Four Engines, Four Backends

ChatGPT, Claude, Perplexity, Gemini are not one target. They are four different search systems wearing chat interfaces — and a win on one rarely transfers.

Why this, for you: "optimize for AI search" sounds like one job. It's four. Before you spend a day tuning a page, know which engine you're targeting and what its backend rewards — otherwise you optimize for a retrieval system that isn't even the one citing you.

Only 11% of cited domains appear on both ChatGPT and Perplexity for identical queries. Each platform runs a different search backend, a different crawler, and different freshness weights.

1 Four backends, four biases

Citation behavior follows retrieval architecture. The engine you're optimizing for inherits the biases of the search index sitting underneath it.

Engine	Backend	What it rewards
ChatGPT	Bing index	Tracks Bing's top-10; retrieves on only ~31% of prompts
Claude	Brave Search	High factual density; ~86.7% overlap with Brave's top results
Perplexity	Own index	Retrieves on every query; strongest freshness weighting
Gemini	Google + Knowledge Graph	Brand-owned sites (52.15% of its citations); multimodal

Signals don't transfer because the ranking mechanisms differ structurally. Winning ChatGPT and Perplexity both takes different content architecture — not the same page with minor tweaks.

2 Citation volume is wildly uneven

How many sources an engine cites per answer sets how competitive a citation slot is. Perplexity is generous; Copilot is a needle's eye.

Platform	Citations / response
Perplexity	21.87
Google AI Overviews	17.93
ChatGPT	7.92
Microsoft Copilot	2.47

Perplexity cites ~2.8× more sources per answer than ChatGPT — a wider door, but one weighted heavily toward fresh and community-generated content.

3 The crawler split: retrieval vs training

Every platform runs separate crawlers for training and for real-time citation. Blocking the wrong bot has no effect on whether you get cited — you can opt out of training while staying citation-eligible.

# Allow retrieval bots → stay eligible for AI answers OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot # Block training bots → opt out of training datasets GPTBot, ClaudeBot, Google-Extended

Most AI crawlers can't run JavaScript

Only Googlebot, Applebot, and Bingbot execute JS. OpenAI and Perplexity crawlers see raw HTML only — server-side rendering is mandatory, or your content is invisible to them.

↪ Your win: pick the engine before you tune the page

Name the target backend — Bing (ChatGPT), Brave (Claude), own index (Perplexity), Google (Gemini); each rewards different signals.
Don't assume transfer — only ~11% domain overlap between ChatGPT and Perplexity; one page rarely wins both.
Allow retrieval bots, block training bots — separate user-agents; blocking training doesn't cost you citations.
Server-side render — most AI crawlers can't execute JavaScript; client-rendered content is unreadable to them.

Retrieval practice — recall, don't peek

Question 1Claude's web search is powered by…

Question 2Domain overlap between ChatGPT and Perplexity citations is about…

Question 3Blocking a platform's training bot in robots.txt…

Question 4Because most AI crawlers can't run JavaScript, you must…

Question 5 · spaced recall from Lesson 01The strongest off-site predictor of AI citation is…

Ask me anything. Want the full user-agent reference table, or why Perplexity's freshness bias can fight ChatGPT's authority bias? Next, Part 2 begins: Answer-First, Atomic Pages — structuring content so a retriever can find and quote it.