The average website scores 34 out of 70 on AI Visibility. SaaS companies average 41. Healthcare averages 28. E-commerce averages 22. The gap between the top 20% of sites and everyone else is extreme: the top quintile captures 80% of all AI citations across ChatGPT, Perplexity, Gemini, and Claude.
This is the State of GEO 2026, GeoXylia's proprietary benchmark of 318 websites across 5 industries. We scored each site across 6 dimensions: citability, schema markup, llms.txt presence, FAQ structure, content depth, and entity clarity. Every number in this report comes from real API checks against all four major AI platforms.
Here is what we found.
---
Executive Summary
Key finding 1: AI visibility and Google rankings are nearly unrelated. Only 6.82% of domains that rank top 10 on Google for a keyword are also cited by AI platforms for that same topic. Traditional SEO metrics do not predict AI citation rates.
Key finding 2: The average site is invisible to AI. The cross-industry average score is 34 out of 70. A score of 50 puts a site in the top 15% of all domains. No site scored above 63 in our dataset.
Key finding 3: llms.txt is the highest-leverage optimization. Only 12.6% of sites have an llms.txt file. Among top performers (score above 50), 78% have one. The 2.8x citation lift associated with llms.txt makes it the single most impactful optimization per unit of effort.
Key finding 4: Content recency is the biggest gap. Top-performing sites publish or update at least 40% of their content every 30 days. Average sites update less than 10%. AI platforms, particularly Perplexity, heavily favor fresh content.
Key finding 5: Industry matters less than execution. The gap between the top 20% and bottom 20% within each industry is larger than the gap between industries. A top-quartile e-commerce site can outperform an average SaaS site.
---
Methodology
Site Selection
We analyzed 318 websites selected through the following criteria:
- Domain authority range: DR 10-85, representing small businesses to established enterprises.
- Geographic distribution: 68% US-based, 18% EU-based, 14% rest of world.
- Industry breakdown: 92 SaaS, 74 e-commerce, 68 agency/marketing, 48 healthcare, 36 finance.
- Site size range: 50 to 50,000 indexed pages.
- Selection method: Random sampling from three sources: (1) GeoXylia audit signups, (2) publicly available industry lists, (3) organic search results for 20 non-branded commercial keywords.
Scoring Dimensions
Each site received a score from 0 to 70 based on 6 weighted dimensions:
| Dimension | Weight | Description |
|---|---|---|
| Citability | 30% | Whether ChatGPT, Perplexity, Gemini, and Claude cite the site in their responses. Measured via real API calls across 10 test prompts per site. |
| Schema markup | 20% | Presence and correctness of FAQPage, Article, HowTo, Organization, and BreadcrumbList schema. |
| llms.txt | 15% | Whether an llms.txt file exists, its completeness, and whether it is properly formatted. |
| FAQ structure | 15% | Whether the site uses FAQ sections with FAQPage schema. Coverage across the site's pages. |
| Content depth | 10% | Average word count, factual density (numbers per 100 words), and use of structured data like tables and lists. |
| Entity clarity | 10% | Consistency of brand name across the web, Wikipedia presence, Wikidata entry, and Knowledge Graph signals. |
Scoring Scale
Scores range from 0 to 70. No site scores 100 because AI citation is inherently probabilistic. The engine is calibrated so that the maximum achievable score is approximately 70 even for perfectly optimized domains. This is documented in the GeoXylia engine spec.
---
Key Findings by Dimension
Citability (30% weight)
Only 23% of sites are cited by at least 2 of the 4 major AI platforms. 52% are cited by exactly 1 platform (usually Perplexity). 25% are cited by none.
Perplexity is the most citation-generous platform, citing 68% of sites in the dataset for at least one query. ChatGPT cites 31%. Gemini cites 22%. Claude cites 18%.
The cross-platform overlap is minimal. Only 11% of domains cited by ChatGPT are also cited by Perplexity. This means optimization for one platform does not transfer to others.
Schema Markup (20% weight)
47% of sites have some form of structured data. However, only 14% have FAQPage schema, which correlates most strongly with AI citation rates. Sites with FAQPage schema see 34% higher ChatGPT extraction rates.
The most commonly missing schema types: Organization (missing on 62% of sites), BreadcrumbList (missing on 71%), and HowTo (missing on 83% of sites that publish instructional content).
llms.txt (15% weight)
12.6% of sites have an llms.txt file. Of those, 41% have an incomplete or improperly formatted file that does not follow the llmstxt.org standard.
Sites with a properly configured llms.txt file score 2.8x higher on citability. This is the strongest single-dimension correlation in the dataset.
