Why Claude cites the brands it cites — Anthropic's selection logic

Claude is the most source-verification-obsessed of the major engines#

Anthropic trained Claude with explicit instructions to avoid inventing facts and to hedge when certainty is low. That has two side effects for brand citations:

First, Claude is slower to mention a brand by name unless the training data contained consistent, verifiable information about it. Brand inconsistency — different descriptions on different pages, different founding dates, conflicting pricing — makes Claude hedge.

Second, when Claude does cite a brand, it cites with high confidence and often includes a qualifier ("according to their public pricing" or "as described on their site"). This makes Claude citations harder to earn but more valuable when earned.

Factor 1: Consistency across your own properties#

Claude penalizes inconsistency. If your homepage says "9 AI platforms" and your pricing page says "8 platforms," Claude is likely to avoid naming a specific number. Audit every mention of key facts (number of features, pricing tiers, founding date, engines tracked) for consistency across pages.

Action: list your top 10 "brand facts" and grep every page to ensure they are stated identically. This seems tedious but Claude rewards it measurably in our scan data.

Factor 2: Schema + visible content agreement#

Claude cross-references structured data against visible content. If your Organization schema says "founded in 2024" but your About page says "founded in 2026," Claude treats the page as low-trust. Align JSON-LD facts with the rendered HTML.

Action: generate your Organization, SoftwareApplication, and FAQ schema from the same source of truth as the visible page text. Do not let schema drift from content.

Factor 3: ClaudeBot + Claude-SearchBot access#

Anthropic operates two crawlers. ClaudeBot collects for training. Claude-SearchBot supports Claude's web search capability (launched in 2024, expanded in 2025). Blocking either reduces citation likelihood for the corresponding mode.

Action: explicitly Allow both in robots.txt. BrandCited's existing robots.txt on brandcited.ai does this correctly — use it as a template.

Factor 4: Long-form content depth#

Claude handles long context exceptionally well (up to 1M tokens in its Opus tier). It will read and synthesize across your entire site if given the chance. Short, thin content is disadvantaged. Claude rewards pages with comprehensive coverage of a topic — 2,000+ word cornerstone pages beat ten 300-word blog posts for the same subject.

Action: identify the 3–5 topics most central to your category and build a cornerstone page for each (at least 2,000 words, full schema, outbound citations, internal links to related content).

What BrandCited measures specifically for Claude#

For every scan, BrandCited queries Claude via the Anthropic API with branded and category prompts at a low temperature (0.2) for consistency. We parse mentions, confidence qualifiers ("according to," "reportedly"), and co-cited sources. Claude's mention count tends to be lower than ChatGPT's but more stable week over week — ideal for trend tracking.