Search has split into three philosophies: answer engines that synthesize a response with citations, AI-augmented indexes that bolt synthesis onto the world's largest link graph, and chatbots with search that let you interrogate results in the same breath. After several months of using all three as daily drivers, the honest conclusion is that there's no winner — there's a routing table. Here's how they actually differ and which queries each one wins.
How the Three Models Work
- Perplexity — purpose-built answer engine. Runs your query through an LLM that decides what to retrieve, searches the web in real time, then synthesizes a cited answer. Inline numbered citations are the core product, and the focus/Focus modes (Web, Academic, Social, Finance) bias retrieval deliberately.
- Google AI Mode — synthesis layered on top of Google's index and Knowledge Graph, which remains unmatched in freshness, local data, and sheer coverage. Strength shows in queries where the underlying index is the differentiator: news, prices, maps, logistics, niche non-English content.
- ChatGPT Search — the most conversational. Best when the question is multi-turn and contextual: "now compare that against last year's model," or "which of those options fits my constraints." Search is retrieved on demand inside a chat you're already reasoning in.
The Tests That Separate Them
Freshness and specificity
Anything time-sensitive or long-tail — "what changed in this week's SDK release," local opening hours, a product restock — still favors Google's crawl-first architecture. Answer engines sometimes summarize a stale cached page and present it as current. Rule of thumb: if the date of the source matters more than the synthesis, go Google first.
Citation trustworthiness
Perplexity made citations the product, and its linking discipline is the most consistent of the three. Google AI Mode tends to favor its own index results, which is fine until the synthesis runs ahead of what the cited page actually says. ChatGPT Search is the loosest here — citations exist but sometimes serve as decoration for a paragraph the model half-wrote from memory. Spot-check any claim that matters.
Hallucination under pressure
The failure mode to respect: when retrieval comes back thin — obscure topic, paywalled sources, conflicting data — all three systems tend to fill the gap with fluent confidence. This looks identical to a well-sourced answer. Long-tail factual claims (exact numbers, names, niche specs) deserve a click-through to the source regardless of engine.
Research workflows
For open-ended research, Perplexity's iteration loop (query → follow-up narrows the cited base) is the strongest starting experience. ChatGPT Search wins once the research turns into analysis — "given these three sources, draft the comparison" — because the chat holds the context. Google AI Mode is fastest for one-hop factual lookups.
The Routing Table
- Fresh facts, local, time-sensitive → Google AI Mode (or plain Google, which still wins when you want links, not prose).
- Research with sources you'll want to cite → Perplexity.
- Multi-turn, opinionated, "help me think" queries → ChatGPT Search.
- High-stakes facts (health, legal, money) → any of them for orientation, then primary sources. Never an AI summary alone.
- Shopping and comparisons → AI search is genuinely better than ten tabs for spec-sheets; verify prices and availability with the actual retailer page, which is often newer than the index.
Privacy and Economics
All three log queries and all three have faced scrutiny over what retrieval requests reveal. Perplexity's browser history feature and Google's ad-ecosystem entanglement both earned press coverage; ChatGPT queries persist in chat history unless deleted. If a query is sensitive, assume it's not ephemeral in any of them.
The Bottom Line
AI search is a real upgrade for synthesis-heavy questions and a real downgrade risk for freshness-critical ones. The skill that pays this decade isn't picking a champion — it's routing: synthesis questions to answer engines, fresh facts to indexes, and reasoning conversations to chat, with a click-through habit for anything you'd bet money on. Fluent search is knowing all three are lying to you a little bit, and checking accordingly.