← Blog·ResearchMarch 28, 2026· 12 min read

We Tested 50 Websites on AI Search — Here Are the Shocking Results

We took 50 real business websites — from SaaS startups to e-commerce stores to local services — and ran every single one through Searchiva's GEO audit engine. The results reveal a massive gap between what businesses think works for AI search and what actually gets cited.

50

Websites tested

Across 12 industries

38

Average GEO score

Out of 100

92%

Missing schema

Had zero JSON-LD

4

Scored above 70

Just 8% of sites

How We Ran This Study

We selected 50 websites across 12 industries: SaaS (10), e-commerce (8), healthcare (5), finance (5), education (4), real estate (4), legal (3), travel (3), food & beverage (3), fitness (2), B2B services (2), and media (1).

Selection criteria: all sites had existing Google traffic (verified via public data), were live and publicly accessible, and ranged from small businesses to mid-size companies. No enterprise sites with dedicated SEO teams.

Each website was audited using Searchiva's GEO engine, which analyzes 25+ AI search ranking factors including schema markup, heading structure, content extractability, factual density, meta quality, and technical signals.

📋 What we measured:

  • • GEO Score (0–100 composite)
  • • Schema markup presence & quality
  • • Heading hierarchy compliance
  • • Content depth & extractability
  • • Meta description quality
  • • AI crawler access (robots.txt)
  • • Factual density & citation-readiness

The Results: A Brutal Reality Check

The average GEO score across all 50 websites was 38 out of 100. To put that in context, our data suggests that a score below 50 means a page is unlikely to ever be cited by ChatGPT, Gemini, or Perplexity.

Score Distribution

0–20 (Critical)11 sites(22%)
21–40 (Poor)20 sites(40%)
41–60 (Average)15 sites(30%)
61–80 (Good)3 sites(6%)
81–100 (Excellent)1 sites(2%)

62% of websites scored below 40 — meaning the vast majority of businesses we tested are essentially invisible to AI search engines. Only 4 sites (8%) scored above 70.

Performance by Industry

Some industries performed significantly better than others:

IndustrySitesAvg ScoreTop Issue
SaaS1052/100Missing FAQ schema
Education448/100Poor heading hierarchy
Healthcare544/100No author authority signals
Finance541/100Blocked AI crawlers
E-commerce835/100Zero structured data
Real Estate432/100JavaScript-only rendering
Food & Bev328/100Image-heavy, text-light
Legal326/100Dense text, no structure
Others831/100Various

SaaS companies led the pack with an average score of 52 — still below the citation threshold, but significantly better than other industries. This makes sense: SaaS companies tend to have better technical infrastructure and more content-focused marketing.

E-commerce and local businesses scored worst, primarily because their pages are product-focused with minimal text content, no schema markup, and often JavaScript-rendered with no server-side alternative for crawlers.

The 5 Most Common Failures We Found

01

92% Had Zero Schema Markup

This was the single most shocking finding. 46 out of 50 websites had absolutely no JSON-LD structured data. No Article schema, no Organization schema, no FAQ schema — nothing. AI models rely heavily on schema to understand content type and relationships. Without it, your content is a black box to AI search engines.

✅ THE FIX:

Add JSON-LD Article schema to every content page. Add Organization schema to your homepage. Add FAQPage schema to any page with Q&A content.

02

76% Had Broken Heading Hierarchy

38 sites had heading hierarchy issues — skipping from H1 to H3, using multiple H1s, or having no heading structure at all. AI models parse headings to understand content boundaries. Broken hierarchy means AI can't accurately extract section-level answers.

✅ THE FIX:

Use exactly one H1 per page. Follow strict H1→H2→H3 nesting. Make headings descriptive and specific (not 'Our Services' but 'Cloud Migration Services for Healthcare').

03

68% Were Blocking AI Crawlers

34 websites had robots.txt rules that blocked GPTBot, PerplexityBot, or both — often unintentionally through overly restrictive wildcard rules. You literally cannot be cited by ChatGPT if GPTBot can't access your pages.

✅ THE FIX:

Review your robots.txt file. Explicitly allow GPTBot, PerplexityBot, and Google-Extended. Remove overly broad Disallow rules.

04

84% Had Weak Meta Descriptions

42 sites had meta descriptions that were either missing, auto-generated generic text, or over 300 characters. AI retrieval systems use meta descriptions as one of the first signals to assess page relevance. A weak meta = a missed opportunity.

✅ THE FIX:

Write unique, descriptive meta descriptions under 160 characters for every page. Include the primary topic and a clear value statement.

05

60% Had Unfocused, Wall-of-Text Content

30 sites had long, unstructured paragraphs without lists, tables, or clear section breaks. AI models extract at the paragraph level — if your key facts are buried in a 500-word block, they'll never be cited.

✅ THE FIX:

Break content into scannable sections. Use bullet points for lists, tables for comparisons, and keep paragraphs to 2-3 sentences. Put the most important fact first.

What the Top 4 Websites Did Right

The 4 websites that scored above 70 shared specific patterns that the rest didn't:

Comprehensive JSON-LD schema on every page (Article + Organization + FAQ)

Clean heading hierarchy with descriptive, keyword-rich H2/H3 headings

Answer-first content structure — key facts in the first sentence of every section

Rich internal linking between topically related pages (5+ links per page)

Visible 'Last Updated' dates with content refreshed within the last 60 days

Allowed all AI crawlers — GPTBot, PerplexityBot, Google-Extended

Used tables and bullet points for data-heavy content

Author bios with credentials and linked professional profiles

The pattern is clear: the top-performing sites weren't necessarily the ones with the best traditional SEO or the most backlinks. They were the ones that had structured, extractable, well-organized content — exactly what AI models need to generate accurate citations.

Key Takeaways

1

AI search readiness is abysmal across all industries — 92% of sites we tested are fundamentally unoptimized for AI citation.

2

Schema markup is the single biggest gap. It's also the easiest to fix — adding JSON-LD takes 30 minutes and can increase your GEO score by 15-25 points.

3

Blocking AI crawlers is shockingly common. Over two-thirds of businesses are accidentally preventing ChatGPT from ever citing them.

4

The bar is still low. Because so few businesses optimize for AI search, even basic GEO optimization puts you ahead of 90%+ of competitors.

5

Traditional SEO success doesn't guarantee AI search visibility. Several sites ranking #1-3 on Google scored below 30 on our GEO audit.

The Bottom Line

AI search is no longer a "future trend" — it's here. ChatGPT has 300M+ weekly users. Google AI Overviews are expanding rapidly. Perplexity is the fastest-growing search product in the world.

Yet 92% of the websites we tested aren't ready. That's both a warning and a massive opportunity. If you optimize now — while your competitors are still sleeping — you can capture a disproportionate share of AI-driven traffic and citations.

The optimizations aren't complex. Add schema. Fix your headings. Unblock AI crawlers. Structure content for extraction. These changes take hours, not months — and the impact is immediate.

How Does Your Website Score?

Run the same GEO audit we used in this study. Free. Takes 30 seconds.

Run Your Free Audit →