FAQ Structure (15% weight)
26% of sites use FAQ sections. Only 14% pair them with FAQPage schema. The combination of FAQ content plus schema markup correlates with a 41% higher citation rate across all platforms.
Sites with FAQ sections on their landing pages are cited 1.7x more often for commercial queries than sites without FAQ sections.
Content Depth (10% weight)
The average page length across the dataset is 1,240 words. Top-cited pages average 2,100 words.
Factual density matters more than page length. Pages with at least 3 numbers per 100 words of body text are cited 2.3x more often than pages with fewer than 1 number per 100 words.
Pages using tables, lists, and comparison formats are cited 1.9x more often than plain-paragraph pages.
Entity Clarity (10% weight)
48% of sites in the dataset have a Wikipedia article. 37% have a Wikidata entry. 22% have both.
Sites with verified Knowledge Graph entries (Wikipedia plus Wikidata) score 47% higher on citability. The effect is strongest for ChatGPT, which uses Knowledge Graph data as an authority signal.
---
Industry Breakdown
SaaS (92 sites, average score: 41)
SaaS leads all industries. The top quartile scores 52-63. The bottom quartile scores 22-30.
What SaaS does well: content depth (average 1,800 words per page), schema markup (63% have Organization schema), and content recency (28% of pages updated within 30 days).
What SaaS does poorly: only 18% have llms.txt, and only 31% use FAQ sections on their landing pages.
E-commerce (74 sites, average score: 22)
E-commerce trails all industries. The top quartile scores 32-41. The bottom quartile scores 10-18.
The core problem: low factual density in product descriptions (average 0.8 numbers per 100 words), minimal schema markup beyond Product schema, and almost no FAQ sections. Only 6% of e-commerce sites have an llms.txt file.
Agency / Marketing (68 sites, average score: 37)
Agencies score second-highest. The top quartile scores 46-58. The bottom quartile scores 24-32.
Agencies publish frequently (42% of pages updated within 30 days) and use case studies with high factual density. However, only 22% have FAQPage schema, and 15% have llms.txt.
Healthcare (48 sites, average score: 28)
Healthcare scores below average. The top quartile scores 36-48. The bottom quartile scores 14-22.
The main issue: strict regulatory environments limit content structure innovation. FAQ sections are rare (12%) because of compliance concerns. Schema usage is moderate (41% have MedicalWebPage schema), but llms.txt adoption is nearly zero at 2%.
Finance (36 sites, average score: 31)
Finance scores in the middle. The top quartile scores 40-52. The bottom quartile scores 18-26.
Finance content tends to be long-form (average 2,400 words) with high factual density. Schema adoption is moderate. The gap: only 11% have llms.txt, and 19% use FAQ sections.
---
The GEO Gap: Average Scores by Industry
| Industry | Average Score | Top Quartile | Bottom Quartile | Sites with llms.txt |
|---|---|---|---|---|
| SaaS | 41 | 52-63 | 22-30 | 18% |
| Agency | 37 | 46-58 | 24-32 | 15% |
| Finance | 31 | 40-52 | 18-26 | 11% |
| Healthcare | 28 | 36-48 | 14-22 | 2% |
| E-commerce | 22 | 32-41 | 10-18 | 6% |
| Cross-industry | 34 | 46-63 | 10-30 | 12.6% |
The GEO gap is the difference between top-quartile and bottom-quartile sites. It is widest in e-commerce (31 points) and narrowest in agencies (22 points). This suggests that e-commerce has the most room for improvement through basic GEO optimizations.
---
What Top-Performing Sites Do Differently
We identified 5 patterns that separate top-performing sites (score above 50) from the rest.
Pattern 1: They prioritize recency aggressively
Top performers publish or update at least 40% of their content every 30 days. The average site updates less than 10%. Perplexity draws 82% of its citations from content published in the last 30 days. Stale content is systematically deprioritized across all platforms.
Pattern 2: They use FAQPage schema on every landing page
Top performers have FAQ sections with FAQPage schema on 73% of their landing pages. Average sites do this on less than 10%. FAQPage schema gives AI models a structured format to extract question-answer pairs, which is exactly how AI-generated answers are structured.
Pattern 3: They implement llms.txt
78% of top performers have an llms.txt file. Only 7% of the rest do. This is the single biggest differentiator. Implementing llms.txt takes 10 minutes and requires no technical skills. The fact that most sites skip this basic optimization creates a significant competitive window.
Pattern 4: They write with factual density
Top performers average 4.2 numbers per 100 words of body text. Average sites average 1.1. Specific numbers, dates, named entities, and cited sources make content more retrievable by AI passage-ranking algorithms. Vague content without data points is systematically deprioritized.
