GEO Tracking Experiment: 30 Days Across 4 AI Engines
GEO tracking is the practice of monitoring your brand across AI engines. I ran a 30-day experiment tracking sanjayshankar.me across ChatGPT, Claude, Gemini, and Perplexity. The results changed how I think about generative engine optimization.
Here’s the problem every SEO professional faces right now: we know AI search is growing. We have read the reports. ChatGPT alone handles 10x the queries it did a year ago, and Perplexity has become a default research tool for technical buyers.
But nobody actually knows what drives AI citation. Does ChatGPT prefer recent press? Does Claude care about backlinks? Does Gemini prioritize structured data?
I set up visibility.so’s AI visibility tracking on my own site. 60 prompts across 4 AI engines, running for 30 days. Here is what I found.
The GEO Tracking Experiment Setup
| Parameter | Value |
|---|---|
| Duration | 30 days (May 15 – June 14, 2026) |
| Target | sanjayshankar.me and visibility.so |
| AI Engines | ChatGPT, Claude, Gemini, Perplexity |
| Prompts | 60 total (15 per engine) |
| Prompt types | “Best SEO platform for agencies”, “AI SEO tools 2026”, “how to automate SEO audits” |
| Tool used | visibility.so AI Visibility monitoring |
The prompts were designed to be questions a real buyer would ask – not keyword-stuffed queries. Things like “What’s the best way to automate SEO audits?” and “Which SEO platform actually helps agencies scale?”
How Each AI Engine Cites Sources in GEO
This is where things get interesting. Each AI model has a fundamentally different citation behavior. Understanding these differences is the core of generative engine optimization.
ChatGPT
Source Preference: Recent press and high-authority domains
Citation Consistency: Moderate – varies by conversation context
Response Stability: Changes noticeably week-to-week
ChatGPT showed a strong bias toward sources published in the last 30-60 days. Older but more authoritative content was ignored in favor of newer pieces, even from lower-authority domains. This aligns with how ChatGPT’s training data cutoff works – recent fine-tuning data carries disproportionate weight.
What worked: Publishing a detailed guide and sharing it on X within the same week. ChatGPT picked it up faster than any other engine.
Claude
Source Preference: Long-form authoritative content
Citation Consistency: High – same prompts produced similar results
Response Stability: Most stable across the 30 days
Claude was the most consistent engine by far. It favored in-depth guides (2,000+ words) with clear structure, proper heading hierarchy, and factual depth. It was also the most likely to cite the same source repeatedly across different conversations.
What worked: Detailed technical posts with comprehensive sections. Claude cited my Agentic SEO guide in 12 of 15 prompts.
Gemini
Source Preference: Structured data and Google-verified entities
Citation Consistency: Low – highly variable response
Response Stability: Least stable – changed frequently
Gemini was the most unpredictable. Its responses varied significantly from day to day, and it showed a strong preference for sites with well-structured data markup. If your site has proper JSON-LD, FAQ schema, and Organization markup, Gemini notices.
What worked: Adding FAQ schema to key pages. Visibility in Gemini was directly correlated with structured data completeness.
Perplexity
Source Preference: Citation count and source diversity
Citation Consistency: High – same sources cited repeatedly
Response Stability: Moderate – updates when new sources appear
Perplexity behaves differently from the other three. It explicitly cites sources and links back to them. It also heavily weights how many other sources reference the same information. If three articles say the same thing, Perplexity will cite all three.
What worked: Getting cited by at least 2-3 other relevant sources on the same topic. Perplexity’s algorithm appears to treat multi-source confirmation as a quality signal.
5 Essential GEO Tracking Patterns I Did Not Expect
1. AI Engines Have “Favorite” Content Formats
| Engine | Best Performing Content | Worst Performing Content |
|---|---|---|
| ChatGPT | Recent tutorials (under 30 days old) | Old but authoritative guides |
| Claude | Long-form deep dives (2k+ words, structured) | Short listicles |
| Gemini | Structured, schema-rich pages | JavaScript-heavy, slow pages |
| Perplexity | Multi-cited, well-referenced content | Single-source claims |
Takeaway for brand tracking AI search: There is no single “optimize for AI” strategy. You need to optimize differently per engine.
