Manuel B. Garcia

Manuel B. Garcia serves as the Senior Director for Educational Technology and Digital Learning at FEU Institute of Technology, Manila, Philippines. Read More

Contact Info

1607, FEU Tech Building,
P. Paredes St, Sampaloc,
Manila, Philippines
mbgarcia@feutech.edu.ph

Follow Me

Supplement D · Guide 8 of 80

What Research Tasks Can Generative AI Help Researchers With?

Generative AI can assist with tasks across much of the research process, including exploration, literature work, coding, data preparation, analysis support, writing, and communication. Its usefulness depends on the task, the system, the evidence available for verification, and the consequences of getting the output wrong.

By Manuel B. Garcia
08
Research Tasks Generative AI Can Help With Guide 8 of 80
01 · The Question

Where Can Generative AI Actually Help in a Research Project?

Once researchers move past asking whether generative AI should be used at all, a more practical question appears: what is it actually useful for?

The answer is broader than “writing.” Generative AI can help researchers generate possibilities, transform information, work with supplied documents, produce or explain code, organize material, support some analytical tasks, revise communication, and interact with other research tools.

But capability is not the same as suitability. A system being able to produce an answer does not mean that answer should determine a research decision. The useful question is therefore not merely “Can AI do this task?” but “What role can AI appropriately play in this task, and how will I know whether its output is good enough?”

02 · The Short Answer

Generative AI Can Help Across Much of the Research Lifecycle

In Brief

Generative AI can assist researchers with idea exploration, terminology development, literature-related work, coding, data preparation, preliminary analysis support, drafting, revision, translation, summarization, and research communication, provided its role is appropriate and consequential outputs are independently evaluated.

Its strongest role is often as an assistive system that generates candidates, transformations, explanations, or preliminary outputs for researchers to inspect. Some tasks require much stronger safeguards than others, particularly when AI can alter evidence, analysis, interpretation, participant information, or scholarly conclusions.

03 · What You Need to Know

Where Generative AI Can Fit Into the Research Process

Usefulness Depends on the Task, Not Simply the Tool

A general-purpose generative AI system may appear capable of doing almost everything because the same interface can discuss theory, summarize documents, generate code, revise prose, analyze uploaded files, and create visual or textual material.

That versatility can encourage researchers to evaluate the product as a whole: “Is this AI good for research?”

A better evaluation occurs at the task level.

The same system might be excellent at converting a researcher-written explanation into plain language, moderately useful at suggesting search terminology, unreliable for generating bibliographic references from memory, and completely inappropriate for processing identifiable participant information under a particular data-governance arrangement.

Generative AI should therefore be understood as part of an AI-assisted research workflow, with suitability assessed separately for each task.

Generative AI Can Help You Explore a Research Problem

Early-stage research involves considerable conceptual exploration. Researchers may need to turn a broad interest into searchable concepts, identify alternative terminology, consider possible variables, expose assumptions, or examine a problem from competing perspectives.

Generative AI can be useful for producing possibilities quickly. You might ask it to:

  • suggest alternative terms for a concept;
  • generate possible ways of narrowing a broad topic;
  • identify questions you may need to investigate;
  • propose counterarguments to an initial position;
  • compare several researcher-supplied conceptual definitions;
  • suggest variables or relationships worth investigating;
  • turn an informal idea into several candidate research questions.

The word candidate matters. A generated research question does not become worthwhile merely because it sounds scholarly. Researchers still need to establish novelty, significance, feasibility, theoretical grounding, ethical acceptability, and methodological answerability.

Generative AI Can Help Develop Search Vocabulary

Researchers often know the phenomenon they want to investigate before they know the terminology used across the literature.

Generative AI can suggest synonyms, broader and narrower concepts, spelling variants, related constructs, and alternative terminology that can subsequently be tested in scholarly information systems.

For example, a researcher interested in why people stop using a technology might explore terms such as discontinuance, abandonment, post-adoption behavior, continued use, resistance, non-use, and disengagement.

