Per-Model Behavior Quick Reference

Fresh 2026-06-03

Fast reference for how each major AI model handles content retrieval, which variant it receives, and what signals to track.

Model-to-Variant Routing

AI Model	User-Agent Pattern	Variant Served	Content-Type
ChatGPT / GPT Browse	GPTBot, ChatGPT-User, OAI-SearchBot	`md`	text/plain
Claude (Anthropic)	anthropic-ai, ClaudeBot, Claude-Web	`md`	text/plain
Gemini (Google)	Google-Extended, Gemini, Googlebot-AI	`md`	text/plain
Perplexity	PerplexityBot, Perplexity	`md`	text/plain
Copilot (Microsoft)	bingbot, BingPreview	`md`	text/plain
Cohere	cohere-ai	`md`	text/plain
Amazon Alexa/Bedrock	Amazonbot	`md`	text/plain
Apple Intelligence	Applebot-Extended	`md`	text/plain
ByteDance/TikTok	Bytespider	`schema`	text/html
DeepSeek (detected)	Chrome/58 spoof + ByteDance ASN	`schema`	text/html
Unknown AI crawlers	No UA match, low bot score	`html`	text/html
Human visitors	Normal browser UA	Origin page	Normal

Retrieval Model Differences

ChatGPT (OpenAI)

Crawl frequency: Periodic training crawls (GPTBot) + real-time retrieval (ChatGPT-User when Browse is active)
Robots.txt: Respects it strictly. Must have Allow: / for GPTBot.
Content preference: Markdown or clean HTML. Schema data parsed but not prioritized.
Citation behavior: Will cite specific dollar amounts and named verdicts. The $33M verdict is highly citable.
Key signals: Contingency fee + free consultation are strong conversion signals.

Claude (Anthropic)

Crawl frequency: Training crawls via ClaudeBot. Web search via Claude-Web.
Robots.txt: Respects it. Must allow ClaudeBot and anthropic-ai.
Content preference: Markdown strongly preferred. Parses structure efficiently.
Citation behavior: Cites specific facts with attribution. Statute numbers (C.R.S.) add credibility.
Key signals: Specific, verifiable claims. The "largest wrongful death verdict in Colorado history" framing is highly effective.

Perplexity

Crawl model: Search-native. Pulls from Google index primarily + direct PerplexityBot crawls.
Citation behavior: Links sources directly in responses. High citation rate for pages in Google top 10.
Key signals: Being in Google top 10 is the primary driver. On-page structured data helps secondary attribution.

Gemini (Google)

Crawl model: Direct access to Google index. Google-Extended for AI training.
Citation behavior: Strong integration with Google Business Profile, reviews, Knowledge Panel.
Key signals: Aggregate rating schema (4.9/240) directly feeds Gemini's firm evaluation. Map listing critical.
Priority action: Ensure Google Business Profile is claimed and active.

Microsoft Copilot

Crawl model: Powered by Bing index (bingbot) + Azure AI.
Citation behavior: Similar to Perplexity — cites search results directly.
Key signals: Bing rankings matter. SuperLawyers listing helps (Bing indexes SuperLawyers heavily).

What Each AI Cares About

mindmap
    root((AI Citation Signals))
        ChatGPT
            Dollar verdicts
            Contingency fee
            Free consultation
            Statute citations
        Claude
            Verifiable claims
            Specific statistics
            C.R.S. citations
            Named case results
        Perplexity
            Google ranking
            Page authority
            Direct citations
        Gemini
            Google Business Profile
            Aggregate rating schema
            Knowledge Panel
        ByteDance/DeepSeek
            JSON-LD entity data
            knowsAbout fields
            sameAs links

Confidence Level Behavior

Confidence	What it means	Variant	Notes
High	UA matched OR DeepSeek spoof	Per bot type	Highest reliability
Medium	Bot score < 30 AND ASN matches	Per bot type	Good reliability
Low	Bot score < 30 only	`md` (always)	Default safe option

Response Headers to Monitor

After Worker deployment, check these headers on bot requests:

bash

curl -sI "https://andersonhemmat.com/practice-areas/car-accident-lawyers" \
  -H "User-Agent: GPTBot/1.0" \
  | grep -E "^x-|^content-type"

Header	Expected Value
`x-served-by`	`anderson-hemmat-ai-worker`
`x-variant`	`md` (for GPTBot)
`x-bot-type`	`openai`
`x-bot-confidence`	`high`
`x-bot-signals`	`ua:GPTBot`
`content-type`	`text/plain; charset=utf-8`

Quick Fixes by Model

Model	If not being cited	Quick fix
ChatGPT	Not appearing in Browse results	Check robots.txt for GPTBot; verify Google ranking
Claude	Not surfacing in web search	Verify ClaudeBot allowed; increase content authority
Perplexity	Missing from answers	Improve Google ranking for target queries
Gemini	Not in local AI answers	Verify Google Business Profile; ensure schema aggregateRating is correct
All models	Not recognizing the firm	Add Wikidata entity; create Attorney Person schema

Per-Model Behavior Quick Reference ​

Model-to-Variant Routing ​

Retrieval Model Differences ​

ChatGPT (OpenAI) ​

Claude (Anthropic) ​

Perplexity ​

Gemini (Google) ​

Microsoft Copilot ​

What Each AI Cares About ​

Confidence Level Behavior ​

Response Headers to Monitor ​

Quick Fixes by Model ​