RanketAI Guide #02: How ChatGPT, Claude & Gemini Each Decide Which Brands to Cite
ChatGPT, Claude, and Gemini use different crawlers, training data, and citation criteria. Why does the same brand appear in one LLM but not another — and how to optimize for all three simultaneously with AEO strategy.
This blog content may use AI tools for drafting and structuring, and is published after editorial review by the RanketAI Editorial Team.
TL;DR: ChatGPT, Claude, and Gemini cite different brands even for identical queries — because their crawlers, training data pipelines, and citation criteria differ fundamentally. RanketAI Guide #02 dissects each LLM's citation algorithm and outlines an AEO optimization strategy for getting cited across multiple LLMs simultaneously.
Why does the same brand appear in ChatGPT but not Claude?
It's a question marketing teams ask constantly: "Our brand shows up consistently in ChatGPT, but our competitor gets mentioned more often in Claude."
This isn't random. ChatGPT, Claude, and Gemini have completely different citation systems.
| Attribute | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Primary crawler | GPTBot + Bing (search) | ClaudeBot | Googlebot |
| Real-time search | ChatGPT Search (Bing-connected) | Claude Web Search (selective) | AI Overview (Google-connected) |
| Training data weight | High | Medium | Low (real-time SERP prioritized) |
| Core citation criterion | · direct-answer paragraphs | Verifiability · source attribution | E-E-A-T · SERP relevance |
| Citation display | Footnote links (Search mode) | Inline source attribution | AI Overview source cards |
Understanding these differences is the starting point for a multi-LLM visibility strategy.
How does ChatGPT decide which brands to cite?
GPTBot: what does it collect?
GPTBot is OpenAI's training data collection crawler. Adding User-agent: GPTBot Disallow: / to your robots.txt blocks its access.
Content characteristics GPTBot prioritizes:
- Direct-answer paragraphs: paragraphs that provide a clear answer to a question
- Data- and statistics-rich content: claims backed by specific numbers and sources
- Authoritative outbound links: links to verified institutions or media
- Well-structured text: content with clear tables, lists, and headings
ChatGPT Search: real-time Bing integration
ChatGPT Search mode references Bing search results in real time. In this mode, Bing search ranking and web content quality matter more than the existing training data corpus.
Core ChatGPT citation optimization:
- Add natural-language question-format headings to FAQ sections
- Cite specific numbers and sources with each claim
- Register in Bing Webmaster Tools and allow crawling
- Provide a direct-answer paragraph (answer within first 200 characters of the page)
ChatGPT does not cite equally across topics
Citation behavior does not only differ between LLMs. It varies sharply by topic inside ChatGPT itself. In Similarweb's 2026 Generative AI Landscape report (May 2026, U.S. desktop) as compiled by Search Engine Journal, how often ChatGPT included links in its answers differed by more than fourfold depending on the subject.
| Topic | Share of ChatGPT answers containing citations |
|---|---|
| Travel | 22.6% |
| Retail | 13.5% |
| Sports | 10.7% |
| Finance | 8.0% |
| Overall average | 6.8% |
| Education | 4.8% |
Travel runs more than three times the average while education sits below it. The gap comes not from content quality in those verticals but from the nature of the questions. The more a topic depends on real-time or local specifics, the more the model reaches for external sources; the more it can be answered from settled knowledge, the more it answers from training data alone.
That splits practice two ways.
- In low-citation topics, accumulating brand mentions that persist in training data is more realistic than waiting for footnote links. In fields like education or technology, link visibility is a poor primary success metric.
- In high-citation topics, making content citable pays back quickly. Where links attach often — travel, retail — direct-answer paragraphs and structured data convert into visibility much faster.
The trend matters too. In the same report, ChatGPT's citation rate rose more than fivefold, from roughly 1.3% in June 2025 to 6.8% in May 2026 — but the climb was uneven, dipping to about 4.5% in February 2026. Citations are trending up without increasing monotonically, so read the trajectory through repeated measurement rather than concluding from a single snapshot.
Why does Claude prioritize verifiability as its citation criterion?
ClaudeBot: Anthropic's approach
ClaudeBot is Anthropic's training data collection crawler. Claude's citation philosophy stems from Anthropic's Constitutional AI principles.
Constitutional AI is designed to ensure Claude provides accurate, verifiable information — which directly influences its citation patterns.