This is a useful generative task because poor suggestions can simply be discarded after checking how the terminology actually appears in the literature.

It is not equivalent to conducting the literature search itself. As explained in the distinction between AI assistants, search engines, research databases, and reference managers, generating plausible scholarly information and retrieving identifiable scholarly records are different operations.

Generative AI Can Support Literature Discovery and Screening

AI-enabled research systems may support semantic search, recommendation, document ranking, question answering over collections, screening assistance, or other forms of literature discovery.

Generative AI can also help researchers work with records they have already retrieved. Depending on the system and workflow, it may assist with extracting candidate information from abstracts, comparing papers, identifying potentially relevant records, or organizing literature according to researcher-defined criteria.

The methodological stakes rise considerably when AI influences which studies enter or leave an evidence base. In a systematic review, for example, a missed eligible study or incorrectly excluded paper may affect the synthesis. Automation can assist screening, but researchers need an appropriate validation strategy rather than assuming that an AI classification is correct.

Generative AI Can Help Researchers Work With Documents They Supply

One of the more useful research applications is asking AI to work with material that is explicitly available in its context.

A researcher might provide a paper and ask for:

  • a structured extraction of its research question, sample, method, and findings;
  • a comparison with another supplied paper;
  • identification of passages discussing a specified concept;
  • a plain-language explanation of a difficult section;
  • a table constructed from explicitly identified information;
  • questions that should be checked while critically reading the paper.

This is generally more traceable than asking a model to describe a publication it may not actually have access to. Still, source grounding does not eliminate error. The system may omit qualifications, misunderstand a table, confuse constructs, or generate an inference not stated in the document.

Generative AI Can Help With Research Design, but Suggestions Need Methodological Evaluation

Researchers can use generative AI to generate possible instrument items, interview questions, experimental procedures, sampling considerations, variables, analytical approaches, or threats to validity.

This can function much like structured brainstorming. For example, after developing a survey instrument, a researcher might ask an AI system to identify potentially ambiguous items or suggest respondent interpretations that were not anticipated.

The AI's suggestion is not validation evidence.

A generated questionnaire does not become psychometrically sound because it looks professional. A generated experimental design does not establish internal validity. A proposed sampling strategy still needs to be justified against the research question, population, design, and inferential goals.

Generative AI Can Help Generate and Debug Research Code

Code generation is one of the more immediately useful applications for researchers who work with R, Python, SQL, MATLAB, JavaScript, or other programming environments.

Generative AI may help:

  • write a first version of a function;
  • translate code between languages;
  • explain unfamiliar syntax;
  • identify possible causes of an error;
  • generate tests;
  • comment or document existing code;
  • suggest more efficient implementations.

Generated code should be treated as code written by an unverified contributor. Run it. Inspect it. Test it on known cases. Examine assumptions. Check package functions against official documentation when necessary.

Code that executes without an error is not necessarily code that implements the correct analysis.

Generative AI Can Assist With Data Preparation

Depending on the type of data and the system being used, AI may support transcription, text normalization, information extraction, categorization, formatting, entity recognition, conversion between structures, and other preparatory tasks.

These applications can save substantial time, particularly with large volumes of unstructured material.

They can also introduce systematic error. An automated transcription system might repeatedly misrecognize technical terminology. An extraction system might omit negative statements. A classifier might perform differently across linguistic or demographic groups.

Researchers should therefore validate AI-assisted transformations rather than assume that preprocessing is methodologically neutral.

Generative AI Can Support Qualitative Analysis Without Becoming the Analyst

Generative AI can work with textual data in ways relevant to qualitative research. Researchers may experiment with candidate classifications, compare passages, identify possible patterns, apply researcher-defined coding frameworks, or use AI output as a contrasting interpretation.

That does not mean a language model can simply be instructed to “do thematic analysis” and return findings that inherit methodological legitimacy from the name of the method.

