We Validated AVS With 816 AI Citations: Score 75 Is the Threshold for Getting Recommended by AI

#TL;DR

We ran an experiment:

155 queries → sent to AI search engine → collected 816 citations
→ scanned 721 cited websites → recorded each site's AVS score
→ analyzed: do higher-AVS websites actually get cited more by AI?

Answer: Yes. But it depends on the query type.

---

#Why We Did This

We built the AI Visibility Score (AVS) — a metric that measures how discoverable your website is to AI search engines. SEO score × 0.5 + AEO score × 0.5 = AVS (0-100).

But a score without validation is just a number.

So we ran what we believe is the first empirical study of AI search citation behavior vs website characteristics.

---

#How the Experiment Worked

Step 1: Design 155 queries
  - 5 domains (tech, finance, health, business, ecommerce)
  - 5 types (informational, comparative, recommendation, howto, local)

Step 2: Submit to OpenAI web_search API
  - Force web search on every query
  - Collect all cited URLs from AI responses

Step 3: Scan every cited URL
  - UltraProbe AVS scanner (pure HTML parsing, zero AI, < 50ms)
  - Record AVS, SEO, and AEO scores

Step 4: Statistical analysis

Total cost: $0.09 (API tokens)
Scan cost: $0 (local execution)

---

#Core Findings

#Finding 1: Cited Websites Have a Median AVS of 77 (Grade B)

AVS distribution of AI-cited websites:

  A (90+):  ██                 18 (2.5%)
  B (75+):  ████████████████  413 (57.3%)  ← more than half
  C (60+):  ████████          156 (21.6%)
  D (45+):  █████             113 (15.7%)
  E/F:      █                  21 (2.9%)

Mean AVS:   72.8
Median AVS: 77

60% of cited websites scored B or above. Only 3% scored E/F.

If your website's AVS is below 75, you're at a disadvantage in AI search.

#Finding 2: Recommendation Queries Have the Highest Threshold

This is the study's most significant finding.

Query Type                    Mean AVS of Cited Sites
──────────────────────────────────────────────────────
Recommendation ("best X")    80.2  ████████████████████
Comparative ("X vs Y")       76.8  ████████████████
Informational ("what is X")  75.4  ███████████████
How-to ("how to X")          72.3  ██████████████
Local ("best X in Taipei")   60.0  ████████████

When someone asks AI "recommend the best SEO tool", the cited websites average AVS 80.

When someone asks "best ramen in Taipei", cited websites average AVS 60.

20-point gap.

What this means:

• If you sell products/tools/services → AEO optimization is critical (threshold: 80)
• If you're a local business → AEO optimization has limited impact (threshold: 60)

#Finding 3: Most Websites Have an AEO Gap

Mean scores of cited websites:
  SEO: 80.6
  AEO: 64.5
  Gap: 16 points

Most websites invest heavily in SEO but barely touch AEO. This gap is your opportunity.

#Finding 4: High-AVS Sites Get Cited Repeatedly

Website	AVS	Times Cited
moneygeek.com	90	11
clevelandclinic.org	84	21
healthline.com	80	20
techradar.com	83	19
nerdwallet.com	82	18

High-AVS websites don't just get cited once — they get cited across multiple queries. AI remembers them.

---

#What This Means for Your Business

#If you sell products or services

Your customers are asking AI: "recommend the best [your industry]." AI cites the highest-AVS sites. If that's not you, it's your competitor.

Action: Scan your website → ultralab.tw/probe → Target AVS 80+

#If you're a local business

AEO has limited impact on local queries. Keep focusing on Google Business Profile and local SEO.

#If you're an SEO professional

AEO is the new frontier. None of your clients are optimized for it. Learning now = first-mover advantage.

---

#Limitations (Honest)

1. Single AI engine (OpenAI only). Perplexity and Google AI Overview may differ.
2. No temporal stability test. Same queries may cite different sites next week.
3. English queries only. Chinese queries may show different patterns.
4. Correlation ≠ causation. Higher AVS correlates with more citations, but we can't prove raising AVS causes more citations.
5. On-page only. Backlinks, domain authority, freshness are not measured.

These are directions for the next study.

---

#Open Source

Everything from this experiment is public:

• AVS Specification: github.com/ppcvote/avs-standard
• Full dataset: 155 queries × 816 citations × 721 AVS scores
• Query bank + experiment scripts + analysis code: all in the repo
• Scanner engine: github.com/ppcvote/ultralab-scanners (MIT)
• Free scanner: ultralab.tw/probe

Replication cost: $0.09.

---

#Conclusion

AVS 75 (Grade B) is the threshold for being cited by AI search engines. Recommendation queries demand AVS 80+. Most websites' AEO scores lag far behind their SEO — this is the biggest optimization gap of the AI era.

What's your AI Visibility Score?

→ Free scan

---

Based on the first empirical study of AI search citation behavior. 155 queries, 816 citations, 721 AVS scores. Full data at github.com/ppcvote/avs-standard.