropic recommends explicit Allow / Disallow rules — no default guess.

Respectful crawl rate

Well-behaved by industry standards; low request rates, respects Crawl-delay.

Retrieval vs training

Same crawler is used for both. Blocking removes you from both surfaces.

Citation surface

Claude.ai and API clients surface source citations when retrieval is enabled by the caller.

2. How Claude decides to cite

  • Prefer pages with visible primary-source citations. Claude models 'a good source cites its sources'.
  • Prefer pages that quantify claims. 'Faster' is worse than 'reduced LCP from 3.2s to 1.8s'.
  • Prefer pages that name specifics. Claude penalises vague hedging in extraction.
  • Prefer pages with clean H1/H2 hierarchy — extraction keys on it.
  • Prefer pages under 3,000 words for a given topic. Long pages get partial extraction and partial citation.

3. Traps that lose Claude citations

  • Anonymous or single-name bylines. Claude prefers a named human author with linkable credentials.
  • Pages that hedge every claim with 'may', 'might', 'could'. Extraction rewards commitment.
  • Pages full of first-person marketing ("we believe…"). Claude prefers third-person prose in the extraction path.
  • Broken schema, mismatched title/H1, canonical drift. Claude drops sources with visible integrity issues.

4. How WBP tunes for Claude

WBP's AI Tuning Score includes Claude-specific checks: quantified-claim ratio, primary-source citation count, and hedge density. Templates that regress on any of the three get flagged before the batch publishes.

The EEAT Engine surfaces the author's linkable credentials and past writing on the topic — the exact signals Claude weighs when deciding whether a byline is trustworthy.

Related standards