OpenAI · 2025 · 02 · 02 · Feature · ~2 min read
OpenAI launched Deep Research
What's actually new
- Up to half an hour of automatic browsing per query — multi-page reading, cross-checking, follow-up searches.
- Reports come with inline citations for every source it consulted. Clickable. Dated.
- Powered by a tuned o3 model — the first time OpenAI's thinking line showed up in a normal consumer feature beyond chat.
If you want more
Worth knowing
- Source quality is mixed. The agent will happily cite SEO content farms, low-quality news aggregators, and even other AI-generated articles unless you steer it.
- $200/month Pro tier got generous limits at launch. Plus users got a tiny allocation. Real cost-per-report is high.
- 'Deep' is relative. For genuinely deep questions, this is where a human researcher's first hour goes — not where their final answer lives.
Who should care
Researchers and analysts whose first hour of any project is gathering background. Students writing essays (with disclosure). Journalists pulling source lists. Anyone who pays for hours of reading just to get oriented.
What to do about it
Use Deep Research for the orientation phase of any project, not the conclusion. Always click the citations to verify, especially for any claim you'd repeat in public.
Honest take
Deep Research was the first time 'slow AI' became a normal consumer feature instead of a research demo. The output is genuinely useful for getting oriented in an unfamiliar topic — and genuinely dangerous if you treat it as final. The citations make trust-checking possible: you can tell at a glance whether the agent leaned on solid sources or wandered into farms. That citation discipline is the most under-noticed change in consumer AI in the past year.
Sources
- OpenAI — Introducing Deep Researchvendor
- Simon Willison — OpenAI Deep Researchthird party
- Ethan Mollick — Deep Research is genuinely usefulthird party
Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com