Content characteristics Claude particularly prefers:
- Source attribution: content with clear author name, publishing organization, and date
- Verifiable claims: assertions that can be traced back to source links or primary data
- Balanced presentation: content that presents both strengths and weaknesses
- Uncertainty acknowledgment: expressions like "estimated to be," "according to," that acknowledge uncertainty
Where ChatGPT prioritizes authority, Claude prioritizes verifiability.
Core Claude citation optimization
- Specify author information as structured data:
authorName,updatedAt, etc. - Link each major claim to a primary source via
claimSourceMap - Prefer "according to Study X (source link)" over "our research shows"
- Data-driven prose over superlative claims
What criteria does Gemini use to cite content?
Google AI Overview: where SEO meets GEO
Gemini (especially Google AI Overview) is the closest of the three LLMs to traditional SEO. Google's existing Googlebot crawl data and E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) signals serve as AI Overview's citation criteria.
One important nuance: SERP #1 pages are frequently not cited in AI Overview. Conversely, pages ranking #5–15 in SERP often get cited in AI Overview. The reason: AI Overview prioritizes answer relevance over ranking.
Content Gemini prioritizes for citation
- FAQPage schema: Google's AI parses structured data directly
- HowTo schema: step-by-step guides appear frequently in AI Overview
- dateModified metadata: content with freshness signals gets preferential treatment
- Brand entity registration: having a brand entity in the Google Knowledge Graph is an advantage
Core Gemini citation optimization:
- Apply FAQPage, HowTo, and Article schemas
- Register brand information in the Knowledge Graph
- Monitor crawling status in Google Search Console
- Keep
dateModifiedcurrent at all times
The 5 AEO signals that capture all three LLMs simultaneously
RanketAI measures 5 AEO signals that ChatGPT, Claude, and Gemini evaluate in common.
AEO Signal 1: Question-format heading ratio (core signal for all three LLMs)
The proportion of H2/H3 headings phrased as questions ("What is X?", "How do you do Y?"). All three LLMs use question-format headings as anchor points when extracting answers. This is the signal with the highest overall impact in the RanketAI measurement model.
Benchmark: 40%+ of H2/H3 headings as questions
AEO Signal 2: Direct-answer paragraphs (high-impact signal for ChatGPT & Claude)
Paragraphs within 50–200 characters after a heading that provide the core answer to the heading's question. Both ChatGPT and Claude use these "direct-answer paragraphs" as the basis for citation excerpts.
Benchmark: Direct-answer paragraph present in 70%+ of major H2 sections
AEO Signal 3: FAQ coverage (high-impact signal for Gemini)
Measures whether questions related to the page's topic are covered in a FAQ section. FAQPage schema markup provides a particular advantage for Gemini.
Benchmark: 8+ FAQ items with FAQPage schema applied
AEO Signal 4: Citation signal density (high-impact signal for Claude)
Density of external authority links, claims containing statistics, numbers, and dates, and author information throughout the page. Weighted especially heavily in Claude's evaluation.
Benchmark: 1+ external citation per 1,000 characters; author information present
AEO Signal 5: AI crawler accessibility (prerequisite signal)
Whether GPTBot, ClaudeBot, and Googlebot are permitted access in robots.txt. Blocking crawlers means even the best content will never be reflected in LLMs. This is a prerequisite to verify before any of the other four signals.
Benchmark: All three crawlers confirmed as permitted
How large are the citation pattern differences between LLMs?
Patterns observed in RanketAI dashboard aggregated data (50 brand domains, Jan–Mar 2026):
| Characteristic | ChatGPT-cited brands | Claude-cited brands | Gemini-cited brands |
|---|---|---|---|
| FAQPage schema applied | 61% | 58% | 89% |
| Author information present | 54% | 91% | 63% |
| Direct-answer paragraph present | 87% | 79% | 71% |
| External authority links included | 72% | 88% | 69% |
| dateModified kept current | 58% | 62% | 84% |
Common traits of brands cited across all three LLMs:
- FAQPage schema + direct-answer paragraphs + author information + external citations — present simultaneously
- llms.txt adoption rate 3.2x higher than non-cited brands
How does RanketAI assign grades?
RanketAI combines the 5 AEO signals above with crawlability, structured data, page speed, and direct LLM citation verification to produce a grade-based diagnosis. Grade meanings break down as follows:
- Top tier — regularly cited across major LLMs; maintain and monitor.
