How to Get Cited by Claude and Gemini

Most AEO advice concentrates on two platforms: ChatGPT and Perplexity. That's where the loudest practitioner conversation is happening, and they're the highest-traffic AI platforms by most measures. The guidance isn't wrong -- it's just incomplete.

Claude and Gemini are distinct systems with distinct citation criteria. Optimizing for ChatGPT's source architecture (third-party directories) does not optimize for Gemini. Optimizing for Perplexity's live retrieval does not optimize for Claude. If your audit shows gaps specifically on Claude or Gemini, a generic fix plan built around Reddit mentions and Yelp listings is targeting the wrong layer.

Here's what the data shows about what actually drives citations on both platforms.

Gemini's Source Architecture Is Brand-First

In our ongoing tracking of platform citation behaviors (`knowledge/platform-citation-behaviors.md`, Scout, 2026-05-12), we documented the Yext 2026 AI Visibility Study -- 17.2 million citations across ChatGPT, Perplexity, and Gemini. Its most counterintuitive finding: Gemini cites brand-owned websites for 52% of its citations. That's the highest brand-site share of any major platform.

ChatGPT gets 49% of its citations from third-party directories. Perplexity favors niche expert databases. Gemini is fundamentally different -- it's the most Google-native of the four platforms we track, which means it weights structured brand signals the way Google's other systems do.

What this means in practice: a business with a strong Yelp presence, three directory listings, and a thin website will underperform on Gemini even if it's well-cited on ChatGPT. The fix that works for one platform actively misdirects you on the other.

Google Business Profile is the highest-leverage single intervention for Gemini specifically. GBP category data, service area settings, service descriptions, and review signals feed directly into Gemini's local response layer. A GBP profile with vague category selection and no services listed is leaving Gemini's primary input source half-configured.

Schema investment also has a clearer return on Gemini than on other platforms. Per our audit data in the same tracking file, schema markup and structured data show higher signal weight for Gemini than for ChatGPT -- consistent with Gemini's deeper integration with Google's entity recognition systems. The brand website's structured data is one of the inputs Gemini uses to understand what a business does and where it operates.

Gemini's Citation Position Is More Volatile Than It Looks

One clarification worth making before the Gemini optimization advice sounds too definitive: Gemini's citation behavior can shift abruptly when Anthropic or Google releases a major model update.

In our methodology analysis of the SISTRIX Citation Drift Study (`methodology-recs/2026-07-30-citation-drift-audit-validity.md`, Scout session 92, 2026-07-30) -- 82,619 prompts and 1,548,213 snapshots tracked over 17 weeks -- the dataset recorded that Gemini 3's launch in January 2026 replaced 42% of AI Overviews cited domains overnight. Not over weeks or months. Overnight.

The same SISTRIX dataset shows that brand domains anchor at a 43% consistent rate across all 17 weeks. This is the protection against model transition volatility: businesses with correct brand-level infrastructure (GBP, schema, entity consistency) land in the stable anchor group. Businesses relying on content-level signals -- recent blog posts, social signals, link velocity -- are more exposed to the volatile 57%.

The implication is that Gemini optimization is infrastructure work, not content marketing. You're building the signals that Gemini's entity system treats as stable anchors, not chasing freshness.

Claude Has a Structural Reading Pattern

Claude is not included in the Yext 17.2M citation dataset -- Anthropic doesn't surface citation source data the same way the other platforms do. What we have on Claude comes from our own audit data and the 5W Citation Source Index 2026, a dataset covering 680 million citations across platforms.

The 5W data surfaces a striking behavioral contrast: 36% of Claude's journalism and authority citations come from the past 12 months, versus 56% for ChatGPT. Claude skews conservative on recency -- it trusts established sources with longer publication histories over fresher content from newer sources. For a business trying to build Claude citation presence, this means entity establishment matters more than content velocity. A Wikidata entry and a Wikipedia mention from two years ago outperform five fresh blog posts.

Our audit tracking in `knowledge/platform-citation-behaviors.md` (Scout, 2026-05-12) documents the other key pattern: sentence-level structure is higher-signal for Claude than for any other platform we track. Claude's extraction behavior works at the paragraph level -- the first sentence of a paragraph is treated as the entity's primary claim. A page that buries the main point in sentence three, or that structures information as dense background before surfacing the actual answer, gives Claude's extraction system less to work with.

This is counterintuitive for most content writers. A well-organized piece that builds context before delivering the answer reads naturally to humans. Claude is more likely to cite a page that states the answer in sentence one and uses the paragraph to support it.

For business pages specifically: the service description, the location statement, and the "what we do" claim should each be the first sentence of their paragraph. Not after a scene-setter. Not following a qualifier. First.

The Entity Cluster Signal (It's Not Just sameAs)

One mechanism that matters for both Claude and Gemini is often described as "sameAs links." The actual mechanism is more specific, and getting the distinction wrong leads to the wrong fix.

In our analysis of schema and entity signals (`knowledge/schema-markup-effects.md`, updated session 39, 2026-06-06 -- source: Kurt Fischman, SSRN 6284518, February 2026), the research clarified: schema's value is in entity cluster density, not the sameAs property itself. A sameAs link pointing to a Wikidata Q-item, a Crunchbase profile, and a G2 listing gives AI systems three separate traversal paths to the same entity record. The citation value comes from having verified, populated entries at those URLs -- not from having the JSON-LD property pointing to them.

A business with a Wikidata entry that carries four or five populated statements (instance of, founder, official website, industry, founding date) creates a denser entity cluster than a business whose schema points to a nearly-empty Wikidata stub. The Crunchbase profile that describes what the company does and lists its founders is a traversal node. An empty stub is not.

For Claude specifically, Wikidata presence reduces hedged responses. Our audit data shows that ambiguous brands -- those without clear entity signals -- receive more "I don't have reliable information about this business" responses from Claude. Wikidata and Wikipedia presence don't guarantee citation, but they reduce the uncertainty that causes non-citation.

What a Platform-Specific Audit Surfaces

The practical problem is that aggregate AI visibility scores obscure platform-specific gaps. A business can have a solid composite score while being invisible on Gemini and producing hedged responses on Claude. The fix plans for those two gaps are different from each other and from the fix plan that would help ChatGPT performance.

A Sourcepull Signal Check breaks out scores per platform. If Gemini is the gap, the investigation starts with GBP completeness and schema quality. If Claude is the gap, it starts with entity establishment -- Wikidata, Wikipedia, and page structure. These are specific and different paths, not the same AEO checklist applied twice.

That specificity is what makes the difference between fixing your visibility and running optimization work on the wrong lever.