Search.
How to (best) search with AI. PS: it's not with ChatGPT.
I use AI for (mostly) two tasks: (1) searching things, (2) writing things.
Today, we will focus on searching online.
Table of contents
✦ My method & results.
✦ The best model to search anything.
✦ The best model for each 15 case studies.
✦ Search vs. Deep Research. How to pick?
✦ Best of prompts to search (to copy & paste).
1. My method & results.
So many tools exist to search online using AI: ChatGPT (duh), Perplexity, but also Grok, DeepSeek, Claude, Kimi, Qwen, and probably another 10x LLMs.
Which one is best?
To know, you would need to run dozens of the same search through all of these tools. Let’s say 9x queries… to 7x different tools… that’s 63 queries!
Who’s crazy enough to compare 63 queries?
Well, I am. I’m Ruben, and welcome to “How to AI”.
I spend most of my awake time testing, comparing, and breaking AI tools to give you exactly (1) which one is best (2) and how to use it best.
I tested 7x different AI companies to answer 9x search queries:
ChatGPT, the original AI.
Perplexity, made for search.
Grok, from xAI (which has access to Twitter).
Deepseek, the Chinese sensation that just got an upgrade.
Claude, that also got an upgrade this week with Sonnet 4.5.
Gemini, the ChatGPT of Google, known for its image editing.
Kimi, another AI tool from China, known for being the least biased.
If you want to compare each model against the 15 queries I picked, go to this google doc: https://docs.google.com/document/d/1pWuMCBVQo1zKcgKltX_BZxAr31KgxmOlp3Vzvmc5Hxc/edit?usp=sharing (on the tab “BEST AI TO SEARCH”).
If you want the results directly, keep reading.
2. The best model to search (anything).
Out of 9 case studies…
Perplexity is the clear winner.
1 - It found the most info about me.
2 - It’s the best tool to find YouTube summaries (and not Gemini… how crazy?).
3 - It can easily search through 51 sources, super fast. Much faster than ChatGPT.
4 - Best formatting for its answer + it follows my instructions of “Reddit only”.
5 - Here, it was the only one to give the exact address where I had to send my papers. Fascinating. That’s real help.
3. The best model for each case study.
Here’s the breakdown.
1. Breaking news synthesis
Prompt: “Summarize what happened in the last 72 hours about the Sora 2 release. Use at least 5 unique sources. Show timestamps, contradictions, and cite each source.”The winner for this specific query: DeepSeek, concise, and it has everything I needed.
2. Person dossier (OSINT-light)
Prompt: "Build a current public profile of [Ruben Hassid]. Include employer, verified socials, 3 notable works, and safe contact paths. Cite every item with working links.”The winner for this specific query: Perplexity. I can’t believe the level of depth.
3. High-intent product choice
Prompt: "The best [PRODUCT: ping pong racket, BUDGET: under $300, REGION: available in Tel aviv] today. Give 3 options with pros/cons, live prices, warranty notes, and 5 owner reviews linked.”The winner for this specific query:
ChatGPT (actually $300), Kimi (most reviews), Grok (best overall).
4. YouTube search + synthesis
Prompt: “Find the 5 most authoritative YouTube videos to learn “How to write a thesis with AI”. For each: channel, publish date, why it’s credible, and exact timestamp for the most valuable 2 minutes.”The winner for this specific query: Perplexity. It’s so good for YouTube.
5. Academic papers to practice
Prompt: “List the 7 most cited papers since 2021 on “Prompt Engineering to avoid hallucinations”. Provide BibTeX, open-access/preprint links, and a 2-line takeaway per paper.”The winner for this specific query: No winner, they were all equally good. Maybe Perplexity?
6. Reddit pain-points mining
Prompt: “Extract the top 10 user pain points about [PRODUCT/CATEGORY] from Reddit posts since 2024-01-01. Quote snippets, link threads, and group into themes.”The winner for this specific query: Perplexity and Grok are best.
7. Regulation lookup
Prompt: “What are the live requirements for [LICENSE/PERMIT] in [JURISDICTION]? Provide controlling statutes, last update date, fees, and application links.”The winner for this specific query: Perplexity literally gave me the paper to fill, and where.
8. B2B market size sanity check
Prompt: “Estimate 2026 TAM/SAM/SOM for [NICHE]. Show 3 independent sources, your triangulation math, and explicit assumptions.”The winner for this specific query: Both Claude & Grok are the best. Slight edge for Claude.
9. Travel with constraints
Prompt: “Design a 3-day plan for [Koh Phangan] given [CONSTRAINTS: Remove worker, eat high-quality beef + $2000/month + walking only]. Include booking links, opening hours, and transit time estimates.”The winner for this specific query:
DeepSeek made a diagram. ChatGPT was the most precise.
4. Search vs. Deep Research. How to pick?
There is a (major) difference between search and deep research.
Every tool has this “deeper search” in its toolkit.
Here’s how to get it for ChatGPT:
Pros:
You don’t need strong prompt engineering. It knows what to search better.
Instead of 10-30 sources, a deep research typically includes 50+ (if not 100+).
The output is also much, much longer (sometimes 20+ pages of text!).
Cons:
It takes forever. Like sometimes 15-25 minutes to complete one search.
Well, longer output does not mean better. You have to read everything.
It still does not search “the entire web”. Cf. my blog around myths.
My personal favorite:
1 - I often use the “Research tab” from Perplexity with Claude Sonnet 4.5 Thinking.
2 - Another great one is directly from ChatGPT. Pro: You don’t need an extra subscription. But Perplexity is part of my “must-have” AI on this blog.
3 - Claude's research is great for a concise output + deep search combo. Make sure to use the new Sonnet 4.5 with it.
5. Best-of prompts to search (copy & paste).
Perplexity is the best AI to search.
So here are my favorite copy & paste prompts for it: