This page records SeaSight GEO's ongoing observations of major AI search products, citation mechanisms, and B2B procurement query performance. We use a fixed query set (covering Chinese and English B2B procurement, supplier comparison, and technical evaluation scenarios), executing the same queries across multiple AI search engines weekly and comparing output changes. We only record verifiable public changes, product-visible changes, and stable trends observed in authorized samples — single anomalies are not treated as conclusions.
Recently Observed AI Search Changes
AI Summaries Increasingly Prefer Structured, Well-Sectioned Explanatory Pages
First Observed: March 2025 | Reconfirmed: April 2026. On the same topic, pages with clear definitions, use cases, steps, FAQs, and Schema markup are more readily understood by AI summaries as citable material. Samples since Q1 2026 show: B2B content pages with FAQPage Schema and HowTo Schema have approximately 3× higher probability of appearing in AIO summaries for "how to choose [product]" queries compared to pages without Schema. For B2B independent sites, guide pages, glossary pages, and purchasing decision pages beyond product pages are becoming more GEO-valuable than pure product catalog pages.
Google AIO Significantly Expands into B2B Procurement Queries
November 2025 – May 2026. In the first half of 2025, Google AI Overviews primarily covered consumer-level information and simple factual queries. But since November 2025, we've observed AIO consistently appearing in commercial procurement queries like "industrial [product] supplier comparison" and "B2B [service] pricing guide 2026." A batch test in March 2026 showed AIO appearance rates for 120 Chinese and English B2B long-tail queries rising from ~12% in Q2 2025 to ~41% in Q1 2026. More critically: when AIO appears, the estimated CTR for the #1 traditional blue link drops 30%–55%. This means B2B companies that don't optimize for AIO visibility could lose most of their traffic even ranking #1 in SEO.
Citation Chain Increasingly Values Title, Summary, and Body Text Consistency
First Observed: May 2025 | Updated: January 2026. When a page title promises "guide/definition/comparison" but the body lacks directly answerable paragraphs, Copilot is more likely to skip that page and cite third-party encyclopedias, Q&A sites, or directory pages. January 2026 comparison testing further revealed: Copilot's citation rate for pages with explicit numerical promises in titles (e.g., "5 Key Factors," "Top 10 Suppliers") improved by ~28%, provided the body actually delivers a list or structured answer matching the title exactly. B2B websites need to write procurement questions, technical parameters, and supply capabilities as independently citable answer blocks, rather than relying solely on overall page ranking.
Copilot Shifts from Encyclopedia Sources to Primary Business Sources
September 2025 – March 2026. In H1 2025, Wikipedia and industry encyclopedia sources accounted for ~35% of Bing Copilot's B2B query citations. But since Q4 2025, we've observed a clear shift in citation preferences: brand websites, industry association white papers, and .edu domains rose to ~52% in Q1 2026. This means Copilot's authority assessment of information sources is shifting from "public encyclopedia" to "primary business sources." Key takeaway for B2B companies: building industry white papers, technical documentation, and case study pages is more effective for earning Copilot citations than relying solely on third-party directory listings.
Source Diversity Affects Answer Credibility — 2026 New Data
Tracked Since April 2025 | Updated: May 2026. Perplexity's citation results typically mix official websites, media, encyclopedias, forums, and databases. Our Q1 2026 analysis of 80 B2B queries found: Perplexity averages 6.2 sources per answer (up from 4.8 in Q2 2025), and when answers contain 3+ different source types (e.g., official website + industry media + academic/GitHub), user trust ratings are higher. Brands relying solely on website content without external authoritative sources not only have low citation probability in Perplexity but are also labeled "limited information." This is why SeaSight GEO emphasizes entity consistency, external citations, and multi-platform account verification.
Perplexity Begins Preferring Geographically Diverse Source Combinations
December 2025 – April 2026. Latest observations indicate Perplexity increasingly prefers simultaneously citing sources from different countries/regions when handling cross-border B2B procurement queries (e.g., "European [product] manufacturers vs Asian suppliers"). In an April 2026 test round, when queries involved cross-border supplier comparison, ~73% of Perplexity answers contained sources from at least two different country domains. For cross-border B2B independent sites, this means not only English content is needed, but also localized content for target markets (e.g., .de, .jp domains or corresponding language pages) to improve geographic coverage citation rates in Perplexity.
ChatGPT Search Browsing Mode and Real-Time Search Citation Logic Divergence
October 2025 – May 2026. Since its launch in late 2024, ChatGPT Search's citation behavior has undergone multiple iterations. From Q4 2025, we've observed a clear divergence in citation logic between ChatGPT Search's "browsing mode" (manually triggered webpage reading) and "auto-search mode": auto-search mode prefers summary snippets from high-DA domains with shallow citation depth (avg 1–2 sources); while browsing mode has ChatGPT deeply reading all visible text from specified URLs and generating more detailed, structured answers. A key change in March 2026: ChatGPT Search began more frequently annotating answers with "Information source: [website name]" and displaying multiple cited sources ranked by authority — increasingly converging with Google AIO's citation display logic. B2B strategy: ensure core pages render complete text content without JavaScript, as ChatGPT's browser may not execute complex JavaScript.