Qualitative analysis involves methodological commitments, reflexivity, context, interpretation, engagement with participants' meanings, and decisions about what counts as evidence. AI-generated themes may be interesting candidates, but the researcher's analytical responsibility remains.

Generative AI Can Support Quantitative Analysis

Generative AI may help researchers understand statistical concepts, generate analysis code, translate output into more accessible language, propose diagnostic checks, explain error messages, or suggest possible reasons for an unexpected result.

These uses can be valuable, especially when the output is treated as something to investigate.

The risk appears when an explanation becomes an interpretation merely because it sounds statistically sophisticated. If AI says a coefficient changed because of “suppression,” “multicollinearity,” “mediation,” or another mechanism, researchers need to determine whether the actual data and model support that explanation.

A statistical term generated in the correct grammatical position is not a diagnosis.

Generative AI Can Help Researchers Challenge Their Own Reasoning

Not every valuable use involves producing material for the final research output.

A researcher can ask AI to act as a critical interlocutor: What assumptions does this argument depend on? What alternative explanations fit these results? What would a skeptical reviewer challenge? What evidence would falsify this interpretation? What population does this claim fail to represent?

This can be particularly useful because the generated output serves as intellectual friction rather than authority. The researcher evaluates the objections rather than accepting them.

Used this way, AI may help expose questions that deserve further thought without being asked to decide the answer.

Generative AI Can Help With Drafting and Revision

Writing is the most visible use of generative AI in research, but even here the possible roles differ substantially.

AI can help researchers:

  • revise awkward sentences;
  • reduce repetition;
  • change the level of technicality for a particular audience;
  • suggest alternative structures;
  • shorten researcher-written material;
  • translate or improve language;
  • generate candidate titles or headings;
  • draft text from researcher-provided instructions or evidence.

The last activity is more consequential than the first. As the system moves from transforming researcher-created material to generating substantive scholarly content, questions about AI-assisted versus AI-generated research become increasingly relevant.

Generative AI Can Help With Research Communication

Research often needs to be communicated beyond the journal article. Generative AI can help transform verified research into abstracts, plain-language summaries, presentation outlines, social-media copy, public-facing explanations, teaching materials, or descriptions for different audiences.

This is a transformation task, but transformations can still distort.

A public summary may turn “associated with” into “causes.” A shortened explanation may omit an important limitation. A generated headline may exaggerate a finding because stronger language sounds more compelling.

Researchers should therefore check whether the transformed version preserves the epistemic strength of the original research.

The Best Tasks Usually Have a Clear Verification Path

A useful way to evaluate potential AI tasks is to ask: If the system is wrong, how will I know?

Task Possible verification Typical consequence of error
Suggest alternative keywords Test terms in actual literature searches Usually limited if poor suggestions are discarded
Rewrite researcher-authored prose Compare with the original meaning and evidence Can range from minor to serious if claims change
Generate code Inspect, execute, test, and compare with expected results Potentially serious if code determines results
Summarize a supplied article Check summary against the article Potentially serious if misrepresentation enters synthesis
Generate references from memory Verify each record in authoritative bibliographic sources High risk of introducing nonexistent or incorrect citations
Interpret primary findings Return to data, method, theory, and relevant scholarship High because conclusions may be distorted

This suggests a practical principle: generative AI is often easiest to use responsibly when its output is cheap to verify relative to the benefit of generating it.

Capability Does Not Equal Permission

A system may technically be able to process interview transcripts, unpublished manuscripts, confidential peer-review material, proprietary datasets, personal information, or sensitive research records.

That does not mean researchers are permitted to provide those materials to it.

UNESCO's guidance emphasizes privacy, human agency, ethical validation, and critical evaluation in research uses of generative AI. The European Commission's updated living guidelines likewise center accountability, transparency, responsibility, research integrity, and risks arising from the handling of information in AI-enabled environments.

