summary

Andy Stapleton - How to Use ChatGPT’s Deep Research to Save HOURS on Research

Thumbnail of the video “Andy Stapleton - How to Use ChatGPT’s Deep Research to Save HOURS on Research”
Community upload. AI output has not been independently verified.

Short Summary

Main recommendation: Use ChatGPT's Deep Research feature to produce a first-draft literature review or research introduction, then check and verify the output yourself before using it. The speaker says it can cut hours of research and writing time, with a 10,000-word literature review generated in about 10 minutes from 40 sources. Its main limits for academic use are weak reference export and writing that still sounds somewhat AI-generated.

Key Points

  • Deep Research is available through a new button on ChatGPT and, at the time of the video, costs $200 US; the speaker expects it to roll out to Plus, Premium, and institutional users later.
  • Answering the clarifying questions it asks (scope, time frame, topics, sources, format, length) produced a more tailored result than the earlier standard ChatGPT answer.
  • Its literature review on OPV devices followed the requested theme structure and linked each reference to the section of text it supported, though some sources were web articles rather than peer-reviewed work.
  • For an introduction on solar concentration mirrors, it generated 24 sources in about four minutes; the first output was longer than requested, and after asking for a shorter ~300-word version, the speaker judged the text better than a typical first draft.
  • Its main academic limitations are that citations cannot be exported to LaTeX, BibTeX, or Zotero, and the introduction it wrote did not include references inline.

Details Worth Noting

  • Concrete instructions:
    • Open ChatGPT, click the Deep Research button, enter your prompt, and click go.
    • When it asks clarifying questions, answer them about scope, time frame (the speaker used the past five years), preferred sources (the speaker said "prioritize the best information"), format (the speaker asked for themes such as materials), and length (the speaker asked for 10,000 words).
    • Set out your desired structure before you start, because it followed the requested section structure closely.
    • Click the sources to check where each claim comes from, and read and verify the material yourself, since the speaker noted some non-peer-reviewed sources and possible hallucination.
    • To write a short paper introduction, ask for a condensed version, such as about 300 words, and then use the sources it already listed, since the shorter text did not include references.
    • Export or copy the sources manually, and move them into a reference manager or Overleaf, because direct export to LaTeX, BibTeX, or Zotero was not available.
  • The speaker tested it on the OPV literature review, a 10,000-word request that took about 10 minutes and found 40 sources, and on a solar concentration mirror introduction that took about four minutes and found 24 sources.
  • The speaker also referenced Humanity's Last Exam, where Deep Research scored 26.6% accuracy versus 13% for OpenAI's o3-mini.
  • The speaker suggests specialized tools such as Thesis AI may be more useful because they can export references into Overleaf while giving full control over referencing.

Bottom Line

Deep Research can dramatically speed up the early stages of academic writing, but it works best as a verified first draft that you must check and reference yourself.

Facts

  1. OpenAI's Deep Research scored 26.6% on Humanity's Last Exam, versus 13% for OpenAI's o3-mini, according to Andy Stapleton.

    3:00