Why Claude Prefers Your Competitor's 2022 Article Over Your New Page

Most AI visibility advice treats platforms interchangeably. Add schema, get directory listings, clean your NAP. That's a reasonable foundation -- but it assumes the platforms care about the same things. They do not.

Our June 2, 2026 investigation into citation behavior across platforms (documented in `platform-citation-behaviors.md`, session 35) surfaced data from the 5W Citation Source Index 2026 -- a dataset of 680 million citations across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. The Claude-specific findings in that data are different enough from every other platform that they warrant a separate explanation.

If you have done everything the standard AEO checklist recommends and Claude still does not cite you, the answer is probably in one of three places.

Claude Has a Recency Problem -- In Reverse

Every other AI platform rewards freshness. Perplexity runs live retrieval on every query and heavily weights content published within the last 30 days. ChatGPT, in its hybrid retrieval mode, skews toward recent pages for queries with current-information intent. Most AEO guides respond to this by telling businesses to publish more frequently and update existing content.

That advice is correct for Perplexity. For Claude, it partly inverts.

Our session 35 investigation (2026-06-02) documents this through the 5W Citation Source Index 2026 findings: 36% of Claude's journalism citations are from the past 12 months, compared to 56% for ChatGPT. Claude is significantly more conservative about recency. It prefers older, more established sources -- ones that have accumulated authority through repeated citation, editorial consistency, and long track records, not through publish date.

The practical implication for a business publishing new service content to attract Claude's attention: Claude is not going to weight that content the way Perplexity would. A 2023 article in an established trade publication may outweigh your well-optimized 2026 service page in Claude's citation ranking for the same query.

This is not a criticism of fresh content. It has value for Perplexity, for Google AI Overviews, for building entity footprint. But targeting Claude specifically requires a different strategy than targeting the platforms that reward recency.

Claude Reads at the Sentence Level

The second distinct finding from session 35 concerns how Claude extracts information from a page it has retrieved.

Most AI platforms assess a page holistically -- does this page discuss the topic, does it appear authoritative, is the content aligned with the query? Claude applies an additional filter at the paragraph level. According to our session 35 citation behavior documentation, Claude's extractor wants the key answer surfaced in sentence one of a paragraph. A page with strong overall quality but poorly structured individual paragraphs may be read but not cited.

This is different from the format preferences documented for other platforms. Perplexity responds to freshness and niche expert source signals. ChatGPT responds to directory presence and Bing ranking overlap. Both will often cite a well-organized page even if individual paragraphs bury their main point. Claude, more than either, is extracting at the sentence level and appears to require that the answer be findable in the first sentence before it draws on the surrounding paragraph.

For businesses writing service pages or FAQ content with the goal of AI citation, the structural implication is specific: don't write paragraphs that build to their point. State the answer first, then support it. A paragraph that opens with context and qualifications before reaching the main claim is harder for Claude to extract from -- even if the information is accurate and complete.

This also explains something we observe in audits: businesses with clearly-written, high-quality service pages that get retrieved but not cited. The content is there. The structure is not what Claude needs.

The Authority Gap: Premium Outlets Only

The third finding concerns which external sources Claude weights when those sources are referenced or quoted in content.

Our session 35 documentation from the 5W data identifies a consistent premium outlet preference in Claude's journalism citations: the New York Times, The Atlantic, The New Yorker, and The Economist appear at significantly higher rates than other publications. This is not an accident of the dataset -- it reflects Claude's training-data weighting toward established, editorially rigorous sources.

For businesses trying to build authority signals that Claude recognizes, the implication cuts both ways.

On the positive side: earned media in premium outlets carries real weight for Claude in a way it does not for Perplexity, which weights niche expert sources and comparison articles. A single mention in a premium publication -- even a brief one -- may contribute more to Claude's authority assessment of your business than a dozen mentions in trade publications.

On the less positive side: if your PR and content program is targeting regional publications, industry newsletters, or local news, that work is valuable for other purposes but does not build the type of authority Claude is specifically looking for. The outlet threshold matters.

This is another point of divergence from ChatGPT and Perplexity, which we documented in our platform citation architecture research (Yext 2026 AI Visibility study, session 15, 17.2 million citations). ChatGPT draws 49% of its citations from third-party directories -- Yelp, BBB, G2 -- and mirrors Bing's organic results closely. Perplexity trusts niche expert databases specific to each business category. Neither platform's citation triggers look much like Claude's premium-outlet preference.

Entity Grounding Reduces Claude's Hedging

The fourth behavior is specific to brands that haven't yet established strong entity presence.

Our session 35 internal audit observations document a consistent pattern: Claude is more likely to give cautious, hedged responses for ambiguous brands -- responses that neither cite the business nor explicitly exclude it. The AI has encountered the brand name in enough contexts to know something exists, but not enough to trust the information. The result is a response that effectively omits the business from any recommendation.

The fix is not content. Entity graph presence -- specifically Wikidata entries and Wikipedia mentions -- appears to reduce this hedging pattern. Claude weights Wikipedia heavily for entity grounding, consistent with its preference for established, authoritative sources. A business that is identifiable in Wikipedia or correctly linked in Wikidata presents Claude with enough entity signal to move past the hedged non-answer.

For most small businesses, Wikipedia is not achievable at scale. Wikidata is more accessible, and the notability threshold for Wikidata entity entries is lower. The sameAs links in LocalBusiness schema, pointing to Wikidata and external directory records, build the entity graph that Claude uses to resolve ambiguity.

Why Claude Takes Longer to Respond to Fixes

All of the above compounds with a structural timing issue.

Our session 34 research (2026-06-01) documented platform ingestion estimates for structural fixes -- new directory listings, schema updates, NAP corrections. Perplexity ingests structural changes in 2-7 days because it runs live retrieval per query. ChatGPT takes 7-21 days. Claude, which draws more heavily on training data cycles than either, runs 14-45 days for structural fixes and 30-90 days for reputational signals to register.

If you have implemented Claude-targeted changes -- entity graph work, content restructured for sentence-level extraction -- set your re-audit window at 6-8 weeks, not two. Checking Claude results at two weeks will show no movement and will not tell you whether the fix was wrong or whether it just hasn't propagated.

What a Claude-Specific Gap Looks Like

In practice, a business with a Perplexity gap and a Claude gap are in structurally different situations. The Perplexity gap is often a freshness or niche-directory problem. The Claude gap is usually one of three things: low entity authority, content that buries its key answers, or absence from the source types Claude was trained to trust.

The starting point is knowing which platform gap is largest and what type of gap it is. A Signal Check at sourcepull.ca runs live queries against ChatGPT, Perplexity, Gemini, and Claude, scores each platform separately, and surfaces the specific failure pattern. The Claude score and the Perplexity score often diverge -- a strong Perplexity score does not predict Claude performance, and vice versa. The 680 million citation data makes clear these are different problems with different solutions.