Before using an AI system for a task, researchers therefore need to consider not only whether it can perform the task, but whether the data can be processed there and whether institutional, ethical, contractual, funder, or publication rules permit the proposed use.

04 · A Practical Example

Using Generative AI Across One Research Project Without Giving It the Project

Hypothetical Example

A Mixed-Methods Study of Student Engagement

A researcher is studying how students engage with a university's online learning environment.

Topic exploration The researcher asks AI for alternative terminology related to behavioral, cognitive, and emotional engagement. She checks the terminology against scholarly literature.
Literature work After retrieving relevant papers from research databases, she provides selected articles to an approved system and asks it to extract candidate information into a structured table. She verifies the extraction against each paper.
Quantitative analysis AI helps generate R code for a diagnostic plot. The researcher checks the functions against documentation, runs the code, and verifies that the output represents the intended variables.
Qualitative work The researcher asks the system to apply several researcher-developed candidate codes to de-identified excerpts. She reviews every classification and uses disagreements as prompts for further analytical reflection.
Writing After writing the findings herself, she asks AI to identify repetitive passages and suggest more concise formulations. She rejects revisions that alter the strength or meaning of the claims.
Communication After the paper is complete, AI produces a candidate plain-language summary. The researcher compares it with the manuscript and corrects several statements that overstate the findings.

Generative AI assisted throughout the project, but its role changed from stage to stage. The researcher did not apply one blanket level of trust. Each output was evaluated according to what it could affect and how it could be checked.

05 · What Researchers Often Get Wrong

Common Mistakes When Choosing Research Tasks for Generative AI

Misconception

If AI Can Do a Task, Is That Evidence That It Should?

No. Technical capability does not establish methodological suitability, ethical acceptability, data-governance compliance, or permission under institutional and publication policies. Evaluate the particular use rather than the impressive breadth of the system.

Misconception

Is Writing the Main Research Use of Generative AI?

No. Generative AI can support coding, document processing, information extraction, search-term development, analysis support, critique, translation, data preparation, and communication, among other tasks. Writing is simply one of its most visible applications.

Misconception

If I Give AI the Original Sources, Can I Stop Checking Its Answers?

No. Grounding the system in source material can make verification easier and reduce some forms of unsupported generation, but the system can still omit qualifications, misread evidence, conflate documents, or introduce claims not supported by the sources.

Misconception

If Generated Code Runs, Is It Correct?

No. Code can execute successfully while implementing the wrong transformation, statistical model, variable definition, denominator, or assumption. Research code needs substantive validation, not merely the absence of an error message.

Misconception

Can AI Save Me From Learning the Method?

It can help explain methods and lower some technical barriers, but relying on AI does not remove the need to understand consequential methodological choices. If you cannot determine whether an analysis is appropriate, you cannot provide meaningful oversight simply by approving generated output.

Misconception

If AI Makes Research Faster, Does It Make Research Better?

Not necessarily. Efficiency and research quality are different outcomes. Generative AI may reduce time spent on some tasks while also allowing errors, weak reasoning, or unsupported content to be produced more rapidly. Whether generative AI improves research quality depends on what it changes in the research process and how those changes are evaluated.

06 · What This Means for You

Use Generative AI Where Its Output Can Be Evaluated

Rather than creating a permanent list of “safe” and “unsafe” AI tasks, evaluate the relationship between the task, the system, the evidence, and the consequences of error.

A simple decision framework

If AI is generating possibilities
Use the output as candidates for investigation rather than as established knowledge.
If AI is transforming material you created or supplied
Compare the output with the source and check that important meaning, evidence, and qualifications were preserved.
If AI is producing code, classifications, or extracted data
Validate performance and errors using an appropriate test or reference standard.
If AI output can materially change a finding or conclusion
Raise the standard of verification and retain direct researcher control over the scholarly judgment involved.
If you cannot verify the output adequately
Do not make consequential research decisions depend on it merely because the output appears sophisticated.

