Skip to content
Primores AI wiki

Experiment: AI Visibility Audit on E-commerce Sites

Experiment: AI Visibility Audit on E-commerce Sites

TL;DR: Testing the AI Visibility Audit skill on two Lithuanian e-commerce sites revealed hidden WAF blocks and misconfigurations that standard SEO tools miss — demonstrating the value of UA-spoofed audits.

The Question

Can the AI Visibility Audit skill detect real issues that affect whether AI agents can access and cite e-commerce content?

Hypothesis

E-commerce sites optimized for traditional SEO may have hidden barriers to AI visibility — particularly:

  • CDN/WAF blocks on AI user-agents
  • Client-side rendering hiding content
  • Missing structured data for AI extraction

Method

Setup

  • Tool: AI Visibility Audit skill for Claude
  • Targets: pigu.lt, varle.lt (major Lithuanian e-commerce platforms)
  • Comparison: anthropic.com, example.com (baselines)

Process

  1. Run full 5-dimension audit on each domain
  2. Compare scores across dimensions
  3. Identify patterns specific to e-commerce
  4. Validate findings with manual checks

Sample Size / Duration

  • 4 domains audited
  • Multiple pages per domain (home, category, product, blog)
  • Single audit session (~15 minutes per domain)

Results

Raw Scores

DomainCrawl (25)Render (25)On-page (20)SoV (20)Authority (10)Total
pigu.lt (home)20221812779
varle.lt (blog)18231913780
anthropic.com1518105452
example.com101285338

Key Findings

pigu.lt: Hidden WAF Block

Finding: The homepage scored well (79/100), but deep pages (product, category) returned 403 Forbidden to all AI bot user-agents.

Evidence:

URL: pigu.lt/lt/p/samsung-galaxy-s24
Googlebot: 200 OK
GPTBot: 403 Forbidden
ClaudeBot: 403 Forbidden
PerplexityBot: 403 Forbidden

Implication: The site appears accessible but AI agents cannot access product information. This is invisible to:

  • Standard SEO crawlers (they use Googlebot UA)
  • robots.txt analysis (no blocks listed)
  • Manual browsing

Business impact: Products cannot be recommended by AI shopping agents.

varle.lt: llms.txt Misconfiguration

Finding: The site has an llms.txt file, but it’s configured as a redirect rather than serving content.

Evidence:

GET /llms.txt
Response: 301 → /lt/llms.txt
GET /lt/llms.txt
Response: 404 Not Found

Implication: Attempted AI optimization, but broken implementation provides no benefit.

varle.lt: Cart Login Above Content

Finding: The blog post page had “Prisijungti” (Login) and cart widget text appearing before actual article content in the HTML source.

Impact: AI extracting “first paragraph” gets login prompts instead of article content.

Both Sites: Strong JSON-LD Implementation

Positive finding: Both sites have comprehensive Product and Organization schema markup — well-positioned for AI extraction once access issues are fixed.

Analysis

What Worked

  • ✅ UA spoofing detected real WAF blocks invisible to standard tools
  • ✅ SSR/CSR detection correctly identified both sites as server-rendered
  • ✅ JSON-LD parsing found good structured data coverage
  • ✅ Scoring differentiated between sites meaningfully

What Didn’t Work

  • ❌ Share-of-voice testing limited for Lithuanian-language queries
  • ❌ Authority dimension hard to assess for local market leaders

Surprises

  • 🤔 WAF blocking AI but not Googlebot — Likely misconfigured rate limiting treating AI bots as scrapers
  • 🤔 High scores despite access issues — Homepage-only audit can miss critical deep-page problems

Conclusions

  1. Traditional SEO tools miss AI-specific blocks — UA spoofing is essential
  2. E-commerce sites face unique challenges — Product pages often have stricter security
  3. llms.txt adoption is early and error-prone — Good intention, poor execution
  4. Homepage vs. deep page divergence — Must audit multiple page types

Practical Applications

  1. For e-commerce teams: Test product pages specifically, not just homepage
  2. For security teams: Review WAF rules for AI bot user-agents — blocking them may hurt AI shopping visibility
  3. For SEO consultants: Add AI visibility audits to service offerings — finds issues other tools miss
FixImpactEffort
Whitelist AI bot UAs in WAFCriticalLow
Add llms.txt with product taxonomyMediumLow
Create FAQ schema for common questionsMediumMedium
FixImpactEffort
Fix llms.txt redirect chainMediumTrivial
Move login widget below main contentLowLow
Add FAQ schema to blog postsMediumMedium

Limitations

  • Single audit point-in-time (sites change)
  • Lithuanian market — results may differ in other regions
  • Sample of 4 domains — not statistically significant
  • Share-of-voice limited by query language support

Next Steps

  • Re-audit after recommended fixes
  • Test more e-commerce platforms (Baltic, EU)
  • Compare against AI visibility SaaS tools (Semrush, Peec)
  • Develop e-commerce-specific scoring weights

Key Takeaways

  • WAF/CDN blocks on AI bots are common and invisible to standard tools
  • E-commerce product pages need specific AI visibility attention
  • llms.txt adoption is early — expect misconfigurations
  • The AI Visibility Audit skill successfully identified real, actionable issues

Experiment conducted: 2026-04-15