State of AI Search
Where Google AI Overviews, ChatGPT, Gemini and Perplexity draw their sources, and what kinds of pages they list. Observed source patterns across four AI search engines based on a sample snapshot of more than 250,000 URLs.
Data: SEMAI visibility exports, October 2026. Sample snapshot, not market share.
Executive Summary
Each AI engine draws on a different mix of websites and content. In this sample, Luna shows more documentation and institutional sources. Gemini and Perplexity show more articles and guides. Company websites appear widely across all engines, but their cited pages are not necessarily product pages.
44.4%
Google AI Overviews: articles and guides, share of recorded links
56.9%
ChatGPT 5.6 Luna: queries containing standards bodies
91.4%
Gemini: queries containing company websites, highest observed
56.4%
Perplexity: queries containing social and publishing platforms
Key Findings
- Google AI Overviews: Articles and guides account for 44.4% of recorded links. Company websites appear in 66.9% of Overview searches with captured sources. Learn how to track these with SEMAI analytics plans.
- ChatGPT 5.6 Luna: Documentation and support account for 33.5% of links. Standards bodies appear in 56.9% of matched queries and public authorities in 46.4%.
- Gemini: Company websites appear in 91.4% of matched queries, the highest observed coverage. Articles and guides account for 50.6% of links.
- Perplexity: Social and publishing platforms appear in 56.4% of matched queries, the highest among the LLMs. Articles and guides account for 49.1% of links.
How to Read the Findings
The main distinction is both where sources come from and what kind of page is listed. A company website can supply a technical guide, an article or a product page. Website type and page format should therefore be read together. These are observed source patterns, not proven engine preferences. Domain classification remains incomplete, especially for Gemini and Perplexity. Recorded links are not independently verified final-answer citations.
Cross-Engine Comparison
The LLM domain mix at a glance
Company websites are the most common source type in every LLM. Beyond that, the engines diverge: Luna leans on standards bodies and public authorities, while Gemini and Perplexity lean on news, review and social or publishing platforms.
Exhibit 1: Share of matched queries containing each website type, by LLM (%)
Note: Same queries, countries and
tracked website across the LLMs. Percentages show query coverage, not a share of all
citations. Categories overlap. Google uses a separate, broader sample and is shown on its
own page.
Source: SEMAI visibility exports, October 2026; anonymous
telecom-software sample.
Exhibit 2: Highest-reach LLM for each website type, share of matched queries
| Website type | Highest-reach LLM | Share of queries | Lead over next |
|---|---|---|---|
| Product and service companies | Gemini | 91.4% | 12.0 pts |
| Standards and professional bodies | Luna | 56.9% | 38.8 pts |
| Government and public bodies | Luna | 46.4% | 34.0 pts |
| Social and publishing platforms | Perplexity | 56.4% | 14.5 pts |
| News, reviews and market analysis | Gemini | 44.3% | 12.2 pts |
| Academic and research sites | Perplexity | 16.6% | 3.9 pts |
Note: Derived from Exhibit 1. Lead over next is the gap in percentage points to the second-highest LLM.
Google AI Overviews
Articles and guides lead the URL mix at 44.4%
68.9%
of monitored searches had an AI Overview
75.4%
of those searches had sources recorded
62.0%
of host appearances remain unclassified
Exhibit 3: Website types and content types in Google AI Overviews (%)
Note: Domain classification is
incomplete: 62.0% of host appearances remain unclassified. Content bars show selected
categories and do not total 100%. Categories overlap.
Source: SEMAI
visibility exports, October 2026. Google covers a broader business sample than the matched
LLM comparison. These figures are a sample snapshot, not market shares.
What the Data Shows
- Company sites and social platforms both appear widely; this does not establish which source type causes visibility.
- Percentages show the share of Overview searches with recorded sources that contain each recognised website type. For deeper analytics, explore SEMAI consulting services.