- Upper-mid tier — cited in multiple LLMs with room to optimize a remaining LLM; focus on the weak LLM.
- Lower-mid tier — cited in one specific LLM only; AEO foundation needed.
- Lower tier — low citation frequency; prioritize FAQPage, schema, and crawler access basics.
Key action summary
| Priority | Action item | LLMs affected |
|---|---|---|
| 1 | Confirm GPTBot, ClaudeBot, and Googlebot are allowed in robots.txt | All |
| 2 | Add FAQPage schema + 8+ question-format H2 headings | Gemini, ChatGPT |
| 3 | Source link + author name for each major claim | Claude, Gemini |
| 4 | Direct-answer paragraphs (50–200 character core answer after each heading) | ChatGPT, Claude |
| 5 | Write and deploy an llms.txt file | All |
| 6 | Keep dateModified current + register in Knowledge Graph | Gemini |
FAQ
Q. Why do ChatGPT and Claude cite the same page differently?▾
Their training data and citation criteria differ. ChatGPT prioritizes the clarity of direct-answer paragraphs; Claude prioritizes source attribution and verifiability. Even for identical content, which signals you strengthen determines citation frequency per LLM.
Q. What happens if I block AI crawlers in robots.txt?▾
ChatGPT's and Claude's training data crawlers honor robots.txt directives. Blocking them may exclude you from their training data. That said, ChatGPT Search's Bing integration is a separate channel, so complete blocking is not possible.
Q. Do I need strong SEO to appear in Gemini AI Overview?▾
SEO helps but is not sufficient. Even SERP #1 pages are frequently absent from AI Overview. AI Overview weighs answer relevance — FAQPage schema, direct-answer paragraphs, freshness — more than SERP rank.
Q. Does llms.txt actually increase citations?▾
The isolated effect of llms.txt is still debated. However, it explicitly guides AI crawlers to understand your site's structure, and positive effects are observed when combined with structured data and FAQ sections.
Q. Which gets cited more in AI — content in Korean or English?▾
Currently, ChatGPT, Claude, and Gemini all show higher citation frequency for English content. That said, Korean queries return Korean content citations. If global brand visibility is the goal, running AEO optimization in parallel for English content is recommended.
Q. How is the RanketAI score different from an SEO score?▾
SEO scores measure signals that determine Google search rankings (backlinks, keywords, domain authority). RanketAI measures signals that determine citation likelihood in AI answers (AEO structure, AI crawler accessibility, direct LLM citation verification). High-SEO/low-RanketAI and low-SEO/high-RanketAI brands both exist.
Q. What does the next guide in this series cover?▾
Guide #03 will explain why non-English content tends to have lower AI visibility and outline improvement directions.
Q. How long does it take to see AEO improvement results?▾
It depends on AI crawler re-indexing cycles. Generally, changes from adding FAQPage schema and direct-answer paragraphs are observed within 1–3 months. Gemini AI Overview syncs with Googlebot crawl cycles and tends to reflect changes relatively quickly.
Further reading
- RanketAI Guide #01: Why SEO Alone Isn't Enough in the AI Search Era
- GPT-5 vs Claude vs Gemini: AI Model Comparison as of March 2026
- Can AI Explain What Your Company Does? The 6-Axis Entity Footprint Audit
Update notes
- First published: 2026-03-24
- Last updated: 2026-07-28 — added ChatGPT citation rate variance by topic and its time series; refreshed OpenAI and Anthropic crawler documentation URLs
- Data basis: RanketAI observations (Jan–Mar 2026); crawler official documentation (as of July 2026); Similarweb 2026 Generative AI Landscape (May 2026, U.S. desktop)
- Next update: When LLM crawler policies change or citation rate trends are refreshed
References
- OpenAI: Overview of OpenAI Crawlers
- Anthropic: ClaudeBot Crawling Policy
- Google: AI Overviews and Search
- BrightEdge: AI Search Ranking Factors 2026
- Search Engine Journal: ChatGPT Links Out Most On Travel Questions
Related Guides
Execution Summary
| Item | Practical guideline |
|---|---|
| Core topic | RanketAI Guide #02: How ChatGPT, Claude & Gemini Each Decide Which Brands to Cite |
| Best fit | Prioritize for AI Business, Funding & Market workflows |
| Primary action | Define a measurable success KPI (cost, time, or quality) before starting any AI initiative |
| Risk check | Validate ROI assumptions with a small pilot before committing the full budget |
| Next step | Establish a quarterly review cadence to track KPI movement and adjust scope |
Data Basis
- Citation rate variance by topic: Similarweb's "2026 Generative AI Landscape" report (May 2026, U.S. desktop), as reported by Search Engine Journal — ChatGPT included citations in 22.6% of travel answers, 13.5% retail, 10.7% sports, 8.0% finance, against a 6.8% overall average and 4.8% for education. The time series runs from roughly 1.3% in June 2025 to 6.8% in May 2026, with a dip to about 4.5% in February 2026. The underlying report sits behind a download gate, so figures are cited from the secondary report.
- Cross-verified against LLM crawler analysis reports by Search Engine Journal, Moz, and Ahrefs (2025–2026). Based on technical documentation and research papers covering GPTBot, ClaudeBot, and Googlebot crawl behavior.
- RanketAI dashboard aggregated data: ChatGPT, Claude, and Gemini citation pattern observation across 50 brand domains (Jan–Mar 2026, repeated queries by domain category).
- AEO signal weighting: based on the RanketAI internal measurement model. Cross-verified with external benchmarks (BrightEdge AI Search Report 2026, Conductor GEO Study 2025).
Key Claims and Sources
This section maps key claims to their supporting sources one by one for fast verification. Review each claim together with its original reference link below.
Claim:How often ChatGPT includes citations varies sharply by topic — 22.6% for travel, 13.5% for retail and 8.0% for finance, against a 6.8% overall average and 4.8% for education
Source:Similarweb 2026 Generative AI Landscape / Search Engine Journal (May 2026, U.S. desktop)Claim:ChatGPT citation rates rose more than fivefold from roughly 1.3% in June 2025 to 6.8% in May 2026, but the climb was uneven, dipping to about 4.5% in February 2026
Source:Similarweb 2026 Generative AI Landscape / Search Engine Journal (May 2026)Claim:ChatGPT combines GPTBot crawl data with training data; ChatGPT Search leverages the Bing index in real time
Source:OpenAI: Overview of OpenAI CrawlersClaim:Claude uses the latest web data collected by ClaudeBot and, following Constitutional AI principles, prefers content with verifiable sources
Source:Anthropic ClaudeBot Crawling PolicyClaim:Google Gemini and AI Overview cite content based on Googlebot-sourced SERP data, with E-E-A-T signals as the core citation selection criterion
Source:Google: AI Overviews and Search
External References
The links below are original sources directly used for the claims and numbers in this post. Checking source context reduces interpretation gaps and speeds up re-validation.
Is your site visible in AI search?
See for free how ChatGPT, Perplexity, and Gemini describe your brand.
Start Free Diagnosis →Related Posts
These related posts are selected to help validate the same decision criteria in different contexts. Read them in order below to broaden comparison perspectives.
RanketAI Guide #08: How LLMs Build Answers — 4 Stages Where Your Brand Surfaces
ChatGPT, Perplexity, and Gemini compose answers through the same 4-stage pipeline — understanding, retrieval, grounding, synthesis. This guide maps where your brand surfaces at each stage and why a single prompt cannot measure AI visibility.
The Korean AI Search Landscape 2026 — Where Naver AI Briefing Meets Global LLMs
Korea runs two overlapping AI search ecosystems: Naver AI Briefing (30M monthly users, now carrying ads) and ChatGPT (23.45M Korean app users). One page on the market structure, the July 2026 double monetization, and the measurement asymmetry.
AI Agents Abandon Pages They Can't Read — How Access Errors Hand Over Your Brand Narrative
Research shows that when an AI agent hits an access error or unreadable pricing, it abandons the brand page and pulls numbers from outside sources. An agent bounce costs the right to state your own facts — here is what to inspect on your site.
Rankings Are Not Citations — Bing's Citation Share and Web IQ Make It Official
In 2026 Microsoft shipped the first first-party AI citation report from a major platform, then added Citation Share — a per-query citation percentage — and agent-native Web IQ. Rankings and AI citations are now officially different KPIs.
What Naver AI Briefing Cites — the 'Citation Economy' and Brand Visibility in Korea
Naver is investing 1 trillion won over five years in content and pays creators by AI Briefing citation frequency through 'Naver Mate.' With up to 70% of AI Briefing answers drawn from UGC, here is what gets cited and how Korean brands show up in AI answers.