Generative AI is particularly useful when generation is faster than doing the task manually but verification remains feasible. When checking the output would require expertise or evidence you do not possess, automation can create an illusion of progress rather than genuine research assistance.

07 · A Quick Checklist

Before Giving a Research Task to Generative AI, Check These Points

Before using generative AI for a research task, check:
Define exactly what you want the system to generate, transform, extract, classify, explain, or suggest.
Determine whether the task is primarily exploratory, operational, analytical, interpretive, or communicative.
Identify what evidence or reference standard you can use to evaluate the output.
Consider how an undetected error would affect the data, method, findings, participants, or conclusions.
Check whether the material you intend to provide may be processed by the system under applicable privacy, ethics, contractual, and institutional requirements.
Verify citations, factual claims, code, calculations, and interpretations through appropriate independent sources or tests.
Keep enough documentation to explain consequential AI use and reproduce or audit the workflow where necessary.
Check current funder, institutional, journal, publisher, and ethics requirements before using AI for consequential or sensitive tasks.
08 · Frequently Asked Questions

Frequently Asked Questions About Generative AI Research Tasks

Can generative AI help with a literature review?

Yes, it can assist with terminology development, discovery, document comparison, extraction, screening support, summarization, and organization. It should not be assumed to have performed a comprehensive or reproducible literature search unless the actual retrieval process supports that claim.

Can generative AI help write research questions?

It can generate candidate questions and alternative formulations. Researchers still need to determine whether a question is significant, theoretically grounded, answerable, feasible, ethical, and appropriately scoped.

Can generative AI analyze qualitative data?

It can assist with classification, candidate coding, comparison, extraction, and pattern exploration. Whether those uses are methodologically appropriate depends on the qualitative approach, research question, data governance, validation strategy, and role assigned to human interpretation.

Can generative AI perform statistical analysis?

AI systems may generate statistical code, explain output, invoke computational tools, or assist with analytical workflows. Researchers still need to verify that the method, assumptions, implementation, calculations, and interpretation are appropriate.

Can generative AI write my research paper?

It can technically generate manuscript text, but technical capability does not settle whether extensive delegation is appropriate or permitted. Researchers need to consider authorship, intellectual contribution, factual verification, source accuracy, disclosure requirements, and the responsibilities that cannot simply be transferred to AI.

Can AI help generate research instruments?

It can generate candidate questionnaire items, interview questions, scenarios, or other materials. Those outputs do not constitute evidence of validity or reliability. Instrument development still requires methodological evaluation appropriate to the intended construct and use.

What kinds of AI tasks are easiest to use responsibly?

Tasks tend to be easier to manage when the output is readily verifiable, errors have limited consequences, and the researcher can compare the result against an authoritative source, known expected result, supplied material, or established method.

Are there research tasks I should never delegate completely to AI?

Yes. AI can assist with many consequential activities, but researchers should retain direct responsibility for tasks involving core scholarly judgment, accountability, ethical responsibility, authoritative verification, and final interpretation. The boundary is examined directly in which research tasks should never be delegated completely to AI.

09 · The Bottom Line

Generative AI Is Most Useful When Researchers Can Evaluate What It Produces

The Bottom Line

Generative AI can help with tasks throughout the research process, from exploring ideas and literature to coding, analysis support, writing, and communication, but its usefulness depends on whether the task is suitable and its output can be adequately evaluated.

Do not choose tasks simply because AI can perform them. Look for places where generation or transformation saves meaningful effort while you retain the evidence, expertise, and authority needed to catch errors and make the final scholarly judgment.

10 · Sources and Further Reading

Sources and Further Reading

11 · Cite this Guide

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.

Has the Field Guide helped your research?

If a guide helped clarify a question, inform a research decision, or move your work forward, I would love to hear about your experience. Your story may also help other researchers discover the Field Guide.

Share Your Experience
Takes only a few minutes