ChatGPT 5.6 Luna
Technical and institutional sources stand out
98.8%
of queries had sources recorded
27.8%
of host appearances unclassified
33.5%
of URLs are documentation and support paths
Exhibit 4: Website types and content types in ChatGPT 5.6 Luna (%)
Note: Types overlap. Percentages are
of all matched queries. Basis and caveats: see appendix.
Source: SEMAI
visibility exports, October 2026.
What the Data Shows
- Standards bodies appear in 56.9% of matched queries and public authorities in 46.4%. Both are more common here than in Gemini or Perplexity.
- Company websites appear in 79.4%. A company source can be a technical manual or API guide, so company-site coverage does not mean product-page coverage.
- Documentation and support paths account for 33.5% of URLs; PDF links account for 12.2%.
Gemini
Company sites and article links are widespread
99.9%
of queries had sources recorded
66.8%
of host appearances unclassified
50.6%
of recorded URLs are articles and guides
Exhibit 5: Website types and content types in Gemini (%)
Note: Types overlap. Percentages are
of all matched queries. Basis and caveats: see appendix.
Source: SEMAI
visibility exports, October 2026.
What the Data Shows
- Company sites appear in 91.4% of matched queries, the highest observed coverage. News, review and market-analysis sites appear in 44.3%; social and publishing platforms in 41.9%.
- Articles and guides make up 50.6% of recorded URLs. Documentation is 2.7%. The observed mix is more article-heavy than Luna, although the actual page text was not reviewed.
Perplexity
Social platform reach is higher than other LLMs
80.3%
of queries had sources recorded
62.1%
of host appearances unclassified
49.1%
of recorded URLs are articles and guides
Exhibit 6: Website types and content types in Perplexity (%)
Note: Types overlap. Percentages are
of all matched queries. Basis and caveats: see appendix.
Source: SEMAI
visibility exports, October 2026.
What the Data Shows
- Social and publishing platforms appear in 56.4% of matched queries, compared with 41.9% for Gemini and 11.9% for Luna. This group includes hosted publishing and code-sharing sites, not just discussion forums.
- Articles and guides account for 49.1% of recorded URLs. PDF links account for 6.9%, above Gemini at 1.3% and below Luna at 12.2%.
Appendix
Data tables and methodology
Exhibit A1: Share of matched queries containing each website type, by LLM (%)
| Website type | Luna | Gemini | Perplexity |
|---|---|---|---|
| Product and service companies | 79.4% | 91.4% | 74.1% |
| Standards and professional bodies | 56.9% | 18.1% | 12.3% |
| Government and public bodies | 46.4% | 4.3% | 12.4% |
| Social and publishing platforms | 11.9% | 41.9% | 56.4% |
| News, reviews and market analysis | 8.7% | 44.3% | 32.1% |
| Academic and research sites | 12.7% | 12.0% | 16.6% |
Note: Highest value in each row shown in blue. Percentages show query coverage, not a share of all citations. Categories overlap.
Exhibit A2: Content types and source capture, by engine (%)
| Measure | Luna | Gemini | Perplexity | |
|---|---|---|---|---|
| Sources recorded | 75.4% | 98.8% | 99.9% | 80.3% |
| Unclassified host appearances | 62.0% | 27.8% | 66.8% | 62.1% |
| Articles and guides (URLs) | 44.4% | 9.3% | 50.6% | 49.1% |
| Documentation and support (URLs) | Unknown | 33.5% | 2.7% | 3.4% |
| PDF documents (URLs) | Unknown | 12.2% | 1.3% | 6.9% |
| Unclassified URL paths | 30.9% | 40.2% | 33.1% | 31.4% |
Note: Google sources recorded is the share of Overview searches with an AI Overview. Google content types were not reported for documentation and PDF documents.
Methodology
- Sample size: The total sample is more than 250,000 URLs.
- LLM basis: Identical query, country and tracked-site combinations in an anonymous telecom-software sample.
- Processing: Hosts and URLs are deduplicated within each combination. Repeated exports are pooled; run settings are unknown.
- Classification: URL paths estimate content type. Listed sources are not independently verified answer citations.
- Scope: Google covers a broader business sample than the matched LLM comparison. No causal or market-wide claim is made.