SciSpace Academy
Tutorial

The Researcher's Workflow with SciSpace: AI That Supports Judgement

Jason GulyaEnglishResponsible AI

TL;DR

1-min read

In this tutorial, Jason Gulya, Professor of English and Communications and Chair of the AI and Academic Integrity Committee at Berkeley College, outlines a researcher-led workflow in SciSpace. It is aimed at researchers, educators and students who want to use AI in academic work without handing over their judgement, and it favours a disciplined, transparent approach over quick answers.

Key takeaways

  • Defining the scope of an academic project before turning to AI keeps the work focused and gives you a clear standard for judging whether results are relevant.
  • Research questions should be phrased neutrally so they do not steer the AI toward a predetermined answer; asking what evidence exists works better than asking the tool to prove a point.
  • Running SciSpace queries from a researcher's perspective means approaching results as leads to examine rather than conclusions to accept.
  • Filtering results and tracing which sources link back to the original article helps you distinguish primary studies from secondary mentions.
  • Output should be evaluated critically: open the cited source, confirm the claim appears there, and note the sample, methods and limitations before relying on it.
  • Short summaries and other outputs are most useful for deciding what to investigate next, guiding your reading instead of replacing it.
Apply it

To apply this, write a scoped, neutral question for your next SciSpace query, then trace each summary back to its original source before using it.

How to do it

  1. Define your project's scope first

    Set the scope of your academic project before turning to AI. A defined scope gives you a clear standard for judging whether results are relevant.

  2. Write a neutral research question

    Ask what evidence exists rather than asking SciSpace to prove a point. Neutral wording avoids steering the AI toward a predetermined answer.

  3. Treat query results as leads

    Run your SciSpace query and approach results as leads to examine, not conclusions to accept. This keeps the researcher workflow with AI under your judgement.

  4. Filter and trace original sources

    Filter results and trace which sources link back to the original article, so you can distinguish primary studies from secondary mentions.

  5. Verify claims in the cited source

    Open each cited source, confirm the claim appears there, and note the sample, methods and limitations before relying on it.

  6. Use summaries to plan your reading

    Use short summaries to decide what to investigate next, guiding your reading rather than replacing it.

Watch the tutorial34 min

AI That Supports Judgement: A Researcher-Led Workflow in SciSpacePlay video: AI That Supports Judgement: A Researcher-Led Workflow in SciSpace

Chapters

  1. Welcome: using AI for research without outsourcing your thinking
  2. How to scope a research project before using AI: question, method, output, constraints
  3. Why AI research tools tell you what you want to hear: 3 questions to ask first
  4. How to phrase a research question so AI doesn't just confirm your bias
  5. Should AI write the argument in your research paper?
  6. What happens when you give SciSpace a leading literature review prompt
  7. How to write a prompt for an AI literature review: the GRIT formula
  8. Are SciSpace's clarifying questions too broad? What that says about your research question
  9. Before vs after: how a better prompt changes the AI literature review output
  10. How to verify citations from an AI literature review: green flag vs yellow flag
  11. Checking the reference list of a SciSpace literature review
  12. How to tell if SciSpace has a paper's full text, then save it, chat with it or make a podcast
  13. How to use follow-up prompts to find gaps in an AI literature review

Try this on SciSpace

Frequently asked questions

How do I avoid bias when asking AI research questions?

Phrase questions neutrally rather than embedding the answer you expect. For example, ask what evidence exists on a relationship instead of asking AI to prove one. Request sources for and against a position, define scope clearly, and compare the output with your own reading of key studies to catch one-sided results.

How should researchers evaluate AI-generated research summaries?

Treat a summary as a lead, not a conclusion. Open the cited source and confirm the claim appears there, check whether it is the original study or a secondary mention, and note sample size, methods and limitations. If a summary cannot be traced to a real source, do not use it.

Related tutorials