MAIOS

MAIOSEditorialResearch or university

Guide · Research or university

AI for research and universities: method, sources and verification

AI can accelerate literature work, analysis, coding and documentation. It becomes a research instrument only when its contribution remains traceable and open to verification.

The central question is not simply which model to use. A research workflow must define its inputs, transformation, checks and the limitations that travel with its result.

Editorial noteDraft generated by the system on 14 August 2026. Not yet reviewed.

From context to usable evidenceFive relations to preserve
  1. Question and corpusDefine the question, scope and admissible sources.
  2. TransformationState whether the system searches, extracts, classifies, translates, computes or synthesises.
  3. TraceRetain references, versions, relevant prompts or procedures and intermediate outputs.
Open the mapPrepare a test

Starting question

AI needs an explicit place in the research method.

A request such as “analyse the literature” hides source selection, inclusion criteria, extraction, synthesis and inference. Separating those steps makes it possible to decide where AI assists, where a researcher checks, and which operations must remain reproducible through other means.

A system may surface connections or generate hypotheses. A useful hypothesis is not evidence: provenance, method and intermediate transformations must remain distinct.

Operational relations

Five steps that must remain connected.

01

Question and corpus

Define the question, scope and admissible sources.

02

Transformation

State whether the system searches, extracts, classifies, translates, computes or synthesises.

03

Trace

Retain references, versions, relevant prompts or procedures and intermediate outputs.

04

Verification

Compare the result with sources, baselines, experiments or expert review.

05

Limitations

Record uncertainty, exclusions, known errors and conditions for reuse.

Boundaries and responsibility

Speed and plausibility do not replace method.

Invented citations, inaccurate reconstructions and unstable classifications can enter a workflow in convincing language. Generated text is not a source, and every consequential claim must lead back to something that can be inspected.

Research data, licensed material, personal information and unpublished results also require an authorised route. Local processing may reduce some exposure, but does not by itself settle access, copies, logs, security or responsibility.

Practical object

AI Research Workflow Map

Complete the AI Research Workflow Map. Each field exposes a relationship to verify before extending the system.

01

Question

What must be known or tested?

02

Corpus

Which sources enter, under which criteria?

03

AI operation

What transformation is assigned to the system?

04

Provenance

How can a reader recover source, version and step?

05

Check

What comparison could falsify the result?

06

Owner

Who interprets and approves scientific use?

07

Limitations

What does the method not cover?

08

Reuse

What is required to repeat or transfer the workflow?

First test

A short test should produce knowledge, not merely an output.

  1. 01
    Choose one step

    Test a bounded search, extraction, classification or analysis—not the entire project.

  2. 02
    Prepare a reference

    Use a small corpus and an expected result checked by a competent researcher.

  3. 03
    Repeat the run

    Observe variation, omissions and sensitivity to instructions or data.

  4. 04
    Compare

    Use original sources, baselines and contrary cases; record the verification effort.

  5. 05
    Document the decision

    State where the system helps, where it fails and under which conditions it may be reused.

Public sources

References for verification and further work.

These sources support initial design. Legal, professional, ethical and organisational requirements depend on the case and the responsible functions.

Frequently asked questions

Research or university: questions to clarify before the project.

Can AI output be cited as a scientific source?

Generated output does not replace the source supporting a claim. Tool use should be documented under the relevant publication or research rules, while claims must lead back to verifiable sources.

How can fabricated citations be reduced?

Constrain the corpus, require links to original sources and verify every reference. A link alone does not prove that the source supports the claim.

Can an AI workflow be reproducible?

It can become meaningfully reproducible when data, versions, instructions, parameters, transformations and checks are retained. Models and services change, so dependencies and limitations must be stated.

Can confidential research data be used?

Only through a route consistent with authorisation, ethics, data protection, agreements and security. Minimise the data and verify where it travels and remains before testing.

What is a suitable first experiment?

Choose one bounded step with a controlled corpus, an expected result and a criterion that can expose failure as well as success.