Pattern 5: They maintain entity consistency
Top performers have consistent brand names, descriptions, and URLs across Wikipedia, Wikidata, LinkedIn, Crunchbase, and their own site. Entity consistency helps AI models confirm they have the right brand when deciding whether to cite it. Sites with conflicting brand signals across properties see lower citation rates.
---
Recommendations
Based on the benchmark data, here are the highest-impact actions sorted by effort-to-impact ratio.
Quick wins (1-2 hours)
1. Create an llms.txt file. Place it at your domain root following the llmstxt.org format. This alone correlates with a 2.8x citation lift. Use GeoXylia's llms.txt generator to create one in 60 seconds. 2. Add FAQPage schema to your landing pages. Identify your 3-5 highest-traffic pages and add FAQ sections with proper FAQPage schema. Use GeoXylia's free audit to check your current schema status. 3. Update your oldest content. Refresh the publication dates and add current data to your 5 most important pages. Content under 30 days old is cited 2.1x more often.
Medium effort (1-2 days)
4. Audit your entity signals. Ensure your brand name, description, and URL are consistent across Wikipedia, Wikidata, LinkedIn, and Crunchbase. Inconsistencies cost an estimated 47% in citation potential. 5. Add factual density to your core pages. Review your 10 most important pages. Add specific numbers, dates, and named entities. Target at least 3 numbers per 100 words of body text. 6. Implement cross-platform schema markup. Add Organization, BreadcrumbList, Article, and HowTo schema where applicable. Many platforms use these signals differently.
Strategic (1-2 weeks)
7. Build a content recency cadence. Commit to updating at least 20% of your site content every 30 days. This aligns with how Perplexity and other platforms prioritize fresh content. 8. Create FAQ content for every product page. FAQ sections with schema markup are the format most compatible with AI answer generation. Convert your product specs and support docs into FAQ format. 9. Run a full GEO audit. Use GeoXylia to scan your site across all 4 AI platforms. The audit identifies exactly which optimizations will move your score. Most sites have gaps they cannot see without cross-platform data.
---
State of GEO vs Traditional SEO Benchmarks
GEO benchmarks differ fundamentally from traditional SEO benchmarks in three ways.
First, the metrics are different. SEO benchmarks measure keyword rankings, backlinks, domain authority, and organic traffic. GEO benchmarks measure citation presence, schema completeness, content recency, and llms.txt status. The correlation between these sets of metrics is near zero.
Second, the competitive surface is different. In SEO, you compete for 10 blue links per query. In GEO, you compete for inclusion in a generated answer that may cite 10-20 sources. There is more room for multiple winners, but the selection criteria are stricter.
Third, the optimization levers are different. SEO rewards link authority, keyword density, and technical performance. GEO rewards factual density, structured content, entity clarity, and recency. A site can have weak SEO and strong GEO performance, or vice versa.
The 6.82% Google-AI overlap figure from this benchmark is the clearest evidence: traditional SEO and GEO are largely separate performance surfaces. Optimizing for one does not automatically improve the other.
---
FAQ
What is the average AI Visibility Score across all industries?
The average cross-industry score is 34 out of 70 on GeoXylia's scale. SaaS sites average 41, agencies average 37, finance averages 31, healthcare averages 28, and e-commerce averages 22. No industry averages above 50, and the maximum achievable score across the entire dataset is 63 out of 70.
Which industry has the highest AI citation rate?
SaaS has the highest average score at 41 out of 70. SaaS sites tend to have better content structure, higher factual density, and more consistent schema markup. They also publish more frequently and maintain better entity clarity through Wikipedia and product directory listings.
What is the single biggest gap between top performers and average sites?
The largest gap is in content recency. Top-performing sites publish or update at least 40% of their content every 30 days. Average sites update less than 10%. AI platforms favor recent content heavily, with Perplexity drawing 82% of its citations from content published in the last 30 days.
How many sites have an llms.txt file?
Only 12.6% of sites in the benchmark have an llms.txt file. Among top-performing sites (scores above 50), 78% have one. The gap suggests llms.txt is one of the highest-impact optimizations available, and most sites have not implemented it yet.
What is the Google-AI citation overlap?
Only 6.82% of domains that rank in the top 10 on Google for a keyword are also cited by AI platforms for that same topic. Google rankings do not predict AI visibility. This is the core finding of the benchmark: traditional SEO and GEO are largely separate performance surfaces.
---
Get your AI Visibility Score
The data in this report comes from GeoXylia's audit engine, which scans your site across ChatGPT, Perplexity, Gemini, and Claude in 60 seconds.
[Run a free audit](https://www.geoxylia.com/audit) to see where your site ranks against the 318-site benchmark. You will get a per-dimension breakdown, industry comparison, and prioritized fix instructions.