What research tools give AI agents better results than plain Google?
Short answer
Give agents AI-native search and scraping tools, and keep them on free tiers until they prove out. I gave my growth agent Exa for AI search and Firecrawl for scraping, both on free tiers, plus a prepaid $50 keyword-data API. A peer swears by the Perplexity API for fresh, source-level results.
A research agent is only as good as what it can see. If all it has is a Google-style scrape, it gets SEO-style results. I wanted the tools enterprises use that most normal entrepreneurs don't know about.
Which tools did I give my research agent?
- Exa for AI search
- Firecrawl for scraping pages
- A keyword-data API, with $50 prepaid so spend can't run away
- An open-source SEO tool
- Yesterday's transcripts and posts as input, so it researches from what we're actually doing
The rule I gave it: stay under the free tiers, for sure. Add cheap or free research tools first and only pay once something earns it.
Is the Perplexity API better than Claude for research?
A peer made the case to me, and his example was convincing. He says the Perplexity API returns the most recent source-level data instead of SEO-style Google scraping. It caught a three-day-old documentation change in an accounting API that Claude missed, and that change was what was breaking his build. He also pointed to Exa (1,000 free requests, then $7 per 1,000) and Firecrawl as worth testing. Those are his findings, so test them on your own work.
Which model should do the research?
My standing rule is Sonnet by default because it allows far more chats before hitting limits, but Opus for newsletter research prompts. My media buyer pushed back that Sonnet doesn't research as thoroughly, and I agreed.
For one-off deep reports, I use ChatGPT's deep research mode with a prompt the AI wrote for me. It takes about ten minutes and reads 100-plus websites: how meta-prompting works.
How do I research links, people and competitors fast?
I keep a dedicated link-analysis bot. All I do is drop interesting links, websites, YouTube videos or tweets, ask how we could use this for our business, and it writes a long document.
I also make specialized bots for studying specific marketers and strategies, separate from my main chief-of-staff bot. You should be doing research on people occasionally, and I created a bot specifically for it instead of using the main one.
What should a growth agent research?
Broadly. A growth-hacker agent should understand the internet in all its capacity: website, SEO, rank and rent, lead gen, social media, Reddit, Quora, Pinterest, cold email and cold DMs. When mine drifted into newsletter-platform minutiae instead of finding growth channels, I pulled it back.
Answered by TJ Larkin. From agent setup sessions, partner calls and team calls in 2026. Last updated .