Citation Mechanism Comparison of Emerging AI Search Engines
December 2025 – June 2026. From late 2025 to early 2026, DeepSeek, Grok (xAI), and Claude (Anthropic) successively strengthened their web search capabilities. Our preliminary comparison test (May 2026, 50 Chinese and English B2B queries) showed: DeepSeek has significantly better citation quality for Chinese B2B queries than English, preferring Zhihu, CSDN, and brand website Chinese content; Grok excels at real-time information aggregation but lacks industry depth in citation sources, favoring news media over technical documentation; Claude's search mode is closer to a "research assistant" — fewer citation sources but deeper analysis, suitable for complex technical evaluation queries. A noteworthy finding: all three search engines rely heavily on Baidu Baike and 1688/Alibaba product pages when citing Chinese B2B content, meaning independent site brands need more systematic Chinese knowledge graph construction to gain visibility on these platforms.
Measurable Impact of Schema Structured Markup on AI Citation Rates
June 2025 – May 2026 (12-month controlled experiment). This is the core topic we've invested the most resources in tracking. In June 2025, we selected 30 B2B independent site product guide pages — 15 deployed with complete FAQPage, HowTo, Product, and Organization Schema, and 15 with only basic WebPage Schema. After 12 months, comparison results showed: pages in the complete Schema group had significantly higher combined citation counts in Google AIO and Perplexity (median 47 citations/page) versus the basic Schema group (median 12 citations/page). Particularly noteworthy is the effect of FAQPage Schema — pages with FAQPage markup had approximately 4× higher probability of appearing as "People also ask" extended sources in AIO. A new Q1 2026 finding: Bing Copilot has begun recognizing and prioritizing display of technical content with Citation and ScholarlyArticle Schema — an important signal for B2B companies publishing white papers and industry reports.
Systematic Differences in AI Search Citation Behavior Across Multilingual Environments
August 2025 – May 2026. We systematically compared AI search results for the same set of B2B queries in English, Chinese, Japanese, German, and Spanish (40 queries per language, executed monthly). Three key findings: First, significant language bias. English queries received on average 38% more citation sources than Chinese queries, with higher website authority scores for English sources. Second, insufficient cross-language citation. A B2B website with only English content is nearly invisible in Chinese AI search queries — even if the brand is a global industry leader. Perplexity and ChatGPT Search currently do minimal cross-language content citation. Third, localized Schema is the breakthrough. In March 2026 testing, after adding hreflang tags and corresponding language Organization Schema to the same English content pages, the brand's visibility in German AI search went from 0 to appearing in 12% of queries. For cross-border B2B independent sites, creating core landing pages in at least target market languages with localized Schema is the minimum threshold for multilingual AI SEO.
How Do We Determine Whether a Change Is Worth Recording?
Our observation methodology follows three principles: verifiable, comparable, and strategically meaningful. Below is a detailed explanation of our judgment criteria.
We maintain a fixed B2B query corpus containing approximately 200 Chinese and English queries, covering industrial equipment procurement, SaaS selection, logistics provider comparison, and technical specification queries. Each query is executed at least weekly across Google AI Overviews, Bing Copilot, Perplexity, ChatGPT Search, DeepSeek, and Grok, recording AI summary text, citation URLs, source ordering, and display format changes. Query results are archived in a structured database supporting cross-temporal comparison and statistical significance testing.
All observations are performed using incognito browser sessions from four geographic nodes (US, Singapore, Germany, Japan) to eliminate interference from login status, search history, and regional personalization. For each AI search platform, we additionally record response time, citation source domain type (official website / media / encyclopedia / forum / academic / government / other), and answer length, used to analyze cross-platform behavioral differences.
| Judgment Dimension | Recording Criteria | Non-Recording Cases |
|---|---|---|
| Verifiability | Public pages, product UI changes, or multi-round authorized samples with repeatable observation; at least two independent query dates confirming the same behavior | Single chat results, unreproducible screenshots, quickly rolled-back transient changes |
| B2B Relevance | Affects supplier discovery, product comparison, purchasing decisions, brand citations, or technical evaluation queries | Pure entertainment, general consumer content, knowledge Q&A without commercial intent |
| Strategic Impact | Changes B2B website content structure, Schema deployment, link building strategy, or AI visibility monitoring methods | UI tweaks or announcements with no apparent impact on GEO practice |
| Statistical Significance | Same trend observed across 20+ queries, with measurable control group differences | Isolated query anomalies or singular phenomena without control groups |
| Cross-Platform Consistency | When 2+ AI search engines show similar behavioral changes, marked as industry trend | Isolated changes on a single platform without cross-platform validation |
Related Reading
- Google AI Overviews Changelog & In-Depth Analysis
- Bing Copilot Citation Patterns & Monitoring Report
- Perplexity Citation Preference Changes Tracking
- ChatGPT Search B2B Visibility Research
- SeaSight GEO Original Data Methodology & Protocol
Need to Apply These Changes to Your Website?
Start with an AI search visibility diagnostic — find out where your brand is mentioned across AI search engines, who is citing you, and how you compare to competitors.
Request Free Diagnostic →