2. Consistency Is Engine-Specific
Claude and Perplexity showed high consistency. The same prompt produced the same citation pattern 80%+ of the time. ChatGPT and Gemini varied significantly.
This means if you are not appearing in Gemini today, you might appear tomorrow. And vice versa. Do not over-optimize based on a single day’s check.
3. Brand Recall Matters More Than Keywords
The most interesting finding from this generative engine optimization experiment: when prompts included brand-specific language (“visibility.so” vs “SEO platform”), engines that recalled the brand from training data cited it even without real-time web lookup.
ChatGPT cited visibility.so in 3 of 15 prompts even when the word “visibility” was not in the prompt. It associated the brand with the category.
This is the closest thing to traditional brand building in the AI era.
4. Content Freshness Has a Half-Life
For ChatGPT, content published in the last 7 days had a 40% higher GEO tracking citation rate than content published 30+ days ago. After 60 days, the advantage dropped to near zero.
The shelf life of SEO content is shrinking. What worked last month needs to be refreshed this month. For SEO teams, this means recurring content refreshes are not optional. They are table stakes for generative engine optimization.
5. Cross-Engine Visibility Creates a Feedback Loop
When the same brand was cited by 2+ engines, the likelihood of being cited by a third engine increased. There appears to be a “multi-engine visibility threshold.” Once you are visible in 2 engines, the third and fourth follow faster.
What This Means for Your GEO Strategy
Based on this GEO tracking experiment, here is what I would recommend for any team serious about AI visibility:
| Action | Why |
|---|---|
| Refresh content every 30 days | ChatGPT favors recent content |
| Write deep, structured guides (2k+ words) | Claude rewards depth and structure |
| Add FAQ and Organization schema | Gemini correlates with structured data |
| Get cited by multiple sources | Perplexity weights citation diversity |
| Monitor every 7 days, not monthly | ChatGPT and Gemini change week-to-week |
| Track all 4 engines, not just one | Each engine has different citation patterns |
How visibility.so Makes Brand Tracking AI Search Easy
I ran this GEO tracking experiment using visibility.so’s AI Visibility monitoring. It automatically checks all 4 engines against prompts you define and surfaces changes over time.
Instead of manually checking each engine, you get:
- Daily snapshots of which engines cite your brand
- Change alerts when your visibility drops or improves
- Prompt-level visibility – see exactly which queries surface your brand
- Competitor tracking – see who else gets cited for the same prompts
All included in every plan. No separate AI API costs. No DataForSEO credits.
Frequently Aasked Questions
What is GEO (Generative Engine Optimization)?
GEO is the practice of optimizing your brand’s visibility inside AI-generated answers — in ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. It’s different from traditional SEO because AI citation follows different signals than ranking algorithms.
Which AI engine should I optimize for first?
Start with ChatGPT (largest user base) and Perplexity (most transparent citation behavior). Then expand to Claude and Gemini.
How often should I check AI visibility?
Weekly. ChatGPT and Gemini change frequently. Monthly checks can miss important shifts.
Does traditional SEO help with AI visibility?
Yes, but not directly. A well-structured site with good content is more likely to be cited. But the signals that drive AI citation (content freshness, source diversity, structured data) are different from traditional ranking factors.
Can I track competitors’ AI visibility?
Yes. visibility.so lets you set up prompts and track which brands appear for any topic – not just your own.
How is this different from rank tracking?
Rank tracking shows your position on a search engine results page. GEO tracking shows whether your brand appears in AI-generated answers. They’re complementary – one covers traditional search, the other covers AI search.
Found this useful? Share it with someone still optimizing only for Google. And follow @sanjayshankarr for more GEO research.