The 3-Layer ChatGPT Architecture
ChatGPT does not search the web in real-time. AEO Protocol's research reveals a 3-layer retrieval system: Bing Index, OAI-SearchBot Cache, and User-Triggered Fetches. Most citations come from Layer 2 (cached content). Understanding this architecture is critical for getting your brand recommended by ChatGPT.
Key Takeaways
- Layer 1 (Bing Index): Standard Bing SEO applies, foundation for discovery
- Layer 2 (OAI-SearchBot): Where most citations come from, recrawls daily
- Layer 3 (User Fetches): Real-time fetching, used for cache forcing
- Cache refresh is per-page, not site-wide. Each page needs individual updates.
How ChatGPT Retrieves Information
Most people think ChatGPT searches the web when you ask a question. In reality, there are three distinct layers between your brand and getting recommended.
Bing Index
FoundationChatGPT's web search is powered by Bing. Standard Bing SEO applies here. If Bing cannot find your website, neither can ChatGPT. This is where most businesses stop their optimization efforts.
What This Means For You:
- Submit your sitemap to Bing Webmaster Tools
- Ensure basic Bing SEO fundamentals are in place
- Many sites rank in ChatGPT without strong Bing rankings
OAI-SearchBot Index (Persistent Cache)
OpenAI's own crawler visits your site and stores content in a persistent cache. This is where most citations come from. When ChatGPT recommends your brand and links to your website, that link typically came from Layer 2.
Cache Characteristics:
- Adds
?utm_source=openaito cached URLs - Recrawl frequency: approximately daily for active sites
- Caches FULL page content (not just metadata)
- Dead pages persist in cache (no automatic liveness checking)
Layer 2 is where the game is played. Most brands are stuck in Layer 1.
User-Triggered Fetches (Ephemeral Cache)
Real-TimeReal-time, on-demand fetching when someone specifically requests current information. This is the least reliable layer for organic discovery but can be exploited to force cache updates.
Layer 3 Characteristics:
- No UTM parameter on fetched URLs
- Updates within 5-10 minutes
- Triggered by specific user prompts
- Can be used to force instant cache updates
Cache Fingerprinting: Identify the Source
You can identify which layer served a citation by examining the URL pattern ChatGPT uses when linking to your site.
| URL Pattern | Source Layer | Freshness |
|---|---|---|
yoursite.com/?utm_source=openai | OAI-SearchBot Index (Layer 2) | Stale (days/weeks old) |
yoursite.com/ (no utm) | User-Triggered Fetch (Layer 3) | Fresh (minutes old) |
Critical Finding: Cache Refresh is Per-Page
One of the most important discoveries from our research: when you update your homepage, only the homepage cache refreshes. Other pages remain stale until explicitly force-fetched.
Real-World Example
The Protocol
After any content update, you must force-fetch EACH updated page individually. Do not assume a site-wide crawl will pick up your changes.
ChatGPT vs Gemini: Different Architectures
Gemini uses a completely different retrieval system called "Grounding with Google Search." Understanding both is essential for full AI visibility.
| Aspect | ChatGPT | Gemini |
|---|---|---|
| Primary Focus | Cache injection, force-fetch | Google SEO, E-E-A-T |
| Speed of Updates | Minutes (via prompts) | Dependent on Google indexing |
| Key Ranking Factor | Being in OAI-SearchBot index | Being in Google Top 10 |
| Content Type | Direct answers, structured | Fact-dense, verifiable claims |
| Trust Signals | Less critical | Critical (Grounding requires verification) |